Monthly Newsletter from CNI

Issue - December 2023


Join us at the upcoming talk on "Sequential Clustering of Data Streams from Unknown Distributions" by Prof. Srikrishna Bhashyam from IIT Madras on 18/12/2023 at 17:00 hrs IST.

Join us at the upcoming talk on "Towards a Theory of Exploration in Continuous-Time Reinforcement Learning" by Prof. Harsha Honnappa from Purdue University on 19/12/2023 at 16:00 hrs IST.

Join us at the upcoming talk on "CT Scan for Your Network: Topology Inference from End-to-end Measurements" by Prof. Ting He from Pennsylvania State University on 20/12/2023 at 17:00 hrs IST.

What’s cooking at the CNI ?

We hosted the CNI Workshop on Information Theory on 2nd December 2022, featuring a lineup of distinguished speakers. The day began with insightful sessions by P Vijay Kumar, Amitalok Budkuley, and Srikrishna Bhashyam. This was followed by a keynote lecture on “Shared Information” by Prof Prakash Narayan, who is visiting the ECE department as a Satish Dhawan Chair Professor. The afternoon sessions continued with talks by Vivek Borkar, Bikash Kumar Dey, Vinod Sharma, Vinod Prabhakaran, Manuj Mukherjee, Shashank Vatedka, and Navin Kashyap

A special heartwarming session dedicated to Prof Prakash Narayan's 70th birthday was a heartfelt celebration of his personal warmth and the positive impact he has had on those around him. The workshop concluded with engaging discussions, and our sincere thanks go to all participants and speakers for their valuable contributions.


Past Events

Online Learning for Network Resource Allocation  


In this talk, Dr. Tareq Si Salem explored the challenges of diverse network applications requiring low latency and high bandwidth in the era of connected devices. He delved into Network Resource Allocation (NRA), emphasizing its role in optimizing resource placement to minimize computational overhead. Practical challenges, such as varying network parameters and unpredictable demand, were discussed. The seminar specifically addressed NRA problems related to caching strategies, inference delivery networks, and fairness issues. Techniques from Online learning and adversarial analysis are leveraged to identify online policies, ensuring robust performance even in unpredictable environments lacking statistical regularity.

Collaborative Decision-Making under Adversarial and Information Constraints


In this talk, Dr. Aritra Mitra discusses a linear bandit setup involving collaborating agents striving to minimize regret, despite a fraction of them being adversarial. The challenge lies in balancing the potential benefit of collaboration with the risk of disruption by adversaries. Robust algorithms are introduced to address this tension, leveraging carefully crafted confidence intervals.  When only a small fraction of agents is corrupted, collaboration remains beneficial despite adversarial elements, yielding tight regret bounds. The second part introduces a linear stochastic bandit system over a limited-capacity channel. Agents transmit encoded estimates to minimize cumulative regret, showing O(d) bits capacity suffices for optimal regret in d-dimensional models, and 1-bit capacity works for the simpler multi-armed bandit problem.

AI-Assisted Teaching: Where We Stand and What Lies Ahead


Check out this talk by Dr. Deepak Sekar, CEO, and Co-founder of Prof Jim, as he discusses the role of AI in supporting traditional classroom teaching. This integration not only saves valuable teacher time but is also useful in scenarios with challenging student-teacher ratios, fostering a more personalized and effective learning environment.

Community detection on Block Models with Geometric Kernels


Dr. Vinay Kumar discussed the community recovery problem on random geometric graphs, where each node has both location and community labels. Exploring statistical limits, he introduced a linear time algorithm for inferring latent communities, showing the exact recovery of the community labels up to the information theoretic threshold in one-dimensional cases.  

Scalability in Low Power Wide Area Networks


Prof. Bhuvana Krishnaswamy of the University of Wisconsin-Madison discussed challenges in wireless data delivery over long distances in a recent talk. Emphasizing the growing demand for Low-Power Wide Area Networks (LPWAN), the talk highlighted the need for solutions addressing increased battery life, longer communication range, scalability, and cost-effectiveness. The focus was on collision resolution strategies, particularly within commercially available LPWANs like LoRa, to efficiently tackle these challenges at a large scale.

Facebook icon Instagram icon Twitter icon YouTube icon

Our mailing address is: