When available, the publication has been linked as PDF. To access, please click on the PDF symbol at the bottom right of each listed publication.
2018 |
|
| 1. | Ramaswamy, Arunselvan; Bhatnagar, Shalabh Analysis of gradient descent methods with nondiminishing bounded errors Journal Article IEEE Transactions on Automatic Control, 63 (5), pp. 1465-1471, 2018. Abstract | BibTeX | Tags: Computer Science and Automation | Links: @article{Ramaswamy2018, title = { Analysis of gradient descent methods with nondiminishing bounded errors}, author = {Arunselvan Ramaswamy and Shalabh Bhatnagar}, url = {http://www.rbccps.org/wp-content/uploads/2018/08/08016343.pdf}, doi = {10.1109/TAC.2017.2744598}, year = {2018}, date = {2018-05-31}, journal = {IEEE Transactions on Automatic Control}, volume = {63}, number = {5}, pages = {1465-1471}, abstract = {The main aim of this paper is to provide an analysis of gradient descent (GD) algorithms with gradient errors that do not necessarily vanish, asymptotically. In particular, sufficient conditions are presented for both stability (almost sure boundedness of the iterates) and convergence of GD with bounded (possibly) nondiminishing gradient errors. In addition to ensuring stability, such an algorithm is shown to converge to a small neighborhood of the minimum set, which depends on the gradient errors. It is worth noting that the main result of this paper can be used to show that GD with asymptotically vanishing errors indeed converges to the minimum set. The results presented herein are not only more general when compared to previous results, but our analysis of GD with errors is new to the literature to the best of our knowledge. Our work extends the contributions of Mangasarian and Solodov, Bertsekas and Tsitsiklis, and Tadiόand Doucet. Using our framework, a simple yet effective implementation of GD using simultaneous perturbation stochastic approximations, with constant sensitivity parameters, is presented. Another important improvement over many previous results is that there are no “additional” restrictions imposed on the step sizes. In machine learning applications where step sizes are related to learning rates, our assumptions, unlike those of other papers, do not affect these learning rates. Finally, we present experimental results to validate our theory.}, keywords = {Computer Science and Automation}, pubstate = {published}, tppubtype = {article} } The main aim of this paper is to provide an analysis of gradient descent (GD) algorithms with gradient errors that do not necessarily vanish, asymptotically. In particular, sufficient conditions are presented for both stability (almost sure boundedness of the iterates) and convergence of GD with bounded (possibly) nondiminishing gradient errors. In addition to ensuring stability, such an algorithm is shown to converge to a small neighborhood of the minimum set, which depends on the gradient errors. It is worth noting that the main result of this paper can be used to show that GD with asymptotically vanishing errors indeed converges to the minimum set. The results presented herein are not only more general when compared to previous results, but our analysis of GD with errors is new to the literature to the best of our knowledge. Our work extends the contributions of Mangasarian and Solodov, Bertsekas and Tsitsiklis, and Tadiόand Doucet. Using our framework, a simple yet effective implementation of GD using simultaneous perturbation stochastic approximations, with constant sensitivity parameters, is presented. Another important improvement over many previous results is that there are no “additional” restrictions imposed on the step sizes. In machine learning applications where step sizes are related to learning rates, our assumptions, unlike those of other papers, do not affect these learning rates. Finally, we present experimental results to validate our theory. |
| 2. | Krishnapuram, Raghu; Rajkumar, Arun; Acharya, Adithya; Dhara, Nikhil; Goudar, Manjunath; Sarashetti, Akshay P Online domain adaptation by exploiting labeled features and pro-active learning Conference Proceedings of the 2018 ACM India Joint International Conference on Data Science and Management of Data (CoDS-COMAD), 11.-13.01.18, Goa, 2018. Abstract | BibTeX | Tags: Machine Learning | Links: @conference{Krishnapuram2018, title = { Online domain adaptation by exploiting labeled features and pro-active learning}, author = {Raghu Krishnapuram and Arun Rajkumar and Adithya Acharya and Nikhil Dhara and Manjunath Goudar and Akshay P. Sarashetti}, url = {http://www.rbccps.org/wp-content/uploads/2018/08/p137-krishnapuram.pdf}, doi = {10.1145/3152494.3152507}, year = {2018}, date = {2018-01-13}, booktitle = {Proceedings of the 2018 ACM India Joint International Conference on Data Science and Management of Data (CoDS-COMAD), 11.-13.01.18, Goa}, pages = {137-145}, abstract = {Domain adaptation algorithms have gained considerable interest in recent years due to their empirical success. However, most of the existing literature assumes the availability of a pool of unlabeled target domain data which is not the case in several practical scenarios. We consider the problem of online domain adaptation where the target (test) data arrives one at a time, and there is a fixed budget available for obtaining their labels. We propose a novel online domain adaptation algorithm which uses the budget judiciously by balancing the cost and reliability of the oracles which provide the labels. The proposed algorithm uses a probabilistic model which allows the seamless integration of previously unseen features (i.e., features that may appear during test time) into the model. Experiments on several benchmark real world datasets empirically establish the efficacy of the proposed algorithm.}, keywords = {Machine Learning}, pubstate = {published}, tppubtype = {conference} } Domain adaptation algorithms have gained considerable interest in recent years due to their empirical success. However, most of the existing literature assumes the availability of a pool of unlabeled target domain data which is not the case in several practical scenarios. We consider the problem of online domain adaptation where the target (test) data arrives one at a time, and there is a fixed budget available for obtaining their labels. We propose a novel online domain adaptation algorithm which uses the budget judiciously by balancing the cost and reliability of the oracles which provide the labels. The proposed algorithm uses a probabilistic model which allows the seamless integration of previously unseen features (i.e., features that may appear during test time) into the model. Experiments on several benchmark real world datasets empirically establish the efficacy of the proposed algorithm. |
| 3. | Lakshminarayanan, Chandrashekar; Bhatnagar, Shalabh; Szepesvári, Csaba A linearly relaxed approximate linear program for Markov decision processes Journal Article IEEE Transactions on Automatic Control, 63 (4), pp. 1185-1191, 2018. Abstract | BibTeX | Tags: Computer Science and Automation | Links: @article{Lakshminarayanan2018, title = {A linearly relaxed approximate linear program for Markov decision processes}, author = {Chandrashekar Lakshminarayanan and Shalabh Bhatnagar and Csaba Szepesvári}, url = {http://www.rbccps.org/wp-content/uploads/2018/08/08014469.pdf}, doi = {10.1109/TAC.2017.2743163}, year = {2018}, date = {2018-04-30}, journal = {IEEE Transactions on Automatic Control}, volume = {63}, number = {4}, pages = {1185-1191}, abstract = {Approximate linear programming (ALP) and its variants have been widely applied to Markov decision processes (MDPs) with a large number of states. A serious limitation of ALP is that it has an intractable number of constraints, as a result of which constraint approximations are of interest. In this paper, we define a linearly relaxed approximation linear program (LRALP) that has a tractable number of constraints, obtained as positive linear combinations of the original constraints of the ALP. The main contribution is a novel performance bound for LRALP.}, keywords = {Computer Science and Automation}, pubstate = {published}, tppubtype = {article} } Approximate linear programming (ALP) and its variants have been widely applied to Markov decision processes (MDPs) with a large number of states. A serious limitation of ALP is that it has an intractable number of constraints, as a result of which constraint approximations are of interest. In this paper, we define a linearly relaxed approximation linear program (LRALP) that has a tractable number of constraints, obtained as positive linear combinations of the original constraints of the ALP. The main contribution is a novel performance bound for LRALP. |
| 4. | Raseena, K T; Ghosh, Prasanta Kumar A maximum likelihood formulation to exploit heart rate variability for robust heart rate estimation from facial video Conference Forthcoming Proceedings of the 40th International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC 2018), 17.-21.07., Honolulu (USA), Forthcoming. Abstract | BibTeX | Tags: Electrical Engineering @conference{Raseena2018, title = {A maximum likelihood formulation to exploit heart rate variability for robust heart rate estimation from facial video}, author = {K. T. Raseena and Prasanta Kumar Ghosh}, year = {2018}, date = {2018-07-21}, booktitle = {Proceedings of the 40th International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC 2018), 17.-21.07., Honolulu (USA)}, abstract = {The problem of estimating the heart rate (HR) from a facial video is considered. A typical approach for this problem is to use independent component analysis (ICA) on the red, blue, green intensity profiles averaged over the facial region. This provides estimates of the underlying source signals, whose spectral peaks are used to predict HR in every analysis window. In this work, we propose a maximum likelihood formulation to optimally select a source signal in each window such that the predicted HR trajectory not only corresponds to the most likely spectral peaks but also ensures a realistic HR variability (HRV) across analysis windows. The likelihood function is efficiently optimized using dynamic programming in a manner similar to Viterbi decoding. The proposed scheme for HR estimation is denoted by vICA. The performance of vICA is compared with a typical ICA approach as well as a recently proposed sparse spectral peak tracking (SSPT) technique that ensures that the predicted HR does not vary drastically across analysis windows. Experiments are performed in a five fold setup using facial videos of 15 subjects recorded using two types of smartphones (Samsung Galaxy and iPhone) at three different distances (6inches, 1foot, 2feet) between the phone camera and the subject. Mean absolute error (MAE) between the original and predicted HR reveals that the proposed vICA scheme performs better than the best of the baseline schemes, namely SSPT by -8.69%, 52.77% and 8.00% when Samsung Galaxy phone was used at a distance of 6inches, 1foot, and 2feet respectively. These improvements are 12.13%, 13.59% and 18.34% when iPhone was used. This, in turn, suggests that the HR predicted from a facial video becomes more accurate when the smoothness of HRV is utilized in predicting the HR trajectory as done in the proposed vICA.}, keywords = {Electrical Engineering}, pubstate = {forthcoming}, tppubtype = {conference} } The problem of estimating the heart rate (HR) from a facial video is considered. A typical approach for this problem is to use independent component analysis (ICA) on the red, blue, green intensity profiles averaged over the facial region. This provides estimates of the underlying source signals, whose spectral peaks are used to predict HR in every analysis window. In this work, we propose a maximum likelihood formulation to optimally select a source signal in each window such that the predicted HR trajectory not only corresponds to the most likely spectral peaks but also ensures a realistic HR variability (HRV) across analysis windows. The likelihood function is efficiently optimized using dynamic programming in a manner similar to Viterbi decoding. The proposed scheme for HR estimation is denoted by vICA. The performance of vICA is compared with a typical ICA approach as well as a recently proposed sparse spectral peak tracking (SSPT) technique that ensures that the predicted HR does not vary drastically across analysis windows. Experiments are performed in a five fold setup using facial videos of 15 subjects recorded using two types of smartphones (Samsung Galaxy and iPhone) at three different distances (6inches, 1foot, 2feet) between the phone camera and the subject. Mean absolute error (MAE) between the original and predicted HR reveals that the proposed vICA scheme performs better than the best of the baseline schemes, namely SSPT by -8.69%, 52.77% and 8.00% when Samsung Galaxy phone was used at a distance of 6inches, 1foot, and 2feet respectively. These improvements are 12.13%, 13.59% and 18.34% when iPhone was used. This, in turn, suggests that the HR predicted from a facial video becomes more accurate when the smoothness of HRV is utilized in predicting the HR trajectory as done in the proposed vICA. |
| 5. | Kallatt, Sangeeth; Nair, Smitha; Majumdar, Kausik Asymmetrically encapsulated vertical ITO/MoS2/Cu2O photodetector with ultrahigh sensitivity Journal Article Small: Micro and Nano, 14 (3), pp. 1-7, 2018. Abstract | BibTeX | Tags: Photonics | Links: @article{Kallatt2018, title = {Asymmetrically encapsulated vertical ITO/MoS2/Cu2O photodetector with ultrahigh sensitivity }, author = {Sangeeth Kallatt and Smitha Nair and Kausik Majumdar}, url = {http://www.rbccps.org/wp-content/uploads/2018/08/Kallatt_et_al-2018-Small.pdf}, doi = {10.1002/smll.201702066}, year = {2018}, date = {2018-01-31}, journal = {Small: Micro and Nano}, volume = {14}, number = {3}, pages = {1-7}, abstract = {Strong light absorption, coupled with moderate carrier transport properties, makes 2D layered transition metal dichalcogenide semiconductors promising candidates for low intensity photodetection applications. However, the performance of these devices is severely bottlenecked by slow response with persistent photocurrent due to long lived charge trapping, and nonreliable characteristics due to undesirable ambience and substrate effects. Here ultrahigh specific detectivity (D*) of 3.2 × 10^14 Jones and responsivity (R) of 5.77 × 10^4 A W^−1 are demonstrated at an optical power density (Pop) of 0.26 W m^−2 and external bias (Vext) of −0.5 V in an indium tin oxide/MoS2/copper oxide/Au vertical multi‐heterojunction photodetector exhibiting small carrier transit time. The active MoS2 layer being encapsulated by carrier collection layers allows us to achieve repeatable characteristics over large number of cycles with negligible trap assisted persistent photocurrent. A large D* > 10^14 Jones at zero external bias is also achieved due to the built‐in field of the asymmetric photodetector. Benchmarking the performance against existing reports in literature shows a viable pathway for achieving reliable and highly sensitive photodetectors for ultralow intensity photodetection applications. }, keywords = {Photonics}, pubstate = {published}, tppubtype = {article} } Strong light absorption, coupled with moderate carrier transport properties, makes 2D layered transition metal dichalcogenide semiconductors promising candidates for low intensity photodetection applications. However, the performance of these devices is severely bottlenecked by slow response with persistent photocurrent due to long lived charge trapping, and nonreliable characteristics due to undesirable ambience and substrate effects. Here ultrahigh specific detectivity (D*) of 3.2 × 10^14 Jones and responsivity (R) of 5.77 × 10^4 A W^−1 are demonstrated at an optical power density (Pop) of 0.26 W m^−2 and external bias (Vext) of −0.5 V in an indium tin oxide/MoS2/copper oxide/Au vertical multi‐heterojunction photodetector exhibiting small carrier transit time. The active MoS2 layer being encapsulated by carrier collection layers allows us to achieve repeatable characteristics over large number of cycles with negligible trap assisted persistent photocurrent. A large D* > 10^14 Jones at zero external bias is also achieved due to the built‐in field of the asymmetric photodetector. Benchmarking the performance against existing reports in literature shows a viable pathway for achieving reliable and highly sensitive photodetectors for ultralow intensity photodetection applications. |
