6. Counter-adversarial inference, control and learning: New Frontiers, Newer Challenges
Organizers: Arpan Chattopadhyay, Kumar Vijay Mishra, John S. Baras, P. R. Kumar, Vivek S. Borkar
Location: Roselle Junior 4611
Website: https://sites.google.com/site/arpanchattop/cdc-2023-workshop
Abstract: Modern day inference and control theory has its origins in the 19th and early 20th centuries, with a growth of applications in the post-industrial-revolution’s automation and defence requirements. The 20th century witnessed the development of reinforcement learning techniques for control in uncertain environments, game theory to model multiple selfish or cooperating agents, stochastic filtering techniques for state tracking, networked control systems to extend the support of wireless networks to control systems, and machine learning techniques to infer from data. During the past two decades, rapid developments in multi-agent control, team decision theory, and big data analytics expanded the scope of these tools in the non-traditional areas. These theoretical developments happened in parallel with a massive improvement in the intelligence, sensing, communication, computation and storage capacities of the devices involved in inference and control systems. The increased affordability of autonomous systems has also led to the situation where autonomous systems owned by different entities need to interact with each other, often in an adversarial manner. Examples range from the competitive cognition and inverse cognition operations between a smart radar and an intelligent target, to the competition between a defender and a stealthy cyber-attacker, to a more futuristic scenario where two autonomous teams of UAVs fight each other in a battlefield. The inference and control operations in these systems need to be performed in uncertain environments. This necessitates the development of new theories of counter-adversarial inference, control and learning.
Counter-adversarial inference, control and learning theory has received significant traction over the past few years. However, this area is far from being mature. Technological advancement is increasingly catering to the complexity and scalability of these problems and solution techniques, and the changing nature of the interaction among agents in new technological domains are giving rise to more challenging problems. This is the motivation behind the proposed workshop-eminent researchers from diverse backgrounds will meet, present their work and discuss future research directions. While the workshop will not exhaustively cover all emerging research directions in this domain, it exhibits its uniqueness in choosing the technical topics spanning a large new spectrum including robust control and learning, decision making against adversaries, multi-agent control over networks, consensus, security, trust, control for MIMO communication (contrary to popularly studied problems on control over communication network) and mean field games between teams.
Despite being such a vibrant field, there has been no special issue from the IEEE control society on counter-adversarial inference, control and learning in recent times, though papers on some of these topics are published in control and learning theory venues in isolated manner. A special session is not sufficient to cover so many topics, and hence a workshop will be an ideal venue for dissemination of
knowledge in this field.
Lecture Schedule:
Lecture Schedule:
- 09:00 ~ 09:30 Scalable control for distributed MIMO communications (Speaker: Soura Dasgupta, Univ. of Iowa)
- 09:30 ~ 10:00 Stealthy Sensor Attacks: Characterization and Moving Target Defense (Speaker: Henrik Sandberg, KTH)
- 10:00 ~ 10:30 Unveiling the Impact: Exploring Grounding Effects on Scalable Consensus Networks (Speaker: Yamin Yan, HKUST)
- 10:30 ~ 11:00 Break
- 11:00 ~ 11:30 Multi-scale Network Games: Modeling, Analysis, and Control (Speaker: Mingyan Liu, University of Michigan, Ann Arbor)
- 11:30 ~ 12:00 Mean field game among teams (Speaker: Aditya Mahajan, McGill University)
- 12:00-13:30 Lunch
- 13:30-14:00 Robust Machine Learning, Reinforcement Learning and Autonomy: A Unifying Theory via Performance and Risk Tradeoff (Speaker: John. S. Baras, Univ. of Maryland)
- 14:00-14:30 Performance Guarantees for Learning based Decision Making (Speaker: Edwin K.P. Chong, Colorado State University)
- 14:30-15:00 Quickest detection of deception attacks in networked control systems with watermarking (Speaker: Subhrakanti Dey, Uppsala University)
- 15:00-15:30 Break
- 15:30-16:00 Deceptive Decision-Making: Inference, Strategies, and Environment Co-Design (Speaker: Melkior Ornik, University of Illinois at Urbana Champaign)
- 16:00-16:30 Reputation-based information design for increasing prosocial behaviour (Speaker: Rajesh Sundaresan, Indian
Institute of Science Bangalore) - 16:00-17:00 Adversarial Robustness Considerations in Black-Box (Speaker: Jonathan Scarlett, National University of Singapore)
- 17:00-17:30 Discussion session