Thu07Dec201716.00-17.00Robert Bosch Centre for Cyber-Physical Systems, Seminar HallShow details
Several finite blocklength converses in information theory have been discovered for several loss criteria using a variety of arguments. What is perhaps unsatisfactory is the absence of a common framework using which converses can be found for any loss criterion. We present a linear programming based framework for obtaining converses for finite blocklength lossy joint source-channel coding problems. The framework applies for any loss criterion, generalizes certain previously known converses, and also extends to multi-terminal settings. The finite blocklength problem is posed equivalently as a nonconvex optimization problem and using a lift-and-project-like method, a close but tractable LP relaxation of this problem is derived. Lower bounds on the original problem are obtained by the construction of feasible points for the dual of this LP relaxation. A particular application of this approach leads to new converses that improve on the converses of Kostina and Verdu ́ for joint source-channel coding and lossy source-coding, and imply the converse of Polyanksiy, Poor and Verdu for channel coding. Another construction leads to a new general converse for finite blocklength joint source-channel coding that shows that the LP is tight for all blocklengths for the "matched setting" of minimization of the expected average bit-wise Hamming distortion of a q-ary uniform source over a q-ary symmetric memoryless channel.
The tightness of the LP relaxation for canonical problems in information theory shows that optimal coding in these problems has an associated "dual" viewpoint: namely, the optimal packing of "source flows" and "channel flows" that are throttled by an error density bottleneck. In the multi-terminal setting, using the language of these flows we derive improvements to converses of Miyake and Kanaya for Slepian-Wolf coding, the converse of Zhou et al for the successive refinement problem and new tight converses for compound and averaged channels. Coincidentally, the recent past has seen a spurt of results on using duality to obtain outer bounds in combinatorial coding theory (including the author's own nonasymptotic upper bounds for zero-error codes for the deletion channel). We speculate that these and our results hold the promise of a unified, duality-based theory of converses for problems in information theory.
About the Speaker
Ankur Kulkarni is an Assistant Professor (since 2013) with the Systems and Control Engineering Group at the Indian Institute of Technology Bombay (IITB).
He received his B.Tech. in Aerospace Engineering from IITB in 2006, M.S. in 2008 and Ph.D. in 2010, both from the University of Illinois at Urbana-Champaign (UIUC). From 2010-2012 he was a post-doctoral researcher at the Coordinated Science Laboratory at UIUC.
His research interests include the role of information in stochastic control, game theory, information theory, combinatorial coding theory problems, optimization and variational inequalities, and operations research. He is an Associate (from 2015-2018) of the Indian Academy of Sciences, Bangalore, a recipient of the INSPIRE Faculty Award of the Department of Science and Technology, Government of India (2013), best paper award at the National Conference on Communications in Chennai (2017), and the William A. Chittenden Award at UIUC (2008). He is a consultant to the Securities and Exchange Board of India on the regulation of high frequency trading.
Mon04Dec201716.00-17.00Robert Bosch Centre for Cyber-Physical Systems, Seminar HallShow details
Sustainability and security considerations have lead to an increased deployment of renewable generation in grids all over the world. However, limited control capabilities and uncertainty associated with renewables poses a challenge to the conventional “load-following” operational strategy adopted by the grid operators. Engaging the demand-side in power grid operations is a potential solution to address this challenge. This talk provides an overview of ways in which consumers can participate in grid management and the benefits associated with their participation. The talk also describes technology solutions being developed at IITB to equip consumers with information and decisions support tools needed to facilitate their participation in grid operations.
About the Speaker
Anupama Kowli is an Assistant Professor at the Indian Institute of Technology Bombay. She is a researcher in the area of electricity markets, energy economics, resource planning and power system operation and control.
She received her Masters and Ph.D. from University of Illinois at Urbana Champaign in 2009 and 2013, respectively. Anupama was a visiting scholar at University of Florida. She interned as an energy consultant at KEMA Inc and as a control engineer at the Pacific Northwest National Laboratory.
Thu30Nov201710.00-11.00Robert Bosch Centre for Cyber-Physical Systems, Seminar HallShow details
Cyber-Physical Systems (CPS) are engineered systems resulting from a seamless integration between physical processes and cyber technologies such as communication networks and computational hardware. This tight integration exposes the CPS to a variety of attacks, both on the physical and cyber components, which can result in significant performance degradation. Further, CPS usually consist of multiple agents that collaborate and share information with each other, thus making them vulnerable to privacy breach and leakage of confidential data. This talk will focus on the need, design and analysis of security and privacy mechanisms in CPS.
In the first part of the talk, we will present a security problem for real-time resource- constrained autonomous systems (for example, a UAV), which can reserve only limited computational resources and time for security and control purposes. In such scenarios, the control and security tasks usually compete with each other for limited resources and there exists a trade-off between security and control performance. We characterize the optimal trade-off and identify attack regimes in which the system should prefer control tasks over security tasks, and vice-versa.
The second part will focus on privacy in cooperative dynamical multi-agent CPS. We present a noise adding differentially private mechanism to preserve the privacy of agents’ state over time, and analyze the effect of the privacy mechanism on the system performance. Next, we show that a fundamental trade-off exists between privacy and cooperation level, and it is beneficial for the agents to reduce cooperation if they want to be more private.
About the Speaker
Vaibhav Katewa is a postdoctoral scholar in the Department of Mechanical Engineering at University of California, Riverside. He received his M.S. and Ph.D. in Electrical Engineering from University of Notre Dame in 2012 and 2016, respectively, advised by Prof. Vijay Gupta. He received his B.Tech. from IIT Kanpur in 2007 in Electrical Engineering.
His research interests include design and analysis of security and privacy methods for cyber-physical systems and complex networks, decentralized and sparse feedback control, and protocol design for networked control systems.