| 6. | Gulhare, Siddhartha; Verma, Ashish; Chakroborty, Partha Comparison of pedestrian data of single file movement collected from controlled pedestrian experiment and from field in mass religious gathering Journal Article Collective Dynamics, 3 , pp. A16: 1-14, 2018. Abstract | BibTeX | Tags: The Kumbh Mela Experiment | Links: @article{Gulhare2018, title = {Comparison of pedestrian data of single file movement collected from controlled pedestrian experiment and from field in mass religious gathering}, author = {Siddhartha Gulhare and Ashish Verma and Partha Chakroborty}, url = {http://www.rbccps.org/wp-content/uploads/2018/08/27-278-1-PB.pdf}, doi = {10.17815/CD.2018.16}, year = {2018}, date = {2018-05-04}, journal = {Collective Dynamics}, volume = {3}, pages = {A16: 1-14}, abstract = {Managing and controlling crowd during mass religious gathering is a challenge for organizers. With good computational capabilities, it is possible to create tools to simulate crowd in real time to aid crowd management. These tools need to be first calibrated and validated with pedestrian empirical data. The empirical data collection from field is difficult and therefore, data collection through controlled pedestrian experiments have become a convenient substitute. However, the ability of experiment data to reproduce actual crowd behavior needs to be examined. This study compared the experiment data with field data collected from mass religious gathering named Kumbh Mela held in India, 2016. The single file movement (pedestrians moving along a single line; SFM) experiment was conducted and its results were compared with the field SFM results. The speed in the field was found to be generally higher than in the experiment for a given density. The results clearly indicate that the pedestrians in the field are motivated to achieve a purpose but participants in the experiments lack the motivation. The pedestrian dynamics of the experiment was found to be different from the field. Hence, the results of pedestrian experiments should not be extrapolated to understand panic, crowd risk situations.}, keywords = {The Kumbh Mela Experiment}, pubstate = {published}, tppubtype = {article} } Managing and controlling crowd during mass religious gathering is a challenge for organizers. With good computational capabilities, it is possible to create tools to simulate crowd in real time to aid crowd management. These tools need to be first calibrated and validated with pedestrian empirical data. The empirical data collection from field is difficult and therefore, data collection through controlled pedestrian experiments have become a convenient substitute. However, the ability of experiment data to reproduce actual crowd behavior needs to be examined. This study compared the experiment data with field data collected from mass religious gathering named Kumbh Mela held in India, 2016. The single file movement (pedestrians moving along a single line; SFM) experiment was conducted and its results were compared with the field SFM results. The speed in the field was found to be generally higher than in the experiment for a given density. The results clearly indicate that the pedestrians in the field are motivated to achieve a purpose but participants in the experiments lack the motivation. The pedestrian dynamics of the experiment was found to be different from the field. Hence, the results of pedestrian experiments should not be extrapolated to understand panic, crowd risk situations. |
| 7. | Rao, Suman; Thankachan, Prashanth; Amrutur, Bharadwaj; Washington, Maryann; Mony, Prem Continuous, real-time monitoring of neonatal position and temperature during Kangaroo Mother Care using a wearable sensor: A techno-feasibility pilot study Journal Article Pilot and Feasibility Studies, 99 (4), pp. 1-7, 2018. Abstract | BibTeX | Tags: Health, Remote Neonatal Monitoring and Intervention | Links: @article{Rao2018, title = {Continuous, real-time monitoring of neonatal position and temperature during Kangaroo Mother Care using a wearable sensor: A techno-feasibility pilot study}, author = {Suman Rao and Prashanth Thankachan and Bharadwaj Amrutur and Maryann Washington and Prem Mony}, url = {http://www.rbccps.org/wp-content/uploads/2018/06/Rao_et_al-2018-Pilot_and_Feasibility_Studies.pdf}, doi = {10.1186/s40814-018-0293-5}, year = {2018}, date = {2018-05-21}, journal = {Pilot and Feasibility Studies}, volume = {99}, number = {4}, pages = {1-7}, abstract = {Background: Remote biomonitoring of vital parameters in hospitals and homes has the potential to improve coverage and quality of maternal and neonatal health. Wearable sensors coupled with modern information and communication technology now offer an opportunity to monitor temperatures and kangaroo mother care (KMC) adherence in a continuous and real-time manner remotely for several days’ duration in hospital and home settings. Using an innovative remote biomonitoring device to measure both temperature and baby position, we undertook a techno-feasibility study in preparation for a clinical trial. Methods: We designed and developed a wearable sensor for tracking KMC adherence and neonatal temperature, using social innovation design principles. After screening mother-infant dyads using clinical and logistic eligibility criteria, we piloted this wearable sensor along with a gateway device and the commercial cellular network. The dyads were recruited during hospitalization and followed up in the hospital and home phases for several days. Simple descriptive statistical analysis was undertaken. Results: Recruitment rate was 50% (6/12), and consenting rate was 83% (5/6) during a 2-month period. These five neonates contributed a total of 39 study days (15 hospital days and 24 home days). Their mean [± standard deviation (S.D.)] birth weight was 1490 (± 244) g. The mean (± S.D.) of the vital signs for the five babies was temperature [36.5 °C (± 0.3)], heart rate [146.5/min (± 14)], and oxygen saturation [94% (± 4)]. No severe or moderate side-effects were noted; one baby developed mild dermatitis under the device that was transient and self-limiting, yielding an incidence proportion of 20% and incidence rate of 2.6/100 person-days. None of the mothers reported any discomfort with the use of the device. Temperatures detected from 81 paired readings revealed that those from the wearable sensor were 0.2 °C lower than those detected by clinical thermometers [36.4 (± 0.7) vs 36.6 (± 0.3); < 0.001]. There was also iterative feedback that was useful for hardware and software design specifications of the wearable sensor, the gateway device, and the analytics platform. Lastly, lessons were learnt with regard to the logistics of research team interactions with healthcare professionals and study participants during the hospitalization and post-discharge home phases of the study. Conclusions: The pilot study has shown that it is feasible and acceptable to track KMC adherence as well as maternal and newborn temperatures in a potentially safe manner on a real-time mode for several days’ duration during hospitalization and home phases. The pilot has also helped inform modifications in clinical monitoring, technological modifications, and logistics planning in preparation for the definitive clinical trial.}, keywords = {Health, Remote Neonatal Monitoring and Intervention}, pubstate = {published}, tppubtype = {article} } Background: Remote biomonitoring of vital parameters in hospitals and homes has the potential to improve coverage and quality of maternal and neonatal health. Wearable sensors coupled with modern information and communication technology now offer an opportunity to monitor temperatures and kangaroo mother care (KMC) adherence in a continuous and real-time manner remotely for several days’ duration in hospital and home settings. Using an innovative remote biomonitoring device to measure both temperature and baby position, we undertook a techno-feasibility study in preparation for a clinical trial. Methods: We designed and developed a wearable sensor for tracking KMC adherence and neonatal temperature, using social innovation design principles. After screening mother-infant dyads using clinical and logistic eligibility criteria, we piloted this wearable sensor along with a gateway device and the commercial cellular network. The dyads were recruited during hospitalization and followed up in the hospital and home phases for several days. Simple descriptive statistical analysis was undertaken. Results: Recruitment rate was 50% (6/12), and consenting rate was 83% (5/6) during a 2-month period. These five neonates contributed a total of 39 study days (15 hospital days and 24 home days). Their mean [± standard deviation (S.D.)] birth weight was 1490 (± 244) g. The mean (± S.D.) of the vital signs for the five babies was temperature [36.5 °C (± 0.3)], heart rate [146.5/min (± 14)], and oxygen saturation [94% (± 4)]. No severe or moderate side-effects were noted; one baby developed mild dermatitis under the device that was transient and self-limiting, yielding an incidence proportion of 20% and incidence rate of 2.6/100 person-days. None of the mothers reported any discomfort with the use of the device. Temperatures detected from 81 paired readings revealed that those from the wearable sensor were 0.2 °C lower than those detected by clinical thermometers [36.4 (± 0.7) vs 36.6 (± 0.3); < 0.001]. There was also iterative feedback that was useful for hardware and software design specifications of the wearable sensor, the gateway device, and the analytics platform. Lastly, lessons were learnt with regard to the logistics of research team interactions with healthcare professionals and study participants during the hospitalization and post-discharge home phases of the study. Conclusions: The pilot study has shown that it is feasible and acceptable to track KMC adherence as well as maternal and newborn temperatures in a potentially safe manner on a real-time mode for several days’ duration during hospitalization and home phases. The pilot has also helped inform modifications in clinical monitoring, technological modifications, and logistics planning in preparation for the definitive clinical trial. |
| 8. | Rathod, Nihesh; Subramanian, Renu; Sundaresan, Rajesh Data-driven and GIS-based coverage estimation in a heterogeneous propagation environment Conference Forthcoming Proceedings of the 2018 IEEE Global Communications Conference: Selected Areas in Communications: Internet of Things, 9.-13.12.18, Abu Dhabi (UAE), Forthcoming. BibTeX | Tags: Electrical Communications Engineering @conference{Rathod2018, title = {Data-driven and GIS-based coverage estimation in a heterogeneous propagation environment}, author = {Nihesh Rathod and Renu Subramanian and Rajesh Sundaresan}, year = {2018}, date = {2018-12-13}, booktitle = {Proceedings of the 2018 IEEE Global Communications Conference: Selected Areas in Communications: Internet of Things, 9.-13.12.18, Abu Dhabi (UAE)}, keywords = {Electrical Communications Engineering}, pubstate = {forthcoming}, tppubtype = {conference} } |
| 9. | Sawant, Omkar; Bhowmick, Anirban; Sreenivas, Thippur V Enhanced directional sensitivity using acoustic dish reflector Conference Forthcoming International Conference on Signal Processing and Communications (SPCOM 2018), 16.-19.07.18, Bangalore (India), Forthcoming. Abstract | BibTeX | Tags: Sensing and Actuation Systems | Links: @conference{Sawant2018, title = {Enhanced directional sensitivity using acoustic dish reflector}, author = {Omkar Sawant and Anirban Bhowmick and Thippur V. Sreenivas}, url = {http://www.rbccps.org/wp-content/uploads/2018/05/PID5361497.pdf}, year = {2018}, date = {2018-07-19}, booktitle = {International Conference on Signal Processing and Communications (SPCOM 2018), 16.-19.07.18, Bangalore (India)}, abstract = {An acoustic reflector is a passive device which is used with microphones to reflect and focus sound waves from a longer distance. They are effective for long-range sound recording without intrusion, such as to record wildlife sounds. In this paper, our objective is to investigate the feasibility of using a dish reflector microphone for indoor activity monitoring of large premises. Acoustic gain and frequency response of the microphone with different parabolic, near parabolic and non- parabolic reflectors are compared with that of the same micro- phone without the reflector. It is observed that the use of dish reflector does significantly enhance the directivity and sensitivity of sound recording. Keeping to a moderate size reflector, among various shapes and materials, the best reflector has given a 14 dB gain improvement over microphone without reflector at 0 degree azimuth. We also found that among steel and plastic dish reflector of approximately same size and shape, steel dish reflector does give a higher performance as expected. The increased directivity of the dish reflector is giving a significant improvement in indoor monitoring and also for sound source tracking.}, keywords = {Sensing and Actuation Systems}, pubstate = {forthcoming}, tppubtype = {conference} } An acoustic reflector is a passive device which is used with microphones to reflect and focus sound waves from a longer distance. They are effective for long-range sound recording without intrusion, such as to record wildlife sounds. In this paper, our objective is to investigate the feasibility of using a dish reflector microphone for indoor activity monitoring of large premises. Acoustic gain and frequency response of the microphone with different parabolic, near parabolic and non- parabolic reflectors are compared with that of the same micro- phone without the reflector. It is observed that the use of dish reflector does significantly enhance the directivity and sensitivity of sound recording. Keeping to a moderate size reflector, among various shapes and materials, the best reflector has given a 14 dB gain improvement over microphone without reflector at 0 degree azimuth. We also found that among steel and plastic dish reflector of approximately same size and shape, steel dish reflector does give a higher performance as expected. The increased directivity of the dish reflector is giving a significant improvement in indoor monitoring and also for sound source tracking. |
| 10. | Kumar, Saurabh; Amrutur, Bharadwaj; Asokan, Sundarrajan Evaluation of Fiber Bragg Grating sensor interrogation using InGaAs linear detector arrays and Gaussian approximation on embedded hardware Journal Article Review of Scientific Instruments, 89 (2), 2018. Abstract | BibTeX | Tags: Fiber Bragg Grating Sensors for Distributed Bio-Chemical Sensing | Links: @article{Kumar2018b, title = {Evaluation of Fiber Bragg Grating sensor interrogation using InGaAs linear detector arrays and Gaussian approximation on embedded hardware}, author = {Saurabh Kumar and Bharadwaj Amrutur and Sundarrajan Asokan}, url = {http://www.rbccps.org/wp-content/uploads/2018/06/1.5022548.pdf}, doi = {10.1063/1.5022548}, year = {2018}, date = {2018-02-28}, journal = {Review of Scientific Instruments}, volume = {89}, number = {2}, abstract = {Fiber Bragg Grating (FBG) sensors have become popular for applications related to structural health monitoring, biomedical engineering, and robotics. However, for successful large scale adoption, FBG interrogation systems are as important as sensor characteristics. Apart from accuracy, the required number of FBG sensors per fiber and the distance between the device in which the sensors are used and the interrogation system also influence the selection of the interrogation technique. For several measurement devices developed for applications in biomedical engineering and robotics, only a few sensors per fiber are required and the device is close to the interrogation system. For these applications, interrogation systems based on InGaAs linear detector arrays provide a good choice. However, their resolution is dependent on the algorithms used for curve fitting. In this work, a detailed analysis of the choice of algorithm using the Gaussian approximation for the FBG spectrum and the number of pixels used for curve fitting on the errors is provided. The points where the maximum errors occur have been identified. All comparisons for wavelength shift detection have been made against another interrogation system based on the tunable swept laser. It has been shown that maximum errors occur when the wavelength shift is such that one new pixel is included for curve fitting. It has also been shown that an algorithm with lower computation cost compared to the more popular methods using iterative non-linear least squares estimation can be used without leading to the loss of accuracy. The algorithm has been implemented on embedded hardware, and a speed-up of approximately six times has been observed.}, keywords = {Fiber Bragg Grating Sensors for Distributed Bio-Chemical Sensing}, pubstate = {published}, tppubtype = {article} } Fiber Bragg Grating (FBG) sensors have become popular for applications related to structural health monitoring, biomedical engineering, and robotics. However, for successful large scale adoption, FBG interrogation systems are as important as sensor characteristics. Apart from accuracy, the required number of FBG sensors per fiber and the distance between the device in which the sensors are used and the interrogation system also influence the selection of the interrogation technique. For several measurement devices developed for applications in biomedical engineering and robotics, only a few sensors per fiber are required and the device is close to the interrogation system. For these applications, interrogation systems based on InGaAs linear detector arrays provide a good choice. However, their resolution is dependent on the algorithms used for curve fitting. In this work, a detailed analysis of the choice of algorithm using the Gaussian approximation for the FBG spectrum and the number of pixels used for curve fitting on the errors is provided. The points where the maximum errors occur have been identified. All comparisons for wavelength shift detection have been made against another interrogation system based on the tunable swept laser. It has been shown that maximum errors occur when the wavelength shift is such that one new pixel is included for curve fitting. It has also been shown that an algorithm with lower computation cost compared to the more popular methods using iterative non-linear least squares estimation can be used without leading to the loss of accuracy. The algorithm has been implemented on embedded hardware, and a speed-up of approximately six times has been observed. |
