Synapse

 

Monthly Newsletter from CNI

Issue - July 2024

Join us at the upcoming talk on "Leveraging WiFi for Robust and Resource-Efficient SLAM" by Aditya Arun, PhD student at UCSD on 23/07/2024.

Join us at the upcoming talk by  Prof. Sharayu Mohrair, from IIT Bombay on 30/07/2024.

Fourth Annual CNI Summer School features Course on Algorithmic Structures Emerging Wireless Networks and Statistical Inference by Prof. Jean-Francois Chamberland

Prof. Jean-Francois Chamberland from Texas A&M University presented a course titled "Algorithmic Structures for Emerging Wireless Networks and Statistical Inference in Large Dimensional Spaces" at the fourth annual CNI summer school. The presentation focused on advanced algorithmic techniques and statistical methods essential for the development of wireless networks. The course started with an in-depth review of concepts from linear algebra, probability, and optimization, which set up the tools and notation needed to discuss the problems. Several problems, such as the Uncoordinated Multiple Access Channel, Compressed Sensing, and Sparsifying Collisions, were discussed in detail. He further illustrated the use of algorithmic tools and ideas, such as approximate message passing, graph-based constructions, and data fragmentation, to solve these problems.

Cisco Executives delve into research at CNI and engage with scholars

Harish Krishnan, Managing Director at Cisco Systems India and SAARC Co-Founder, along with Ravishankar Rao, Director of Strategy and Operations at Cisco, and the Cisco CSR team visited CNI to get acquainted with recent developments. During their visit, they also interacted with CNI scholars.

Average Reward Markov Decision Process

In this talk, Prof. Vaneet Aggarwal explored an infinite horizon average reward Markov Decision Process (MDP). Distinguishing his work from previous research, he focused on regret guarantees with a general parameterization, particularly emphasizing policy gradient-based algorithms. He then elucidated the key insights underlying his gradient estimation approach, achieving a regret bound of O(T^0.75). Additionally, he proposed an efficient momentum-based approach that attained a regret bound of O(T^0.5). Finally, he introduced a technique to reduce the dependence on mixing time for this problem.

Designing Robust Algorithms for Data Streams that might be adversarially generated.

In this talk, Prof. Amit Chakrabarti discussed the challenges of modern data stream processing systems, which must efficiently handle continuous data streams while ensuring accurate decision-making that impacts future data flows. The focus was on adversarially robust streaming algorithms, designed to produce correct outputs despite each data element potentially being adversarially chosen based on previous outputs. He also outlined foundational concepts in this field and presented recent findings in two key areas: Missing Item Finding (MIF) and graph coloring. These results highlight how the availability of random bits influences the space efficiency of algorithms in both MIF and graph coloring tasks.

Learning in the presence of strategy

In this talk, Prof. Ganesh Ghalme explored how machine decision-making interacts with human welfare, highlighting the challenges posed by transparent decision rules that can be strategically manipulated by rational agents. This necessitates the development of robust decision strategies. Two frameworks were presented: strategic classification, where agents manipulate their features to influence the system's classifier, and strategic representation, where systems manipulate information to induce favourable decisions from agents or decision-makers. He further discussed various robust algorithmic approaches within these frameworks and examined their limitations.