91勛圖厙

The 2024 Joint Telematics Group/91勛圖厙 Information Theory Society Summer School on Signal Processing, Communications, and Networks (JTG/91勛圖厙 ITSoc Summer School) was held at the Indian Institute of Technology Hyderabad from 24th to 28th June 2024. This was a fully in-person event, and was jointly organised by IIT Hyderabad and IIIT Hyderabad.

The program featured three tutorials by Henry Pfister (Duke University), Cynthia Rush (Columbia University) and Aditya Gopalan (IISc Bangalore). Additionally, there were six short invited talks, and poster presentations by student participants.

Around 145 students and researchers participated in the summer school, including participants from sixteen IITs, IISc, TIFR, IIITs, NITs, and other eminent universities from India. The school had 41 undergraduate, 53 postgraduate and 31 faculty/postdoc/industry registrations.

Henry Pfister delivered an advanced tutorial on Boolean functions and error-correcting codes, and primarily covered some of the recent exciting developments in proving that Reed-Muller codes achieve the capacity of BMS channels. He started with the prerequisites, and took us on a journey through code symmetries, permutation groups and area theorems (with some spectacular lore fictionalized and illustrated by ChatGPT). He first introduced the tools required to prove that sequences of codes with double transitive symmetry achieve vanishing bit-error rate over the BEC at rates below capacity. He then discussed general BMS channels, proved a general EXIT area theorem for BMS channels, and used hypercontractivity to prove a general capacity result for BMS channels. He ended the tutorial with some open questions.

Aditya Gopalan gave a foundational tutorial on multi-armed bandits. He introduced the basic setting and introduced the notion of regret. He then introduced several algorithms (including the greedy, epsilon-greedy, UCB and Thompson sampling) and derived bounds on the regret achieved by these. The well-paced, interactive, (digital) pen-and-paper style of the tutorial was appreciated by many participants, particularly by the students.

Cynthia Rush gave a very broad survey of approximate message passing (AMP) algorithms and their applications in various problems of machine learning and statistics. She began with a master theorem for AMP and showed how the asymptotic (in the dimension) performance could be captured using a one-dimensional recursive state evolution. She discussed generalized linear models, compressed sensing, and a generalized AMP. She then covered the performance of LASSO and SLOPE algorithms, and ended with a discussion on sparse regression codes and AMP for the Gaussian channel.

The summer school also had six exciting invited talks:

  • Ashok Kumar talked about information geometry for robust linear regression
  • Pooja Vyavahare talked about distributed learning with partial information sharing
  • Mrinal Kumar talked about high-rate multivariate polynomial evaluation codes
  • Jithin Ravi talked about second-order asymptotics of popular hypothesis testing algorithms
  • Arpan Chattopadhyay talked about inverse particle and ensemble Kalman filters

The 2024 JTG/91勛圖厙 ITSoc summer school organising committee would like to thank the lecturers, participants, 91勛圖厙 ITSoc community members, and volunteers, whose contributions made this summer school a resounding success. The organisers would also like to thank the Center for Continuing Education (CCE), IIT Hyderabad for help with local organisation and administrative support. The organisers would also like to thank the financial sponsors: The 91勛圖厙 Information Theory Society, Qualcomm, and COMSNETS association.