| 11. | Zhou, Enlu; Bhatnagar, Shalabh Gradient-based adaptive stochastic search for simulation optimization over continuous space Journal Article INFORMS Journal on Computing, 30 (1), pp. 154-167, 2018. Abstract | BibTeX | Tags: Computer Science and Automation | Links: @article{Zhou2018, title = {Gradient-based adaptive stochastic search for simulation optimization over continuous space}, author = {Enlu Zhou and Shalabh Bhatnagar}, url = {http://www.rbccps.org/wp-content/uploads/2018/08/ijoc.2017.0771.pdf}, doi = {10.1287/ijoc.2017.0771}, year = {2018}, date = {2018-01-23}, journal = {INFORMS Journal on Computing}, volume = {30}, number = {1}, pages = {154-167}, abstract = {We extend the idea of model-based algorithms for deterministic optimization to simulation optimization over continuous space. Model-based algorithms iteratively generate a population of candidate solutions from a sampling distribution and use the performance of the candidate solutions to update the sampling distribution. By viewing the original simulation optimization problem as another optimization problem over the parameter space of the sampling distribution, we propose to use a direct gradient search on the parameter space to update the sampling distribution. To improve the computational efficiency, we further develop a two-timescale updating scheme that updates the parameter on a slow timescale and estimates the quantities involved in the parameter updating on the fast timescale. We analyze the convergence properties of our algorithms through techniques from stochastic approximation, and demonstrate the good empirical performance by comparing with two state-of-the-art model-based simulation optimization methods.}, keywords = {Computer Science and Automation}, pubstate = {published}, tppubtype = {article} } We extend the idea of model-based algorithms for deterministic optimization to simulation optimization over continuous space. Model-based algorithms iteratively generate a population of candidate solutions from a sampling distribution and use the performance of the candidate solutions to update the sampling distribution. By viewing the original simulation optimization problem as another optimization problem over the parameter space of the sampling distribution, we propose to use a direct gradient search on the parameter space to update the sampling distribution. To improve the computational efficiency, we further develop a two-timescale updating scheme that updates the parameter on a slow timescale and estimates the quantities involved in the parameter updating on the fast timescale. We analyze the convergence properties of our algorithms through techniques from stochastic approximation, and demonstrate the good empirical performance by comparing with two state-of-the-art model-based simulation optimization methods. |
| 12. | Somasundaran, Biju Venkadath; Soundararajan, Rajiv; Biswas, Soma Image denoising for image retrieval by cascading a deep quality assessment network Conference Proceedings of the 2018 IEEE International Conference on Image Processing, 7.-10.10.18, Athens (Greece), 2018. BibTeX | Tags: Electrical Communications Engineering @conference{Somasundaran2018, title = {Image denoising for image retrieval by cascading a deep quality assessment network}, author = {Biju Venkadath Somasundaran and Rajiv Soundararajan and Soma Biswas}, year = {2018}, date = {2018-10-09}, booktitle = {Proceedings of the 2018 IEEE International Conference on Image Processing, 7.-10.10.18, Athens (Greece)}, keywords = {Electrical Communications Engineering}, pubstate = {published}, tppubtype = {conference} } |
| 13. | Kolathaya, Shishir; James, Aaron D Input-to-state safety with control barrier functions Journal Article IEEE Control Systems Letters, 3 (1), pp. 108-113, 2018. Abstract | BibTeX | Tags: Autonomous Systems | Links: @article{Kolathaya2018, title = {Input-to-state safety with control barrier functions}, author = {Shishir Kolathaya and Aaron D. James}, url = {http://www.rbccps.org/wp-content/uploads/2018/08/08405547.pdf}, doi = {10.1109/LCSYS.2018.2853698}, year = {2018}, date = {2018-07-06}, journal = {IEEE Control Systems Letters}, volume = {3}, number = {1}, pages = {108-113}, abstract = {This letter presents a new notion of input-to-state safe control barrier functions (ISSf-CBFs), which ensure safety of nonlinear dynamical systems under input disturbances. Similar to how safety conditions are specified in terms of forward invariance of a set, input-to-state safety conditions are specified in terms of forward invariance of a slightly larger set. In this context, invariance of the larger set implies that the states stay either inside or very close to the smaller safe set; and this closeness is bounded by the magnitude of the disturbances. The main contribution of the letter is the methodology used for obtaining a valid ISSf-CBF, given a control barrier function. The associated universal control law will also be provided. Towards the end, we will study unified quadratic programs that combine control Lyapunov functions and ISSf-CBFs in order to obtain a single control law that ensures both safety and stability in systems with input disturbances.}, keywords = {Autonomous Systems}, pubstate = {published}, tppubtype = {article} } This letter presents a new notion of input-to-state safe control barrier functions (ISSf-CBFs), which ensure safety of nonlinear dynamical systems under input disturbances. Similar to how safety conditions are specified in terms of forward invariance of a set, input-to-state safety conditions are specified in terms of forward invariance of a slightly larger set. In this context, invariance of the larger set implies that the states stay either inside or very close to the smaller safe set; and this closeness is bounded by the magnitude of the disturbances. The main contribution of the letter is the methodology used for obtaining a valid ISSf-CBF, given a control barrier function. The associated universal control law will also be provided. Towards the end, we will study unified quadratic programs that combine control Lyapunov functions and ISSf-CBFs in order to obtain a single control law that ensures both safety and stability in systems with input disturbances. |
| 14. | Ramesh, Rakshit; Acharya, Srikrishna; Rajaraman, Vasanth; Babu, Arun; Joglekar, Ashish; Sharma, Abhay; Amrutur, Bharadwaj; Namekar, Prashant Interoperable middleware for smartcities - Streetlighting on LoRaWAN as a case study Conference 10th International Conference on Communication Systems and Networks (COMSNETS), 03.-07.01.18, Bangalore, 2018. Abstract | BibTeX | Tags: Smart City | Links: @conference{Ramesh2018, title = {Interoperable middleware for smartcities - Streetlighting on LoRaWAN as a case study}, author = {Rakshit Ramesh and Srikrishna Acharya and Vasanth Rajaraman and Arun Babu and Ashish Joglekar and Abhay Sharma and Bharadwaj Amrutur and Prashant Namekar}, url = {http://www.rbccps.org/wp-content/uploads/2018/08/AnInteroperableMiddleware.pdf}, year = {2018}, date = {2018-01-07}, booktitle = {10th International Conference on Communication Systems and Networks (COMSNETS), 03.-07.01.18, Bangalore}, abstract = {A smart city, is like every other organism a structured and organized entity. An organism comprises of a brain that performs its tasks via its appendages, just as smart cities should. However the brain needs a spine through which it communicates information to its appendages, becomes a point of confluence of different sensory and actuator pathways and facilitates activities. We herein put forth our work involving the building of such a spine for a smart city, and focus on the last mile communication aspects, especially using an LPWAN network and an implementation example with smart streetlights.}, keywords = {Smart City}, pubstate = {published}, tppubtype = {conference} } A smart city, is like every other organism a structured and organized entity. An organism comprises of a brain that performs its tasks via its appendages, just as smart cities should. However the brain needs a spine through which it communicates information to its appendages, becomes a point of confluence of different sensory and actuator pathways and facilitates activities. We herein put forth our work involving the building of such a spine for a smart city, and focus on the last mile communication aspects, especially using an LPWAN network and an implementation example with smart streetlights. |
| 15. | Diddigi, Raghuram Bharadwaj; Prabuchandran, K J; Bhatnagar, Shalabh Novel sensor scheduling scheme for intruder tracking in energy efficient sensor networks Journal Article IEEE Wireless Communications Letters, pp. 1-4, 2018. Abstract | BibTeX | Tags: Computer Science and Automation | Links: @article{Diddigi2018, title = {Novel sensor scheduling scheme for intruder tracking in energy efficient sensor networks}, author = {Raghuram Bharadwaj Diddigi and K. J. Prabuchandran and Shalabh Bhatnagar}, url = {http://www.rbccps.org/wp-content/uploads/2018/08/08314081.pdf}, doi = {10.1109/LWC.2018.2814576}, year = {2018}, date = {2018-03-12}, journal = {IEEE Wireless Communications Letters}, pages = {1-4}, abstract = {We consider the problem of tracking an intruder using a network of wireless sensors. For tracking the intruder at each instant, the optimal number and the right configuration of sensors has to be powered. As powering the sensors consumes energy, there is a trade off between accurately tracking the position of the intruder at each instant and the energy consumption of sensors. This problem has been formulated in the framework of Partially Observable Markov Decision Process (POMDP). Even for the state-of-the-art algorithm in the literature, the curse of dimensionality renders the problem intractable. In this paper, we formulate the Intrusion Detection (ID) problem with a suitable state-action space in the framework of POMDP and develop a Reinforcement Learning (RL) algorithm utilizing the Upper Confidence Tree Search (UCT) method to solve the ID problem. Through simulations, we show that our algorithm performs and scales well with the increasing state and action spaces.}, keywords = {Computer Science and Automation}, pubstate = {published}, tppubtype = {article} } We consider the problem of tracking an intruder using a network of wireless sensors. For tracking the intruder at each instant, the optimal number and the right configuration of sensors has to be powered. As powering the sensors consumes energy, there is a trade off between accurately tracking the position of the intruder at each instant and the energy consumption of sensors. This problem has been formulated in the framework of Partially Observable Markov Decision Process (POMDP). Even for the state-of-the-art algorithm in the literature, the curse of dimensionality renders the problem intractable. In this paper, we formulate the Intrusion Detection (ID) problem with a suitable state-action space in the framework of POMDP and develop a Reinforcement Learning (RL) algorithm utilizing the Upper Confidence Tree Search (UCT) method to solve the ID problem. Through simulations, we show that our algorithm performs and scales well with the increasing state and action spaces. |
| 16. | George, Jude Baby; Abraham, Grace; Rashid, Zubin; Amrutur, Bharadwaj; Sikdar, Sujit Random neuronal ensembles can inherently do context dependent coarse conjunctive encoding of input stimulus without any specific training Journal Article Scientific Reports, 8 , pp. 1-10, 2018. Abstract | BibTeX | Tags: Systems Engineering | Links: @article{George2018b, title = {Random neuronal ensembles can inherently do context dependent coarse conjunctive encoding of input stimulus without any specific training}, author = {Jude Baby George and Grace Abraham and Zubin Rashid and Bharadwaj Amrutur and Sujit Sikdar}, url = {http://www.rbccps.org/wp-content/uploads/2018/06/41598_2018_Article_19462.pdf}, doi = {10.1038/s41598-018-19462-3}, year = {2018}, date = {2018-01-23}, journal = {Scientific Reports}, volume = {8}, pages = {1-10}, abstract = {Conjunctive encoding of inputs has been hypothesized to be a key feature in the computational capabilities of the brain. This has been inferred based on behavioral studies and electrophysiological recording from animals. In this report, we show that random neuronal ensembles grown on multi-electrode array perform a coarse-conjunctive encoding for a sequence of inputs with the first input setting the context. Such an encoding scheme creates similar yet unique population codes at the output of the ensemble, for related input sequences, which can then be decoded via a simple perceptron and hence a single STDP neuron layer. The random neuronal ensembles allow for pattern generalization and novel sequence classification without needing any specific learning or training of the ensemble. Such a representation of the inputs as population codes of neuronal ensemble outputs, has inherent redundancy and is suitable for further decoding via even probabilistic/random connections to subsequent neuronal layers. We reproduce this behavior in a mathematical model to show that a random neuronal network with a mix of excitatory and inhibitory neurons and sufficient connectivity creates similar coarse-conjunctive encoding of input sequences.}, keywords = {Systems Engineering}, pubstate = {published}, tppubtype = {article} } Conjunctive encoding of inputs has been hypothesized to be a key feature in the computational capabilities of the brain. This has been inferred based on behavioral studies and electrophysiological recording from animals. In this report, we show that random neuronal ensembles grown on multi-electrode array perform a coarse-conjunctive encoding for a sequence of inputs with the first input setting the context. Such an encoding scheme creates similar yet unique population codes at the output of the ensemble, for related input sequences, which can then be decoded via a simple perceptron and hence a single STDP neuron layer. The random neuronal ensembles allow for pattern generalization and novel sequence classification without needing any specific learning or training of the ensemble. Such a representation of the inputs as population codes of neuronal ensemble outputs, has inherent redundancy and is suitable for further decoding via even probabilistic/random connections to subsequent neuronal layers. We reproduce this behavior in a mathematical model to show that a random neuronal network with a mix of excitatory and inhibitory neurons and sufficient connectivity creates similar coarse-conjunctive encoding of input sequences. |
| 17. | Mony, Prem; Thankachan, Prashanth; Bhat, Swarnarekha; Rao, Suman; Washington, Maryann; Antony, Sumi; Thomas, Annamma; Nagarajarao, Sheela C; Rao, Hiteshwar; Amrutur, Bharadwaj Remote biomonitoring of temperatures in mothers and newborns: Design, development and testing of a wearable sensor device in a tertiary-care hospital in southern India Journal Article BMJ Innovations, pp. 1-8, 2018. Abstract | BibTeX | Tags: Health, Remote Neonatal Monitoring and Intervention | Links: @article{Mony2018b, title = {Remote biomonitoring of temperatures in mothers and newborns: Design, development and testing of a wearable sensor device in a tertiary-care hospital in southern India}, author = {Prem Mony and Prashanth Thankachan and Swarnarekha Bhat and Suman Rao and Maryann Washington and Sumi Antony and Annamma Thomas and Sheela C Nagarajarao and Hiteshwar Rao and Bharadwaj Amrutur}, url = {http://www.rbccps.org/wp-content/uploads/2018/06/bmjinnov-2016-000153.full_.pdf}, doi = {10.1136/bmjinnov-2016-000153}, year = {2018}, date = {2018-02-14}, journal = {BMJ Innovations}, pages = {1-8}, abstract = {Objective: Newer technologies such as wearables, sensors, mobile telephony and computing offer opportunities to monitor vital physiological parameters and tackle healthcare problems, thereby improving access and quality of care. We describe the design, development and testing of a wearable sensor device for remote biomonitoring of body temperatures in mothers and newborns in southern India. Methods: Based on client needs and technological requirements, a wearable sensor device was designed and developed using principles of ‘social innovation’ design. The device underwent multiple iterations in product design and engineering based on user feedback, and then following preclinical testing, a techno- feasibility study and clinical trial were undertaken in a tertiary-care teaching hospital in Bangalore, India. Clinical trial phases I and IIa for evaluation of safety and efficacy were undertaken in the following sequence: 7 healthy adult volunteers; 18 healthy mothers; 3 healthy babies; 10 stable babies in the neonatal care intensive unit and 1 baby with morbidities. Time stamped skin temperature readings obtained at 5 min intervals over a 1-hour period from the device secured on upper arms of mothers and abdomen of neonates were compared against readings from thermometers used routinely in clinical practice. Results: Devices were comfortably secured on to adults and neonates, and data were efficiently transmitted via the gateway device for secure storage and retrieval for analysis. The mean skin temperatures in mothers were lower than the axillary temperatures by 2°C; and in newborns, there was a precision of –0.5°C relative to axillary measurements. While occasional minimal adverse events were noted in healthy volunteers, no adverse events were noted in mothers or neonates. Conclusions: This proof-of-concept study shows that this device is promising in terms of feasibility, safety and accuracy (with appropriate calibration) with potential for further refinements in device accuracy and pursuit of further phases of clinical research for improved maternal and neonatal health.}, keywords = {Health, Remote Neonatal Monitoring and Intervention}, pubstate = {published}, tppubtype = {article} } Objective: Newer technologies such as wearables, sensors, mobile telephony and computing offer opportunities to monitor vital physiological parameters and tackle healthcare problems, thereby improving access and quality of care. We describe the design, development and testing of a wearable sensor device for remote biomonitoring of body temperatures in mothers and newborns in southern India. Methods: Based on client needs and technological requirements, a wearable sensor device was designed and developed using principles of ‘social innovation’ design. The device underwent multiple iterations in product design and engineering based on user feedback, and then following preclinical testing, a techno- feasibility study and clinical trial were undertaken in a tertiary-care teaching hospital in Bangalore, India. Clinical trial phases I and IIa for evaluation of safety and efficacy were undertaken in the following sequence: 7 healthy adult volunteers; 18 healthy mothers; 3 healthy babies; 10 stable babies in the neonatal care intensive unit and 1 baby with morbidities. Time stamped skin temperature readings obtained at 5 min intervals over a 1-hour period from the device secured on upper arms of mothers and abdomen of neonates were compared against readings from thermometers used routinely in clinical practice. Results: Devices were comfortably secured on to adults and neonates, and data were efficiently transmitted via the gateway device for secure storage and retrieval for analysis. The mean skin temperatures in mothers were lower than the axillary temperatures by 2°C; and in newborns, there was a precision of –0.5°C relative to axillary measurements. While occasional minimal adverse events were noted in healthy volunteers, no adverse events were noted in mothers or neonates. Conclusions: This proof-of-concept study shows that this device is promising in terms of feasibility, safety and accuracy (with appropriate calibration) with potential for further refinements in device accuracy and pursuit of further phases of clinical research for improved maternal and neonatal health. |
| 18. | K.T., Raseena; Ghosh, Prasanta Kumar Robust heart rate estimation from facial video based on a maximum likelihood formulation using multiple spectral peaks exploiting heart rate variability Journal Article Forthcoming Biomedical Signal Processing and Control, Forthcoming. @article{K.T.2018b, title = {Robust heart rate estimation from facial video based on a maximum likelihood formulation using multiple spectral peaks exploiting heart rate variability}, author = {Raseena K.T. and Prasanta Kumar Ghosh}, year = {2018}, date = {2018-07-31}, journal = {Biomedical Signal Processing and Control}, keywords = {Health}, pubstate = {forthcoming}, tppubtype = {article} } |
| 19. | Kundu, Atreyee Stabilizing switched nonlinear systems under restricted switching Conference Proceedings of the 21st International Conference on Hybrid Systems: Computation and Control, 2018. Abstract | BibTeX | Tags: Control and Optimisation and Policy | Links: @conference{Kundu2018, title = {Stabilizing switched nonlinear systems under restricted switching}, author = {Atreyee Kundu}, url = {http://www.rbccps.org/wp-content/uploads/2018/08/p101-Kundu-1.pdf}, doi = {10.1145/3178126.3178130}, year = {2018}, date = {2018-04-13}, booktitle = {Proceedings of the 21st International Conference on Hybrid Systems: Computation and Control}, pages = {101-110}, abstract = {This paper deals with input/output-to-state stability (IOSS) of continuous-time switched nonlinear systems under restricted switching. The switching signals obey restrictions on: (i) transitions between subsystems, and (ii) dwell time on subsystems. Given a family of systems, possibly containing non-IOSS dynamics, we present an algorithm to construct a time-dependent switching signal that guarantees IOSS of the resulting switched system under these restrictions. The main apparatus for our analysis are multiple Lyapunov-like functions and graph-theoretic tools.}, keywords = {Control and Optimisation and Policy}, pubstate = {published}, tppubtype = {conference} } This paper deals with input/output-to-state stability (IOSS) of continuous-time switched nonlinear systems under restricted switching. The switching signals obey restrictions on: (i) transitions between subsystems, and (ii) dwell time on subsystems. Given a family of systems, possibly containing non-IOSS dynamics, we present an algorithm to construct a time-dependent switching signal that guarantees IOSS of the resulting switched system under these restrictions. The main apparatus for our analysis are multiple Lyapunov-like functions and graph-theoretic tools. |
| 20. | Yaji, Vinayaka G; Bhatnagar, Shalabh Stochastic recursive inclusions with non-additive iterate-dependent Markov noise Journal Article Stochastics. An International Journal of Probability and Stochastic Processes, 90 (3), pp. 330-363, 2018. Abstract | BibTeX | Tags: Computer Science and Automation | Links: @article{Yaji2018, title = {Stochastic recursive inclusions with non-additive iterate-dependent Markov noise}, author = {Vinayaka G. Yaji and Shalabh Bhatnagar}, doi = {10.1080/17442508.2017.1353984}, year = {2018}, date = {2018-03-30}, journal = {Stochastics. An International Journal of Probability and Stochastic Processes}, volume = {90}, number = {3}, pages = {330-363}, abstract = {In this paper we study the asymptotic behaviour of stochastic approximation schemes with set-valued drift function and non-additive iterate-dependent Markov noise. We show that a linearly interpolated trajectory of such a recursion is an asymptotic pseudotrajectory for the flow of a limiting differential inclusion obtained by averaging the set-valued drift function of the recursion w.r.t. the stationary distributions of the Markov noise. The limit set theorem by Benaim is then used to characterize the limit sets of the recursion in terms of the dynamics of the limiting differential inclusion. We then state two variants of the Markov noise assumption under which the analysis of the recursion is similar to the one presented in this paper. Scenarios where our recursion naturally appears are presented as applications. These include controlled stochastic approximation, subgradient descent, approximate drift problem and analysis of discontinuous dynamics all in the presence of non-additive iterate-dependent Markov noise.}, keywords = {Computer Science and Automation}, pubstate = {published}, tppubtype = {article} } In this paper we study the asymptotic behaviour of stochastic approximation schemes with set-valued drift function and non-additive iterate-dependent Markov noise. We show that a linearly interpolated trajectory of such a recursion is an asymptotic pseudotrajectory for the flow of a limiting differential inclusion obtained by averaging the set-valued drift function of the recursion w.r.t. the stationary distributions of the Markov noise. The limit set theorem by Benaim is then used to characterize the limit sets of the recursion in terms of the dynamics of the limiting differential inclusion. We then state two variants of the Markov noise assumption under which the analysis of the recursion is similar to the one presented in this paper. Scenarios where our recursion naturally appears are presented as applications. These include controlled stochastic approximation, subgradient descent, approximate drift problem and analysis of discontinuous dynamics all in the presence of non-additive iterate-dependent Markov noise. |
| 21. | Karthik, Girija Ramesan; Ghosh, Prasanta Kumar Subband selection for binaural speech source localization Conference Proceedings of the 18th Annual Conference of the International Speech Communication Association (INTERSPEECH): Situated Interaction, 20.-24.08.17, Stockholm (Sweden), 2018. Abstract | BibTeX | Tags: Electrical Engineering | Links: @conference{Karthik2018, title = {Subband selection for binaural speech source localization}, author = {Girija Ramesan Karthik and Prasanta Kumar Ghosh}, url = {http://www.rbccps.org/wp-content/uploads/2018/08/c330f2de2a9b57bad9b268c85d9a77b1742d.pdf}, doi = {10.21437/Interspeech.2017-954}, year = {2018}, date = {2018-02-28}, booktitle = {Proceedings of the 18th Annual Conference of the International Speech Communication Association (INTERSPEECH): Situated Interaction, 20.-24.08.17, Stockholm (Sweden)}, pages = {1929-1933}, abstract = {We consider the task of speech source localization using bin- aural cues, namely interaural time and level difference (ITD & ILD). A typical approach is to process binaural speech us- ing gammatone filters and calculate frame-level ITD and ILD in each subband. The ITD, ILD and their combination (ITLD) in each subband are statistically modelled using Gaussian mix- ture models for every direction during training. Given a binaural test-speech, the source is localized using maximum likelihood criterion assuming that the binaural cues in each subband are in- dependent. We, in this work, investigate the robustness of each subband for localization and compare their performance against the full-band scheme with 32 gammatone filters. We propose a subband selection procedure using the training data where sub- bands are rank ordered based on their localization performance. Experiments on Subject 003 from the CIPIC database reveal that, for high SNRs, the ITD and ITLD of just one subband centered at 296Hz is sufficient to yield localization accuracy identical to that of the full-band scheme with a test-speech of duration 1sec. At low SNRs, in case of ITD, the selected sub- bands are found to perform better than the full-band scheme.}, keywords = {Electrical Engineering}, pubstate = {published}, tppubtype = {conference} } We consider the task of speech source localization using bin- aural cues, namely interaural time and level difference (ITD & ILD). A typical approach is to process binaural speech us- ing gammatone filters and calculate frame-level ITD and ILD in each subband. The ITD, ILD and their combination (ITLD) in each subband are statistically modelled using Gaussian mix- ture models for every direction during training. Given a binaural test-speech, the source is localized using maximum likelihood criterion assuming that the binaural cues in each subband are in- dependent. We, in this work, investigate the robustness of each subband for localization and compare their performance against the full-band scheme with 32 gammatone filters. We propose a subband selection procedure using the training data where sub- bands are rank ordered based on their localization performance. Experiments on Subject 003 from the CIPIC database reveal that, for high SNRs, the ITD and ITLD of just one subband centered at 296Hz is sufficient to yield localization accuracy identical to that of the full-band scheme with a test-speech of duration 1sec. At low SNRs, in case of ITD, the selected sub- bands are found to perform better than the full-band scheme. |
| 22. | Karmakar, Prasenjit; Bhatnagar, Shalabh Two time-scale stochastic approximation with controlled Markov noise and off-policy temporal difference learning Journal Article Mathematics of Operations Research, 43 (1), pp. 130-151, 2018. Abstract | BibTeX | Tags: Computer Science and Automation | Links: @article{Karmakar2018, title = {Two time-scale stochastic approximation with controlled Markov noise and off-policy temporal difference learning}, author = {Prasenjit Karmakar and Shalabh Bhatnagar}, url = {http://www.rbccps.org/wp-content/uploads/2018/08/moor.2017.0855.pdf}, doi = {0.1287/moor.2017.0855}, year = {2018}, date = {2018-02-28}, journal = {Mathematics of Operations Research}, volume = {43}, number = {1}, pages = {130-151}, abstract = {We present for the first time an asymptotic convergence analysis of two time- scale stochastic approximation driven by “controlled” Markov noise. In particular, the faster and slower recursions have nonadditive controlled Markov noise components in addition to martingale difference noise. We analyze the asymptotic behavior of our framework by relating it to limiting differential inclusions in both time scales that are defined in terms of the ergodic occupation measures associated with the controlled Markov processes. Finally, we present a solution to the off-policy convergence problem for temporal-difference learning with linear function approximation, using our results.}, keywords = {Computer Science and Automation}, pubstate = {published}, tppubtype = {article} } We present for the first time an asymptotic convergence analysis of two time- scale stochastic approximation driven by “controlled” Markov noise. In particular, the faster and slower recursions have nonadditive controlled Markov noise components in addition to martingale difference noise. We analyze the asymptotic behavior of our framework by relating it to limiting differential inclusions in both time scales that are defined in terms of the ergodic occupation measures associated with the controlled Markov processes. Finally, we present a solution to the off-policy convergence problem for temporal-difference learning with linear function approximation, using our results. |
| 23. | Karthika, P S; Aparna, P M; Verma, Ashish Understanding crowd dynamics at Ghat regions during world's largest mass religious gathering, Kumbh Mela Journal Article International Journal of Disaster Risk Reduction, 31 , pp. 918-925, 2018. Abstract | BibTeX | Tags: The Kumbh Mela Experiment | Links: @article{Karthika2018, title = {Understanding crowd dynamics at Ghat regions during world's largest mass religious gathering, Kumbh Mela}, author = {P. S. Karthika and P. M. Aparna and Ashish Verma}, doi = {10.1016/j.ijdrr.2018.08.005}, year = {2018}, date = {2018-10-01}, journal = {International Journal of Disaster Risk Reduction}, volume = {31}, pages = {918-925}, abstract = {In this paper a porous flow approach on a cul de sac is proposed to understand the dynamics of crowd at ghat regions (banks of the sacred river) in mass religious gatherings. Kumbh Mela, one of the mankind's largest religious gathering encompassing almost all possible crowd scenarios, provides a unique opportunity to explore the crowd dynamics along all facets. Here, Cul-de-sac refers to the ghat region where people gather with the intention to take holy dip. The data used for this study was collected during Kumbh Mela held during 22nd April to 21st May 2016. Visual observations from the video data shows a high degree of complexity probably due to the nature of activities at the study location, e.g., lane formation, creeping behavior etc. The proposed porous flow approach divides the entire study area into pores, and it is assumed that pilgrims traverse this network through interconnected vacant pores. The pedestrian data from video sequences (entry time, exit time, and flow) is extracted manually and time series analysis of pore occupancy is done to get an approximate measure of local density. Further, using Poisson regression analysis it was found that both the inflow and the duration of holy dip are significant factors in influencing the number of arrivals into the pore. Since behavioral aspects of a pedestrian is a significant governing factor of crowd dynamics, these microscopic parameters can be used to get a measure of criticality of the system in terms of crowd risk.}, keywords = {The Kumbh Mela Experiment}, pubstate = {published}, tppubtype = {article} } In this paper a porous flow approach on a cul de sac is proposed to understand the dynamics of crowd at ghat regions (banks of the sacred river) in mass religious gatherings. Kumbh Mela, one of the mankind's largest religious gathering encompassing almost all possible crowd scenarios, provides a unique opportunity to explore the crowd dynamics along all facets. Here, Cul-de-sac refers to the ghat region where people gather with the intention to take holy dip. The data used for this study was collected during Kumbh Mela held during 22nd April to 21st May 2016. Visual observations from the video data shows a high degree of complexity probably due to the nature of activities at the study location, e.g., lane formation, creeping behavior etc. The proposed porous flow approach divides the entire study area into pores, and it is assumed that pilgrims traverse this network through interconnected vacant pores. The pedestrian data from video sequences (entry time, exit time, and flow) is extracted manually and time series analysis of pore occupancy is done to get an approximate measure of local density. Further, using Poisson regression analysis it was found that both the inflow and the duration of holy dip are significant factors in influencing the number of arrivals into the pore. Since behavioral aspects of a pedestrian is a significant governing factor of crowd dynamics, these microscopic parameters can be used to get a measure of criticality of the system in terms of crowd risk. |
2017 |
|
| 24. | Sekuboyina, Anjany Kumar; Devarakonda, Surya Teja; Seelamantula, Chandra Sekhar A convolutional neural network approach for abnormality detection in wireless capsule endoscopy Conference Proceedings of the 2017 IEEE International Symposium on Biomedical Imaging, 18.-21.04.2017, Melbourne (Australia), 2017. Abstract | BibTeX | Tags: Cyber Surgery and Remote Patient Care, Health | Links: @conference{Sekuboyina2017, title = {A convolutional neural network approach for abnormality detection in wireless capsule endoscopy}, author = {Anjany Kumar Sekuboyina and Surya Teja Devarakonda and Chandra Sekhar Seelamantula}, url = {http://www.rbccps.org/wp-content/uploads/2017/10/07950698.pdf}, doi = {10.1109/ISBI.2017.7950698}, year = {2017}, date = {2017-06-19}, booktitle = {Proceedings of the 2017 IEEE International Symposium on Biomedical Imaging, 18.-21.04.2017, Melbourne (Australia)}, pages = {1057-1060}, abstract = {In wireless capsule endoscopy (WCE), a swallowable miniature optical endoscope is used to transmit color images of the gastrointestinal tract. However, the number of images transmitted is large, taking a significant amount of the medical expert’s time to review the scan. In this paper, we propose a technique to automate the abnormality detection in WCE images. We split the image into several patches and extract features pertaining to each block using a convolutional neural network (CNN) to increase their generality while overcoming the drawbacks of manually crafted features. We intend to exploit the importance of color information for the task. Experiments are performed to determine the optimal color space components for feature extraction and classifier design. We ob- tained an area under receiver-operating-characteristic (ROC) curve of approximately 0.8 on a dataset containing multiple abnormalities, namely, aphthae, bleeding, chylous cysts, lymphangiectasias, polypoids, stenoses, ulcers, and villous oedema.}, keywords = {Cyber Surgery and Remote Patient Care, Health}, pubstate = {published}, tppubtype = {conference} } In wireless capsule endoscopy (WCE), a swallowable miniature optical endoscope is used to transmit color images of the gastrointestinal tract. However, the number of images transmitted is large, taking a significant amount of the medical expert’s time to review the scan. In this paper, we propose a technique to automate the abnormality detection in WCE images. We split the image into several patches and extract features pertaining to each block using a convolutional neural network (CNN) to increase their generality while overcoming the drawbacks of manually crafted features. We intend to exploit the importance of color information for the task. Experiments are performed to determine the optimal color space components for feature extraction and classifier design. We ob- tained an area under receiver-operating-characteristic (ROC) curve of approximately 0.8 on a dataset containing multiple abnormalities, namely, aphthae, bleeding, chylous cysts, lymphangiectasias, polypoids, stenoses, ulcers, and villous oedema. |
| 25. | Bhargava, Srivatsa; Gorur, Pushkar; Amrutur, Bharadwaj A distributed object detector-tracker aided video encoder for smart camera networks Conference Proceedings of the 11th International Conference on Distributed Smart Cameras (ICDSC), 05.-07.09.17, Stanford (USA), 2017. Abstract | BibTeX | Tags: Efficient architectures and algorithms for distributed video analytics for Smart Cities, Modeling and Simulation and Analytics | Links: @conference{Srivatsa2017, title = {A distributed object detector-tracker aided video encoder for smart camera networks}, author = {Srivatsa Bhargava and Pushkar Gorur and Bharadwaj Amrutur}, doi = {10.1145/3131885.3131920}, year = {2017}, date = {2017-09-07}, booktitle = {Proceedings of the 11th International Conference on Distributed Smart Cameras (ICDSC), 05.-07.09.17, Stanford (USA)}, pages = {69-75}, abstract = {In this paper, we propose a Region of Interest (ROI) modulated H.264 video encoder system, based on a distributed object detector-tracker framework, for smart camera networks. Locations of objects of interest, as determined by detector-tracker are used to semantically partition each frame into regions assigned with multiple levels of importance. A distributed architecture is proposed to implement the object detector-tracker framework to mitigate the computational cost. Further, a rate control algorithm with modified Rate-Distortion(RD) cost is proposed to determine Quantization Parameter(QP) and skip decision of Macro Blocks based on their relative levels of importance. Our experiments show that, the proposed system achieves upto 3x reduction in bitrate without significant reduction in PSNR of ROI(head-shoulder region of pedestrians). We also demonstrate the trade-off between total computational cost and compression possible with the proposed distributed detector-tracker framework. }, keywords = {Efficient architectures and algorithms for distributed video analytics for Smart Cities, Modeling and Simulation and Analytics}, pubstate = {published}, tppubtype = {conference} } In this paper, we propose a Region of Interest (ROI) modulated H.264 video encoder system, based on a distributed object detector-tracker framework, for smart camera networks. Locations of objects of interest, as determined by detector-tracker are used to semantically partition each frame into regions assigned with multiple levels of importance. A distributed architecture is proposed to implement the object detector-tracker framework to mitigate the computational cost. Further, a rate control algorithm with modified Rate-Distortion(RD) cost is proposed to determine Quantization Parameter(QP) and skip decision of Macro Blocks based on their relative levels of importance. Our experiments show that, the proposed system achieves upto 3x reduction in bitrate without significant reduction in PSNR of ROI(head-shoulder region of pedestrians). We also demonstrate the trade-off between total computational cost and compression possible with the proposed distributed detector-tracker framework. |
| 26. | Ramaswamy, Arunselvan; Bhatnagar, Shalabh A generalization of the Borkar-Meyn theorem for stochastic recursive inclusions Journal Article Mathematics of Operations Research, 42 (3), pp. 648-661, 2017. Abstract | BibTeX | Tags: Control and Optimisation and Policy | Links: @article{Ramaswamy2017, title = {A generalization of the Borkar-Meyn theorem for stochastic recursive inclusions}, author = {Arunselvan Ramaswamy and Shalabh Bhatnagar}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/moor.2016.0821.pdf}, doi = {10.1287/moor.2016.0821}, year = {2017}, date = {2017-08-31}, journal = {Mathematics of Operations Research}, volume = {42}, number = {3}, pages = {648-661}, abstract = {In this paper, the stability theorem of Borkar and Meyn is extended to include the case when the mean field is a set-valued map. Two different sets of sufficient conditions are presented that guarantee the “stability and convergence” of stochastic recursive inclusions. Our work builds on the works of Benaïm, Hofbauer and Sorin as well as Borkar and Meyn. As a corollary to one of the main theorems, a natural generalization of the Borkar and Meyn theorem follows. In addition, the original theorem of Borkar and Meyn is shown to hold under slightly relaxed assumptions. As an application to one of the main theorems, we discuss a solution to the “approximate drift problem.” Finally, we analyze the stochastic gradient algorithm with “constant-error gradient estimators” as yet another application of our main result.}, keywords = {Control and Optimisation and Policy}, pubstate = {published}, tppubtype = {article} } In this paper, the stability theorem of Borkar and Meyn is extended to include the case when the mean field is a set-valued map. Two different sets of sufficient conditions are presented that guarantee the “stability and convergence” of stochastic recursive inclusions. Our work builds on the works of Benaïm, Hofbauer and Sorin as well as Borkar and Meyn. As a corollary to one of the main theorems, a natural generalization of the Borkar and Meyn theorem follows. In addition, the original theorem of Borkar and Meyn is shown to hold under slightly relaxed assumptions. As an application to one of the main theorems, we discuss a solution to the “approximate drift problem.” Finally, we analyze the stochastic gradient algorithm with “constant-error gradient estimators” as yet another application of our main result. |
| 27. | Rathod, Nihesh; Subramanian, Renu; Sundaresan, Rajesh A GIS- and measurement-aided framework for placement of relays in a heterogeneous propagation environment Conference Forthcoming Proceedings of the 9th International Conference on Communication Systems and Networks (COMSNETS), 04.-08.01.17, Bangalore, Forthcoming. BibTeX | Tags: Design of large-scale IoT networks, Smart Campus Water Management: Closing the IoT Loop from Network to Knowledge @conference{Rathod2017, title = {A GIS- and measurement-aided framework for placement of relays in a heterogeneous propagation environment}, author = {Nihesh Rathod and Renu Subramanian and Rajesh Sundaresan}, year = {2017}, date = {2017-01-08}, booktitle = {Proceedings of the 9th International Conference on Communication Systems and Networks (COMSNETS), 04.-08.01.17, Bangalore}, keywords = {Design of large-scale IoT networks, Smart Campus Water Management: Closing the IoT Loop from Network to Knowledge}, pubstate = {forthcoming}, tppubtype = {conference} } |
| 28. | Joseph, Ajin George; Bhatnagar, Shalabh A model based search method for prediction in model-free Markov decision process Conference Proceedings of the 2017 International Joint Conference on Neural Networks (IJCNN), 14.-19.05.17, Anchorage (USA), 2017. Abstract | BibTeX | Tags: Control and Optimisation and Policy | Links: @conference{Joseph2017, title = {A model based search method for prediction in model-free Markov decision process}, author = {Ajin George Joseph and Shalabh Bhatnagar}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/07965851.pdf}, doi = {10.1109/IJCNN.2017.7965851}, year = {2017}, date = {2017-07-03}, booktitle = {Proceedings of the 2017 International Joint Conference on Neural Networks (IJCNN), 14.-19.05.17, Anchorage (USA)}, pages = {170-177}, abstract = {In this paper, we provide a new algorithm for the problem of prediction in the model-free MDP setting, i.e., estimating the value function of a given policy using the linear function approximation architecture, with memory and computation costs scaling quadratically in the size of the feature set. The algorithm is a multi-timescale variant of the very popular cross entropy (CE) method which is a model based search method to find the global optimum of a real-valued function. This is the first time a model based search method is used for the prediction problem. A proof of convergence using the ODE method is provided. The theoretical results are supplemented with experimental comparisons. The algorithm achieves good performance fairly consistently on many benchmark problems.}, keywords = {Control and Optimisation and Policy}, pubstate = {published}, tppubtype = {conference} } In this paper, we provide a new algorithm for the problem of prediction in the model-free MDP setting, i.e., estimating the value function of a given policy using the linear function approximation architecture, with memory and computation costs scaling quadratically in the size of the feature set. The algorithm is a multi-timescale variant of the very popular cross entropy (CE) method which is a model based search method to find the global optimum of a real-valued function. This is the first time a model based search method is used for the prediction problem. A proof of convergence using the ODE method is provided. The theoretical results are supplemented with experimental comparisons. The algorithm achieves good performance fairly consistently on many benchmark problems. |
| 29. | Gayarthi, H; Aparna, P M; Verma, Ashish A review of studies on understanding crowd dynamics in the context of crowd safety in mass religious gatherings Journal Article International Journal of Disaster Risk Reduction, 2017. Abstract | BibTeX | Tags: The Kumbh Mela Experiment | Links: @article{Gayarthi2017, title = {A review of studies on understanding crowd dynamics in the context of crowd safety in mass religious gatherings}, author = {H. Gayarthi and P. M. Aparna and Ashish Verma}, url = {http://www.rbccps.org/wp-content/uploads/2017/10/1-s2.0-S2212420917302212-main.pdf}, doi = {10.1016/j.ijdrr.2017.07.017}, year = {2017}, date = {2017-08-03}, journal = {International Journal of Disaster Risk Reduction}, abstract = {Understanding the principles and applications of crowd dynamics in mass gatherings is very important, specifically with respect to crowd risk analysis and crowd safety. Historical trends from India and other countries suggest that the stampedes in mass gatherings, especially in religious events occur frequently highlighting the importance of studying the crowd behaviour more scientifically. This is required to support appropriate and timely crowd management principles, in the planning of crowd control measures and provision of early warning systems at mass gatherings. Common pedestrian behaviours in crowd like group formation, self-organization, leader follower effect, queue formation and bottleneck conditions have substantial influence on crowd dynamics. It is important not to let a single aspect go overlooked with respect to mass gatherings since it can lead to major stampedes. Kumbh Mela, one of the largest mass religious gatherings in the world, features these different crowd scenarios observed often in the same event area and thus provides a unique opportunity to study the crowd behaviour in a holistic way. Understanding these pedestrian behaviours and having a clear understanding of the normal behaviour may provide opportunities to change crowd dynamics and overcome the adverse effects resulting in safer mass religious gatherings in future. This paper provides an exhaustive review of the current understanding of crowd dynamics and explores the modelling techniques that are available to enhance crowd safety. The purpose of this literature review is to improve the understanding of crowd dynamics,and highlight the research gaps in the context of crowd safety in mass religious gatherings like Kumbh Mela.}, keywords = {The Kumbh Mela Experiment}, pubstate = {published}, tppubtype = {article} } Understanding the principles and applications of crowd dynamics in mass gatherings is very important, specifically with respect to crowd risk analysis and crowd safety. Historical trends from India and other countries suggest that the stampedes in mass gatherings, especially in religious events occur frequently highlighting the importance of studying the crowd behaviour more scientifically. This is required to support appropriate and timely crowd management principles, in the planning of crowd control measures and provision of early warning systems at mass gatherings. Common pedestrian behaviours in crowd like group formation, self-organization, leader follower effect, queue formation and bottleneck conditions have substantial influence on crowd dynamics. It is important not to let a single aspect go overlooked with respect to mass gatherings since it can lead to major stampedes. Kumbh Mela, one of the largest mass religious gatherings in the world, features these different crowd scenarios observed often in the same event area and thus provides a unique opportunity to study the crowd behaviour in a holistic way. Understanding these pedestrian behaviours and having a clear understanding of the normal behaviour may provide opportunities to change crowd dynamics and overcome the adverse effects resulting in safer mass religious gatherings in future. This paper provides an exhaustive review of the current understanding of crowd dynamics and explores the modelling techniques that are available to enhance crowd safety. The purpose of this literature review is to improve the understanding of crowd dynamics,and highlight the research gaps in the context of crowd safety in mass religious gatherings like Kumbh Mela. |
| 30. | Lakshminarayanan, Chandrashekar; Bhatnagar, Shalabh A stability criterion for two-timescale stochastic approximation schemes Journal Article Automatica, 79 , pp. 108-114, 2017. Abstract | BibTeX | Tags: Control and Optimisation and Policy | Links: @article{Lakshminarayanan2017, title = {A stability criterion for two-timescale stochastic approximation schemes}, author = {Chandrashekar Lakshminarayanan and Shalabh Bhatnagar}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/1-s2.0-S0005109816305222-main.pdf}, doi = {10.1016/j.automatica.2016.12.014}, year = {2017}, date = {2017-05-31}, journal = {Automatica}, volume = {79}, pages = {108-114}, abstract = {We present the first sufficient conditions that guarantee stability of two-timescale stochastic approxima- tion schemes. Our analysis is based on the ordinary differential equation (ODE) method and is an extension of the results in Borkar and Meyn (2000) for single-timescale schemes. As an application of our result, we show the stability of iterates in a two-timescale stochastic approximation scheme arising in reinforce- ment learning.}, keywords = {Control and Optimisation and Policy}, pubstate = {published}, tppubtype = {article} } We present the first sufficient conditions that guarantee stability of two-timescale stochastic approxima- tion schemes. Our analysis is based on the ordinary differential equation (ODE) method and is an extension of the results in Borkar and Meyn (2000) for single-timescale schemes. As an application of our result, we show the stability of iterates in a two-timescale stochastic approximation scheme arising in reinforce- ment learning. |
| 31. | A., Prashanth L; Bhatnagar, Shalabh; Fu, Michael; Marcus, Steve Adaptive system optimization using random directions stochastic approximation Journal Article IEEE Transactions on Automatic Control, 62 (5), pp. 2223-2238, 2017. Abstract | BibTeX | Tags: Control and Optimisation and Policy | Links: @article{Prashanth2017, title = {Adaptive system optimization using random directions stochastic approximation}, author = {Prashanth L. A. and Shalabh Bhatnagar and Michael Fu and Steve Marcus}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/07544612.pdf}, doi = {10.1109/TAC.2016.2600643}, year = {2017}, date = {2017-05-31}, journal = {IEEE Transactions on Automatic Control}, volume = {62}, number = {5}, pages = {2223-2238}, abstract = {We present new algorithms for simulation op- timization using random directions stochastic approxima- tion (RDSA). These include first-order (gradient) as well as second-order (Newton) schemes. We incorporate both continuous-valued as well as discrete-valued perturbations into both types of algorithms. The former are chosen to be independent and identically distributed (i.i.d.) symmet- ric uniformly distributed random variables (r.v.), while the latter are i.i.d. asymmetric Bernoulli r.v.s. Our Newton al- gorithm, with a novel Hessian estimation scheme, requires N -dimensional perturbations and three loss measurements per iteration, whereas the simultaneous perturbation New- ton search algorithm of [1] requires 2N-dimensional per- turbations and four loss measurements per iteration. We prove the asymptotic unbiasedness of both gradient and Hessian estimates and asymptotic (strong) convergence for both first-order and second-order schemes. We also pro- vide asymptotic normality results, which in particular estab- lish that the asymmetric Bernoulli variant of Newton RDSA method is better than 2SPSA of [1]. Numerical experiments are used to validate the theoretical results.}, keywords = {Control and Optimisation and Policy}, pubstate = {published}, tppubtype = {article} } We present new algorithms for simulation op- timization using random directions stochastic approxima- tion (RDSA). These include first-order (gradient) as well as second-order (Newton) schemes. We incorporate both continuous-valued as well as discrete-valued perturbations into both types of algorithms. The former are chosen to be independent and identically distributed (i.i.d.) symmet- ric uniformly distributed random variables (r.v.), while the latter are i.i.d. asymmetric Bernoulli r.v.s. Our Newton al- gorithm, with a novel Hessian estimation scheme, requires N -dimensional perturbations and three loss measurements per iteration, whereas the simultaneous perturbation New- ton search algorithm of [1] requires 2N-dimensional per- turbations and four loss measurements per iteration. We prove the asymptotic unbiasedness of both gradient and Hessian estimates and asymptotic (strong) convergence for both first-order and second-order schemes. We also pro- vide asymptotic normality results, which in particular estab- lish that the asymmetric Bernoulli variant of Newton RDSA method is better than 2SPSA of [1]. Numerical experiments are used to validate the theoretical results. |
| 32. | Joseph, Ajin George; Bhatnagar, Shalabh An incremental fast policy search using a single sample path Conference Proceedings of the 7th International Conference on Pattern Recognition and Machine Intelligence (PReMI), 05.-08.12.17, Kolkata (India), Lecture Notes in Computer Science 2017. Abstract | BibTeX | Tags: Machine Learning | Links: @conference{Joseph2017c, title = {An incremental fast policy search using a single sample path}, author = {Ajin George Joseph and Shalabh Bhatnagar}, doi = {10.1007/978-3-319-69900-4_1}, year = {2017}, date = {2017-11-01}, booktitle = {Proceedings of the 7th International Conference on Pattern Recognition and Machine Intelligence (PReMI), 05.-08.12.17, Kolkata (India)}, pages = {3-10}, series = {Lecture Notes in Computer Science}, abstract = {In this paper, we consider the control problem in a reinforcement learning setting with large state and action spaces. The control problem most commonly addressed in the contemporary literature is to find an optimal policy which optimizes the long run γ-discounted transition costs, where γ∈[0,1). They also assume access to a generative model/simulator of the underlying MDP with the hidden premise that realization of the system dynamics of the MDP for arbitrary policies in the form of sample paths can be obtained with ease from the model. In this paper, we consider a cost function which is the expectation of a approximate value function w.r.t. the steady state distribution of the Markov chain induced by the policy, without having access to the generative model. We assume that a single sample path generated using a priori chosen behaviour policy is made available. In this information restricted setting, we solve the generalized control problem using the incremental cross entropy method. The proposed algorithm is shown to converge to the solution which is globally optimal relative to the behaviour policy.}, keywords = {Machine Learning}, pubstate = {published}, tppubtype = {conference} } In this paper, we consider the control problem in a reinforcement learning setting with large state and action spaces. The control problem most commonly addressed in the contemporary literature is to find an optimal policy which optimizes the long run γ-discounted transition costs, where γ∈[0,1). They also assume access to a generative model/simulator of the underlying MDP with the hidden premise that realization of the system dynamics of the MDP for arbitrary policies in the form of sample paths can be obtained with ease from the model. In this paper, we consider a cost function which is the expectation of a approximate value function w.r.t. the steady state distribution of the Markov chain induced by the policy, without having access to the generative model. We assume that a single sample path generated using a priori chosen behaviour policy is made available. In this information restricted setting, we solve the generalized control problem using the incremental cross entropy method. The proposed algorithm is shown to converge to the solution which is globally optimal relative to the behaviour policy. |
| 33. | Amrutur, Bharadwaj; Rajaraman, Vasanth; Acharya, Srikrishna; Ramesh, Rakshit; Joglekar, Ashish; Sharma, Abhay; Simmhan, Yogesh; Lele, Abhijit; Mahesh, Ashwin; Sankaran, Sathya An open Smart City IoT test bed: Street light poles as Smart City spines Conference Proceedings of the 2nd International Conference on Internet-of-Things Design and Implementation (IoTDI), 18.-21.04.17, Pittsburgh (USA), 2017. Abstract | BibTeX | Tags: Smart City, Systems Engineering | Links: @conference{Amrutur2017, title = {An open Smart City IoT test bed: Street light poles as Smart City spines}, author = {Bharadwaj Amrutur and Vasanth Rajaraman and Srikrishna Acharya and Rakshit Ramesh and Ashish Joglekar and Abhay Sharma and Yogesh Simmhan and Abhijit Lele and Ashwin Mahesh and Sathya Sankaran}, url = {http://www.rbccps.org/wp-content/uploads/2017/10/p323-Amrutur.pdf}, doi = {10.1145/3054977.3057303}, year = {2017}, date = {2017-06-15}, booktitle = {Proceedings of the 2nd International Conference on Internet-of-Things Design and Implementation (IoTDI), 18.-21.04.17, Pittsburgh (USA)}, pages = {323-324}, abstract = {Street light poles will be a key enabler for a smart city's hardware infrastructure, thanks to their ubiquity throughout the city as well as access to power. We propose an IoT test bed around light poles for the city, with a modular hardware and software architecture to enable experimentation with various technologies.}, keywords = {Smart City, Systems Engineering}, pubstate = {published}, tppubtype = {conference} } Street light poles will be a key enabler for a smart city's hardware infrastructure, thanks to their ubiquity throughout the city as well as access to power. We propose an IoT test bed around light poles for the city, with a modular hardware and software architecture to enable experimentation with various technologies. |
| 34. | Ramaswamy, Arunselvan; Bhatnagar, Shalabh Analysis of gradient descent methods with non-diminishing bounded errors Journal Article IEEE Transactions on Automatic Control, 2017. Abstract | BibTeX | Tags: Control and Optimisation and Policy | Links: @article{Ramaswamy2017b, title = {Analysis of gradient descent methods with non-diminishing bounded errors }, author = {Arunselvan Ramaswamy and Shalabh Bhatnagar}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/08016343.pdf}, doi = {10.1109/TAC.2017.2744598}, year = {2017}, date = {2017-08-24}, journal = {IEEE Transactions on Automatic Control}, abstract = {The main aim of this paper is to provide an analysis of gradient descent (GD) algorithms with gradient errors that do not necessarily vanish, asymptotically. In particular, sufficient conditions are presented for both stability (almost sure boundedness of the iterates) and convergence of GD with bounded, (possibly) non-diminishing gradient errors. In addition to ensuring stability, such an algorithm is shown to converge to a small neighborhood of the minimum set, which depends on the gradient errors. It is worth noting that the main result of this paper can be used to show that GD with asymptotically vanishing errors indeed converges to the minimum set. The results presented herein are not only more general when compared to previous results, but our analysis of GD with errors is new to the literature to the best of our knowledge. Our work extends the contributions of Mangasarian and Solodov, Bertsekas and Tsitsiklis and Tadic and Doucet. Using our framework, a simple yet effective implementation of GD using simultaneous perturbation stochastic approximations (SP SA), with constant sensitivity parameters, is presented. Another important improvement over many previous results is that there are no “additional” restrictions imposed on the step-sizes. In machine learning applications where step-sizes are related to learning rates, our assumptions, unlike those of other papers, do not affect these learning rates. Finally, we present experimental results to validate our theory.}, keywords = {Control and Optimisation and Policy}, pubstate = {published}, tppubtype = {article} } The main aim of this paper is to provide an analysis of gradient descent (GD) algorithms with gradient errors that do not necessarily vanish, asymptotically. In particular, sufficient conditions are presented for both stability (almost sure boundedness of the iterates) and convergence of GD with bounded, (possibly) non-diminishing gradient errors. In addition to ensuring stability, such an algorithm is shown to converge to a small neighborhood of the minimum set, which depends on the gradient errors. It is worth noting that the main result of this paper can be used to show that GD with asymptotically vanishing errors indeed converges to the minimum set. The results presented herein are not only more general when compared to previous results, but our analysis of GD with errors is new to the literature to the best of our knowledge. Our work extends the contributions of Mangasarian and Solodov, Bertsekas and Tsitsiklis and Tadic and Doucet. Using our framework, a simple yet effective implementation of GD using simultaneous perturbation stochastic approximations (SP SA), with constant sensitivity parameters, is presented. Another important improvement over many previous results is that there are no “additional” restrictions imposed on the step-sizes. In machine learning applications where step-sizes are related to learning rates, our assumptions, unlike those of other papers, do not affect these learning rates. Finally, we present experimental results to validate our theory. |
| 35. | Shyam, Pranav; Gupta, Shubham; Dukkipati, Ambedkar Attentive recurrent comparators Conference Proceedings of the 34th International Conference on Machine Learning (ICML 2017), 06.-11.08.17, Sydney (Australia), 70 , 2017. Abstract | BibTeX | Tags: CyberGut: A Bio-CPS approach to understand gut biology | Links: @conference{Shyam2017, title = {Attentive recurrent comparators}, author = {Pranav Shyam and Shubham Gupta and Ambedkar Dukkipati}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/shyam17a.pdf}, year = {2017}, date = {2017-08-11}, booktitle = {Proceedings of the 34th International Conference on Machine Learning (ICML 2017), 06.-11.08.17, Sydney (Australia)}, volume = {70}, pages = {3173-3181}, abstract = {Rapid learning requires flexible representations to quickly adopt to new evidence. We develop a novel class of models called Attentive Recurrent Comparators (ARCs) that form representations of objects by cycling through them and making observations. Using the representations extracted by ARCs, we develop a way of approximating a dynamic representation space and use it for one- shot learning. In the task of one-shot classifi- cation on the Omniglot dataset, we achieve the state of the art performance with an error rate of 1.5%. This represents the first super-human re- sult achieved for this task with a generic model that uses only pixel information.}, keywords = {CyberGut: A Bio-CPS approach to understand gut biology}, pubstate = {published}, tppubtype = {conference} } Rapid learning requires flexible representations to quickly adopt to new evidence. We develop a novel class of models called Attentive Recurrent Comparators (ARCs) that form representations of objects by cycling through them and making observations. Using the representations extracted by ARCs, we develop a way of approximating a dynamic representation space and use it for one- shot learning. In the task of one-shot classifi- cation on the Omniglot dataset, we achieve the state of the art performance with an error rate of 1.5%. This represents the first super-human re- sult achieved for this task with a generic model that uses only pixel information. |
| 36. | Krishnaswamy, S; P.T., Akhil; Arapostathis, A; Shakkottai, S; Sundaresan, Rajesh Augmenting max-weight with explicit learning for wireless scheduling with switching costs Conference Proceedings of the 2017 IEEE International Conference on Computer Communications (INFOCOM), 01.-04.05.17, Atlanta (USA), 2017. Abstract | BibTeX | Tags: Design of large-scale IoT networks | Links: @conference{Krishnaswamy2017, title = {Augmenting max-weight with explicit learning for wireless scheduling with switching costs}, author = {S. Krishnaswamy and Akhil P.T. and A. Arapostathis and S. Shakkottai and Rajesh Sundaresan}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/08056983.pdf}, doi = {10.1109/INFOCOM.2017.8056983}, year = {2017}, date = {2017-10-05}, booktitle = {Proceedings of the 2017 IEEE International Conference on Computer Communications (INFOCOM), 01.-04.05.17, Atlanta (USA)}, abstract = {In small-cell wireless networks where users are connected to multiple base stations (BSs), it is often advantageous to opportunistically switch off a subset of BSs to minimize energy costs. We consider two types of energy cost: (i) the cost of maintaining a BS in the active state, and (ii) the cost of switching a BS from the active state to inactive state. The problem is to operate the network at the lowest possible energy cost (sum of activation and switching costs) subject to queue stability. In this setting, the traditional approach – a Max-Weight algorithm along with a Lyapunov-based stability argument – does not suffice to show queue stability, essentially due to the temporal co-evolution between channel scheduling and the BS activation decisions induced by the switching cost. Instead, we develop a learning and BS activation algorithm with slow temporal dynamics, and a Max-Weight based channel scheduler that has fast temporal dynamics. We show using convergence of time-inhomogeneous Markov chains, that the co-evolving dynamics of learning, BS activation and queue lengths lead to near optimal average energy costs along with queue stability.}, keywords = {Design of large-scale IoT networks}, pubstate = {published}, tppubtype = {conference} } In small-cell wireless networks where users are connected to multiple base stations (BSs), it is often advantageous to opportunistically switch off a subset of BSs to minimize energy costs. We consider two types of energy cost: (i) the cost of maintaining a BS in the active state, and (ii) the cost of switching a BS from the active state to inactive state. The problem is to operate the network at the lowest possible energy cost (sum of activation and switching costs) subject to queue stability. In this setting, the traditional approach – a Max-Weight algorithm along with a Lyapunov-based stability argument – does not suffice to show queue stability, essentially due to the temporal co-evolution between channel scheduling and the BS activation decisions induced by the switching cost. Instead, we develop a learning and BS activation algorithm with slow temporal dynamics, and a Max-Weight based channel scheduler that has fast temporal dynamics. We show using convergence of time-inhomogeneous Markov chains, that the co-evolving dynamics of learning, BS activation and queue lengths lead to near optimal average energy costs along with queue stability. |
| 37. | Kadavankandy, Arun; Avrachenkov, Konstantin; Cottatellucci, Laura; Sundaresan, Rajesh Belief propagation for subgraph detection with imperfect side-information Conference Proceedings of the 2017 IEEE International Symposium on Information Theory (ISIT), 25.-30.06.17, Aachen (Germany), 2017. Abstract | BibTeX | Tags: Modeling and Simulation and Analytics | Links: @conference{Kadavankandy2017, title = {Belief propagation for subgraph detection with imperfect side-information}, author = {Arun Kadavankandy and Konstantin Avrachenkov and Laura Cottatellucci and Rajesh Sundaresan}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/08006800.pdf}, doi = {10.1109/ISIT.2017.8006800}, year = {2017}, date = {2017-08-15}, booktitle = {Proceedings of the 2017 IEEE International Symposium on Information Theory (ISIT), 25.-30.06.17, Aachen (Germany)}, pages = {1603-1607}, abstract = {We propose a local message passing algorithm based on Belief Propagation (BP) to detect a small hidden Erdos-Rényi (ER) subgraph embedded in a larger sparse ER random graph in the presence of side-information. We consider side-information in the form of revealed subgraph nodes called cues, some of which may be erroneous. Namely, the revealed nodes may not all belong to the subgraph, and it is not known to the algorithm a priori which cues are correct and which are incorrect. We show that asymptotically as the graph size tends to infinity, the expected fraction of misclassified nodes approaches zero for any positive value of a parameter λ, which represents the effective Signal-to-Noise Ratio of the detection problem. Previous works on subgraph detection using BP without side-information showed that BP fails to recover the subgraph when λ <; 1/e. Our results thus demonstrate the substantial gains in having even a small amount of side-information.}, keywords = {Modeling and Simulation and Analytics}, pubstate = {published}, tppubtype = {conference} } We propose a local message passing algorithm based on Belief Propagation (BP) to detect a small hidden Erdos-Rényi (ER) subgraph embedded in a larger sparse ER random graph in the presence of side-information. We consider side-information in the form of revealed subgraph nodes called cues, some of which may be erroneous. Namely, the revealed nodes may not all belong to the subgraph, and it is not known to the algorithm a priori which cues are correct and which are incorrect. We show that asymptotically as the graph size tends to infinity, the expected fraction of misclassified nodes approaches zero for any positive value of a parameter λ, which represents the effective Signal-to-Noise Ratio of the detection problem. Previous works on subgraph detection using BP without side-information showed that BP fails to recover the subgraph when λ <; 1/e. Our results thus demonstrate the substantial gains in having even a small amount of side-information. |
| 38. | Joseph, Ajin George; Bhatnagar, Shalabh Bounds for off-policy prediction in reinforcement learning Conference Proceedings of the 2017 International Joint Conference on Neural Networks (IJCNN), 14.-19.05.17, Anchorage (USA), 2017. Abstract | BibTeX | Tags: Control and Optimisation and Policy, Machine Learning | Links: @conference{Joseph2017b, title = {Bounds for off-policy prediction in reinforcement learning}, author = {Ajin George Joseph and Shalabh Bhatnagar}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/07966359.pdf}, doi = {10.1109/IJCNN.2017.7966359}, year = {2017}, date = {2017-07-03}, booktitle = {Proceedings of the 2017 International Joint Conference on Neural Networks (IJCNN), 14.-19.05.17, Anchorage (USA)}, pages = {3991-3997}, abstract = {In this paper, we provide for the first time, error bounds for the off-policy prediction in reinforcement learning. The primary objective in off-policy prediction is to estimate the value function of a given target policy of interest using the linear function approximation architecture by utilizing a sample trajectory generated by a behaviour policy which is possibly different from the target policy. The stability of the off-policy prediction has been an open question for a long time. Only recently, could Yu provide a generalized proof, which makes our results more appealing to the reinforcement learning community. The off-policy prediction is useful in complex reinforcement learning settings, where the sample trajectory is hard to obtain and one has to rely on the sample behaviour of the system with respect to an arbitrary policy. We provide here error bound on the solution of the off-policy prediction with respect to a closeness measure between the target and the behaviour policy.}, keywords = {Control and Optimisation and Policy, Machine Learning}, pubstate = {published}, tppubtype = {conference} } In this paper, we provide for the first time, error bounds for the off-policy prediction in reinforcement learning. The primary objective in off-policy prediction is to estimate the value function of a given target policy of interest using the linear function approximation architecture by utilizing a sample trajectory generated by a behaviour policy which is possibly different from the target policy. The stability of the off-policy prediction has been an open question for a long time. Only recently, could Yu provide a generalized proof, which makes our results more appealing to the reinforcement learning community. The off-policy prediction is useful in complex reinforcement learning settings, where the sample trajectory is hard to obtain and one has to rely on the sample behaviour of the system with respect to an arbitrary policy. We provide here error bound on the solution of the off-policy prediction with respect to a closeness measure between the target and the behaviour policy. |
| 39. | Anjana, G R; Kumar, Mohan M S; Amrutur, Bharadwaj Comparison of different controllers for equitable water supply in water networks Conference Proceedings of the 15th International Computing & Control for the Water Industry Conference (CCWI), 05.-07.09.17, Sheffield (UK), 2017. Abstract | BibTeX | Tags: Water | Links: @conference{Anjana2017, title = {Comparison of different controllers for equitable water supply in water networks}, author = {G. R. Anjana and M. S. Mohan Kumar and Bharadwaj Amrutur}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/F108.pdf}, doi = {10.15131/shef.data.c.3867985.v1}, year = {2017}, date = {2017-09-07}, booktitle = {Proceedings of the 15th International Computing & Control for the Water Industry Conference (CCWI), 05.-07.09.17, Sheffield (UK)}, abstract = {Equitable distribution of water among consumers is one of the most important agenda for many water authorities around the world. Many methods have been listed in literature for equitable distribution of water. In recent years, different automatic controller based approaches were also introduced for flow control in water systems. In this work, a comparative study is carried out among dynamic inversion based PID (DI-PID), model predictive control (MPC) and EnKF based PID controllers for flow control applications in real time for a water distribution network – Bangalore inflow water distribution system. It was observed that the DI-PID and EnKF based PID controllers are better than other controllers for equitable distribution of water in a mega city like Bangalore.}, keywords = {Water}, pubstate = {published}, tppubtype = {conference} } Equitable distribution of water among consumers is one of the most important agenda for many water authorities around the world. Many methods have been listed in literature for equitable distribution of water. In recent years, different automatic controller based approaches were also introduced for flow control in water systems. In this work, a comparative study is carried out among dynamic inversion based PID (DI-PID), model predictive control (MPC) and EnKF based PID controllers for flow control applications in real time for a water distribution network – Bangalore inflow water distribution system. It was observed that the DI-PID and EnKF based PID controllers are better than other controllers for equitable distribution of water in a mega city like Bangalore. |
| 40. | Sharma, Abhay; Murthy, Chandra R Computationally tractable algorithms for finding a subset of non-defective items from a large population Journal Article IEEE Transactions on Information Theory, 63 (11), pp. 7149-7165, 2017. Abstract | BibTeX | Tags: Sensing and Actuation Systems | Links: @article{Sharma2017b, title = {Computationally tractable algorithms for finding a subset of non-defective items from a large population}, author = {Abhay Sharma and Chandra R. Murthy }, url = {http://www.rbccps.org/wp-content/uploads/2018/01/08024008.pdf}, doi = {10.1109/TIT.2017.2748143}, year = {2017}, date = {2017-09-01}, journal = {IEEE Transactions on Information Theory}, volume = {63}, number = {11}, pages = {7149-7165}, abstract = {In the classical non-adaptive group testing setup, pools of items are tested together, and the main goal of a recovery algorithm is to identify the complete defective set given the outcomes of different group tests. In contrast, the main goal of a non-defective subset recovery algorithm is to identify a subset of non-defective items given the test outcomes. In this paper, we present a suite of computationally efficient and analytically tractable non-defective subset recovery algorithms. By analyzing the probability of error of the algorithms, we obtain bounds on the number of tests required for non-defective subset recovery with arbitrarily small probability of error. Our analysis accounts for the impact of both the additive noise (false positives) and dilution noise (false negatives). By comparing with information theoretic lower bounds, we show that the upper bounds on the number of tests are orderwise tight up to a log^2 K factor, where K is the number of defective items. We also provide simulation results that compare the relative performance of the different algorithms and reveal insights into their practical utility. The proposed algorithms significantly outperform the straightforward approaches of testing items one-by-one, and of first identifying the defective set and then choosing the non-defective items from the complement set, in terms of the number of measurements required to ensure a given success rate.}, keywords = {Sensing and Actuation Systems}, pubstate = {published}, tppubtype = {article} } In the classical non-adaptive group testing setup, pools of items are tested together, and the main goal of a recovery algorithm is to identify the complete defective set given the outcomes of different group tests. In contrast, the main goal of a non-defective subset recovery algorithm is to identify a subset of non-defective items given the test outcomes. In this paper, we present a suite of computationally efficient and analytically tractable non-defective subset recovery algorithms. By analyzing the probability of error of the algorithms, we obtain bounds on the number of tests required for non-defective subset recovery with arbitrarily small probability of error. Our analysis accounts for the impact of both the additive noise (false positives) and dilution noise (false negatives). By comparing with information theoretic lower bounds, we show that the upper bounds on the number of tests are orderwise tight up to a log^2 K factor, where K is the number of defective items. We also provide simulation results that compare the relative performance of the different algorithms and reveal insights into their practical utility. The proposed algorithms significantly outperform the straightforward approaches of testing items one-by-one, and of first identifying the defective set and then choosing the non-defective items from the complement set, in terms of the number of measurements required to ensure a given success rate. |
| 41. | Sundaresan, Rajesh Developing a framework for using electricity consumption data to drive energy efficiency in the residential sector Technical Report 2017. BibTeX | Tags: Developing a framework for using electricity consumption data to drive energy efficiency in the residential sector, Energy | Links: @techreport{Sundaresan2017, title = {Developing a framework for using electricity consumption data to drive energy efficiency in the residential sector}, author = {Rajesh Sundaresan}, url = {http://www.rbccps.org/wp-content/uploads/2017/06/KeralaEnergyEfficiencyAluva.pdf}, year = {2017}, date = {2017-03-31}, keywords = {Developing a framework for using electricity consumption data to drive energy efficiency in the residential sector, Energy}, pubstate = {published}, tppubtype = {techreport} } |
| 42. | Bora, Ashish; Borkar, Vivek S; Garg, Dinesh; Sundaresan, Rajesh Edge conductance estimation using MCMC Conference Proceedings of the 54th Annual Allerton Conference on Communication, Control, and Computing, 27.-30.09.16, Monticello (USA), 2017. Abstract | BibTeX | Tags: Developing a framework for using electricity consumption data to drive energy efficiency in the residential sector | Links: @conference{Bora2016, title = {Edge conductance estimation using MCMC}, author = {Ashish Bora and Vivek S. Borkar and Dinesh Garg and Rajesh Sundaresan}, url = {http://www.rbccps.org/wp-content/uploads/2017/10/07852230.pdf}, doi = {10.1109/ALLERTON.2016.7852230}, year = {2017}, date = {2017-02-13}, booktitle = {Proceedings of the 54th Annual Allerton Conference on Communication, Control, and Computing, 27.-30.09.16, Monticello (USA)}, pages = {202-208}, abstract = {We propose an iterative and distributed Markov Chain Monte Carlo scheme for estimation of effective edge conductances in a graph. A sample complexity analysis is provided. The theoretical guarantees on the performance of the proposed algorithm are weak compared to those of existing algorithms. But numerical experiments suggest that the algorithm might still be effective while offering the advantages of low per iterate computation and memory requirements.}, keywords = {Developing a framework for using electricity consumption data to drive energy efficiency in the residential sector}, pubstate = {published}, tppubtype = {conference} } We propose an iterative and distributed Markov Chain Monte Carlo scheme for estimation of effective edge conductances in a graph. A sample complexity analysis is provided. The theoretical guarantees on the performance of the proposed algorithm are weak compared to those of existing algorithms. But numerical experiments suggest that the algorithm might still be effective while offering the advantages of low per iterate computation and memory requirements. |
| 43. | Khandelwal, T; Rajwanshi, K; Bharadwaj, P; Garani, S S; Sundaresan, Rajesh Exploiting appliance state constraints to improve appliance state detection Conference Proceedings of the 8th ACM International Conference on Future Energy Systems (ACM eEnergy), 16.-19.05.17, Hong Kong (China), (111-120), 2017. Abstract | BibTeX | Tags: Developing a framework for using electricity consumption data to drive energy efficiency in the residential sector | Links: @conference{Khandelwal2017, title = {Exploiting appliance state constraints to improve appliance state detection}, author = {T. Khandelwal and K. Rajwanshi and P. Bharadwaj and S. S. Garani and Rajesh Sundaresan}, url = {http://www.rbccps.org/wp-content/uploads/2017/10/p111-Khandelwal.pdf}, doi = {10.1145/3077839.3077859}, year = {2017}, date = {2017-05-19}, booktitle = {Proceedings of the 8th ACM International Conference on Future Energy Systems (ACM eEnergy), 16.-19.05.17, Hong Kong (China)}, number = {111-120}, abstract = {This work deals with non-intrusive load monitoring using a single inexpensive device at the mains. We argue that very low sampling rates (of 1 Hz) may suffice. This enables significant compression and cheaper end-devices. There are challenges when operating at such low sampling rates, of course. To achieve good appliance inference performance we propose improved event detection, feature extraction, and inference algorithms. The inference algorithm exploits state transition constraints and proposes the use of a maximum likelihood sequence detection for improved performance.}, keywords = {Developing a framework for using electricity consumption data to drive energy efficiency in the residential sector}, pubstate = {published}, tppubtype = {conference} } This work deals with non-intrusive load monitoring using a single inexpensive device at the mains. We argue that very low sampling rates (of 1 Hz) may suffice. This enables significant compression and cheaper end-devices. There are challenges when operating at such low sampling rates, of course. To achieve good appliance inference performance we propose improved event detection, feature extraction, and inference algorithms. The inference algorithm exploits state transition constraints and proposes the use of a maximum likelihood sequence detection for improved performance. |
| 44. | Vinerbi, Lorenzo; Babu, Arun Framework for automation of hazard log management on large critical projects Book Chapter Bondavalli, Andrea; Branca, Francesco (Ed.): Certifications of Critical Systems – The CECRIS Experience, Chapter 5, pp. 103-115, River Publishers, 2017, ISBN: 9788793519565. Abstract | BibTeX | Tags: Systems Engineering | Links: @inbook{Vinerbi2017, title = {Framework for automation of hazard log management on large critical projects}, author = {Lorenzo Vinerbi and Arun Babu}, editor = {Andrea Bondavalli and Francesco Branca}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/05_Chapter_05.pdf}, doi = {10.13052/rp-9788793519558}, isbn = {9788793519565}, year = {2017}, date = {2017-09-08}, booktitle = {Certifications of Critical Systems – The CECRIS Experience}, pages = {103-115}, publisher = {River Publishers}, chapter = {5}, abstract = {A hazard (HZ) is any situation that could cause harm to the system or lives. HZ depends on the system and its environment, and the probability of the HZ to cause harm is known as risk. HZs are analyzed by identifying their causes and the possible negative consequences that might ensue. For example, the dangerous failure of a traffic signal could be caused by a logic error in the traffic signaling controller’s software program. The consequence could be conflicting traffic flows simultaneously receiving green signals. A hazard log (HL) is a database of all risk management activities in a project. Maintaining its correctness and consistency on large safety/mission critical projects involving multiple vendors, suppliers, and partners is critical and challenging. IBM DOORS is one of the popular tool used for HZ management in mission critical applications. However, not all stake-holders are familiar with it. Also, it may not always feasible for all stake-holders to provide correct, well structured, and consistent HZ data. IBM DOORS have been reported to be useful in managing DO-178 compliance for avionics. Also, HL in DOORS allows capabilities for tracing requirements and test results. However, DOORS has steeper learning curve and is difficult to use by common people and beginners. Also, they lack validation capabilities. Custom checks may require difficult to use plug-ins which are not generic. This complexity makes it difficult to maintain the rules; preventing reuse in other projects. This chapter demonstrates a modular and extensible way to specify rules for checks locally at the stake-holder side, as well as while combining data from various parties to form the HL. The HZ-LOG automatization tool simplifies the process of HZ data collection on large projects to form the HL, while ensuring data consistency and correctness. The data provided by all parties are collected using a template containing scripts. The scripts check for mistakes/errors based on internal standards of company in charge of the HZ management. The collected data is then subjected to merging in DOORS, which also contain scripts to check and import data to form the HL. The requirements of HL tool are: (i) Perform checks of incoming data from vendors and partners; (ii) It shall allow to collect and keep log for all information related to iden- tified HZs (and related identified mitigations), structuring information accordingly; (iii) It shall be possible to manage the status of the HZs and related mitiga- tions, allowing for the control of risk. Only allowed HZ status transitions shall be possible and logging of the related status transition activity shall be kept in the tool for traceability purposes; (iv) Only RAMS specialist are allowed to manage HZs being necessary no different user profiles for the management of HZs in the tool; (v) A function of the tool shall allow to extract the “current” status of the project system HL by allowing the creation of documentary reports containing the set of necessary information about the predicted HZs, mitigations identified, and the status of all related risk control activities.}, keywords = {Systems Engineering}, pubstate = {published}, tppubtype = {inbook} } A hazard (HZ) is any situation that could cause harm to the system or lives. HZ depends on the system and its environment, and the probability of the HZ to cause harm is known as risk. HZs are analyzed by identifying their causes and the possible negative consequences that might ensue. For example, the dangerous failure of a traffic signal could be caused by a logic error in the traffic signaling controller’s software program. The consequence could be conflicting traffic flows simultaneously receiving green signals. A hazard log (HL) is a database of all risk management activities in a project. Maintaining its correctness and consistency on large safety/mission critical projects involving multiple vendors, suppliers, and partners is critical and challenging. IBM DOORS is one of the popular tool used for HZ management in mission critical applications. However, not all stake-holders are familiar with it. Also, it may not always feasible for all stake-holders to provide correct, well structured, and consistent HZ data. IBM DOORS have been reported to be useful in managing DO-178 compliance for avionics. Also, HL in DOORS allows capabilities for tracing requirements and test results. However, DOORS has steeper learning curve and is difficult to use by common people and beginners. Also, they lack validation capabilities. Custom checks may require difficult to use plug-ins which are not generic. This complexity makes it difficult to maintain the rules; preventing reuse in other projects. This chapter demonstrates a modular and extensible way to specify rules for checks locally at the stake-holder side, as well as while combining data from various parties to form the HL. The HZ-LOG automatization tool simplifies the process of HZ data collection on large projects to form the HL, while ensuring data consistency and correctness. The data provided by all parties are collected using a template containing scripts. The scripts check for mistakes/errors based on internal standards of company in charge of the HZ management. The collected data is then subjected to merging in DOORS, which also contain scripts to check and import data to form the HL. The requirements of HL tool are: (i) Perform checks of incoming data from vendors and partners; (ii) It shall allow to collect and keep log for all information related to iden- tified HZs (and related identified mitigations), structuring information accordingly; (iii) It shall be possible to manage the status of the HZs and related mitiga- tions, allowing for the control of risk. Only allowed HZ status transitions shall be possible and logging of the related status transition activity shall be kept in the tool for traceability purposes; (iv) Only RAMS specialist are allowed to manage HZs being necessary no different user profiles for the management of HZs in the tool; (v) A function of the tool shall allow to extract the “current” status of the project system HL by allowing the creation of documentary reports containing the set of necessary information about the predicted HZs, mitigations identified, and the status of all related risk control activities. |
| 45. | Sankaran, Raman; Bach, Francis R; Bhattacharyya, Chiranjib Identifying groups of strongly correlated variables through smoothed ordered weighted L1-norms Journal Article Proceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS), 20.-22.04.17, Fort Lauderdale (USA), pp. 1123-1131, 2017. Abstract | BibTeX | Tags: Machine Learning | Links: @article{Sankaran2017, title = {Identifying groups of strongly correlated variables through smoothed ordered weighted L1-norms}, author = {Raman Sankaran and Francis R. Bach and Chiranjib Bhattacharyya}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/sankaran17a.pdf}, year = {2017}, date = {2017-08-16}, journal = {Proceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS), 20.-22.04.17, Fort Lauderdale (USA)}, pages = {1123-1131}, abstract = {The failure of LASSO to identify groups of correlated predictors in linear regression has sparked significant research interest. Recently, various norms were proposed, which can be best described as instances of ordered weighted ℓ1 norms (OWL), as an alternative to ℓ1 regularization used in LASSO. OWL can identify groups of correlated variables but it forces the model to be constant within a group. This artifact induces unnecessary bias in the model estimation. In this paper we take a submodular perspective and show that OWL can be posed as the Lovász extension of a suitably defined submodular function. The submodular perspective not only explains the group-wise constant behavior of OWL, but also suggests alternatives. The main contribution of this paper is smoothed OWL (SOWL), a new family of norms, which not only identifies the groups but also allows the model to be flexible inside a group. We establish several algorithmic and theoretical properties of SOWL including group identification and model consistency. We also provide algorithmic tools to compute the SOWL norm and its proximal operator, whose computational complexity O(dlogd) is significantly better than that of general purpose solvers in O(d2logd). In our experiments, SOWL compares favorably with respect to OWL in the regimes of interest.}, keywords = {Machine Learning}, pubstate = {published}, tppubtype = {article} } The failure of LASSO to identify groups of correlated predictors in linear regression has sparked significant research interest. Recently, various norms were proposed, which can be best described as instances of ordered weighted ℓ1 norms (OWL), as an alternative to ℓ1 regularization used in LASSO. OWL can identify groups of correlated variables but it forces the model to be constant within a group. This artifact induces unnecessary bias in the model estimation. In this paper we take a submodular perspective and show that OWL can be posed as the Lovász extension of a suitably defined submodular function. The submodular perspective not only explains the group-wise constant behavior of OWL, but also suggests alternatives. The main contribution of this paper is smoothed OWL (SOWL), a new family of norms, which not only identifies the groups but also allows the model to be flexible inside a group. We establish several algorithmic and theoretical properties of SOWL including group identification and model consistency. We also provide algorithmic tools to compute the SOWL norm and its proximal operator, whose computational complexity O(dlogd) is significantly better than that of general purpose solvers in O(d2logd). In our experiments, SOWL compares favorably with respect to OWL in the regimes of interest. |
| 46. | Chakrabortty, Tuhin; Varma, Manoj Inferring existence of physical feedback mechanisms from short-lived synchronization events in complex systems Conference Forthcoming Proceedings of the 2017 Conference on Complex Systems, 17.-22.09.17, Cancun (Mexico), Forthcoming. BibTeX | Tags: Sensing and Actuation Systems @conference{Chakrabortty2017, title = {Inferring existence of physical feedback mechanisms from short-lived synchronization events in complex systems}, author = {Tuhin Chakrabortty and Manoj Varma}, year = {2017}, date = {2017-09-22}, booktitle = {Proceedings of the 2017 Conference on Complex Systems, 17.-22.09.17, Cancun (Mexico)}, keywords = {Sensing and Actuation Systems}, pubstate = {forthcoming}, tppubtype = {conference} } |
| 47. | Prasad, Oppili; Jha, Prateek; Pillai, Sreelal; Prasad, Mandya; Amrutur, Bharadwaj; Sambandan, Sanjiv Interconnects on elastomers: Optimizing for stretchability, speed and layout area Journal Article Flexible and Printed Electronics, 2 (4), 2017. Abstract | BibTeX | Tags: Systems Engineering | Links: @article{Prasad2017, title = {Interconnects on elastomers: Optimizing for stretchability, speed and layout area}, author = {Oppili Prasad and Prateek Jha and Sreelal Pillai and Mandya Prasad and Bharadwaj Amrutur and Sanjiv Sambandan}, url = {http://www.rbccps.org/wp-content/uploads/2017/11/Prasad_2017_Flex._Print._Electron._2_045007.pdf}, doi = {10.1088/2058-8585/aa91ba}, year = {2017}, date = {2017-11-03}, journal = {Flexible and Printed Electronics}, volume = {2}, number = {4}, abstract = {The design of interconnects on elastomers is of interest in flexible electronics. The geometric design of interconnects must take into account the electrical and mechanical aspects, along with layout area. This work discusses the impact of the geometry of stretchable interconnects on various user specifications, namely layout area, stretchability given by mechanical stiffness coefficient and speed given by electrical impedance. Analytical models have been derived and an optimization methodology is developed to help pick the best design for the given application. The analytical models are corroborated with simulations and experiments. Interconnect designs have been fabricated on a copper-viscoelastic adhesive-elastomer stack architecture, using laser toner-transfer process where the viscoelastic adhesive prevents delamination and in which the process flow is in line with conventional printed circuit board fabrication techniques.}, keywords = {Systems Engineering}, pubstate = {published}, tppubtype = {article} } The design of interconnects on elastomers is of interest in flexible electronics. The geometric design of interconnects must take into account the electrical and mechanical aspects, along with layout area. This work discusses the impact of the geometry of stretchable interconnects on various user specifications, namely layout area, stretchability given by mechanical stiffness coefficient and speed given by electrical impedance. Analytical models have been derived and an optimization methodology is developed to help pick the best design for the given application. The analytical models are corroborated with simulations and experiments. Interconnect designs have been fabricated on a copper-viscoelastic adhesive-elastomer stack architecture, using laser toner-transfer process where the viscoelastic adhesive prevents delamination and in which the process flow is in line with conventional printed circuit board fabrication techniques. |
| 48. | Vaidhiyan, Nidhin K; Arun, S P; Sundaresan, Rajesh Neural dissimilarity indices that predict oddball detection in behaviour Journal Article IEEE Transactions on Information Theory, 63 (8), pp. 4778-4796, 2017. Abstract | BibTeX | Tags: Modeling and Simulation and Analytics | Links: @article{Vaidhiyan2017, title = {Neural dissimilarity indices that predict oddball detection in behaviour}, author = {Nidhin K. Vaidhiyan and S. P. Arun and Rajesh Sundaresan}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/07937887.pdf}, doi = {10.1109/TIT.2017.2707485}, year = {2017}, date = {2017-06-01}, journal = {IEEE Transactions on Information Theory}, volume = {63}, number = {8}, pages = {4778-4796}, abstract = {Neuroscientists have recently shown that images that are difficult to find in visual search elicit similar patterns of firing across a population of recorded neurons. The L1 distance between firing rate vectors associated with two images was strongly correlated with the inverse of decision time in behavior. But why should decision times be correlated with L1 distance? What is the decision-theoretic basis? In our decision theoretic formulation, we model visual search as an active sequential hypothesis testing problem with switching costs. Our analysis suggests an appropriate neuronal dissimilarity index, which correlates equally strongly with the inverse of decision time as the L1 distance. We also consider a number of other possibilities, such as the relative entropy (Kullback-Leibler divergence) and the Chernoff entropy of the firing rate distributions. A more stringent test of equality of means, which would have provided a strong backing for our modeling, fails for our proposed as well as the other already discussed dissimilarity indices. However, test statistics from the equality of means test, when used to rank the indices in terms of their ability to explain the observed results, places our proposed dissimilarity index at the top followed by relative entropy, Chernoff entropy, and the L1 indices. Computations of the different indices require an estimate of the relative entropy between two Poisson point processes. An estimator is developed and is shown to have near unbiased performance for almost all operating regions.}, keywords = {Modeling and Simulation and Analytics}, pubstate = {published}, tppubtype = {article} } Neuroscientists have recently shown that images that are difficult to find in visual search elicit similar patterns of firing across a population of recorded neurons. The L1 distance between firing rate vectors associated with two images was strongly correlated with the inverse of decision time in behavior. But why should decision times be correlated with L1 distance? What is the decision-theoretic basis? In our decision theoretic formulation, we model visual search as an active sequential hypothesis testing problem with switching costs. Our analysis suggests an appropriate neuronal dissimilarity index, which correlates equally strongly with the inverse of decision time as the L1 distance. We also consider a number of other possibilities, such as the relative entropy (Kullback-Leibler divergence) and the Chernoff entropy of the firing rate distributions. A more stringent test of equality of means, which would have provided a strong backing for our modeling, fails for our proposed as well as the other already discussed dissimilarity indices. However, test statistics from the equality of means test, when used to rank the indices in terms of their ability to explain the observed results, places our proposed dissimilarity index at the top followed by relative entropy, Chernoff entropy, and the L1 indices. Computations of the different indices require an estimate of the relative entropy between two Poisson point processes. An estimator is developed and is shown to have near unbiased performance for almost all operating regions. |
| 49. | Heemels, Maurice; Kundu, Atreyee; Daafouz, Jamal On Lyapunov-Metzler inequalities and S-Procedure characterizations for the stabilization of switched linear systems Journal Article IEEE Transactions on Automatic Control, 62 (9), pp. 4593-4597, 2017. Abstract | BibTeX | Tags: Control and Optimisation and Policy | Links: @article{Heemels2017, title = {On Lyapunov-Metzler inequalities and S-Procedure characterizations for the stabilization of switched linear systems}, author = {Maurice Heemels and Atreyee Kundu and Jamal Daafouz}, url = {http://www.rbccps.org/wp-content/uploads/2018/01/07733093.pdf}, doi = {10.1109/TAC.2016.2624987}, year = {2017}, date = {2017-08-28}, journal = {IEEE Transactions on Automatic Control}, volume = {62}, number = {9}, pages = {4593-4597}, abstract = {In this note we present connections between two celebrated tools for the design of stabilising switching laws for continuous-time and discrete-time switched linear systems, namely Lyapunov-Metzler inequalities and S-procedure.}, keywords = {Control and Optimisation and Policy}, pubstate = {published}, tppubtype = {article} } In this note we present connections between two celebrated tools for the design of stabilising switching laws for continuous-time and discrete-time switched linear systems, namely Lyapunov-Metzler inequalities and S-procedure. |
| 50. | Prasanth, V; Parekhji, Rubin; Amrutur, Bharadwaj Safety analysis for integrated circuits in the context of hybrid systems Conference Forthcoming International Test Conference (ITC 2017), 31.10.-02.11.17, Fort Worth (USA), Forthcoming. BibTeX | Tags: Systems Engineering @conference{Prasanth2017, title = {Safety analysis for integrated circuits in the context of hybrid systems}, author = {V. Prasanth and Rubin Parekhji and Bharadwaj Amrutur}, year = {2017}, date = {2017-11-02}, booktitle = {International Test Conference (ITC 2017), 31.10.-02.11.17, Fort Worth (USA)}, keywords = {Systems Engineering}, pubstate = {forthcoming}, tppubtype = {conference} } |
