For my Bengali-speaking friends:.

By the way, here is the paper:

Quite exciting to listen to the AI-generated podcast of one of our recent papers (to appear in COLT 25).Thanks to Kushagra Chatterjee, one of my PhD students, for generating this.

RT @kasperglarsen: The SODA'26 Call for Papers and submission server is now online! Deadline July 14 AOE. Please share and submit your awes….

RT @BarmanSiddharth:

RT @ccanonne_: The 35th International Symposium on Algorithms and Computation (ISAAC'24), will be held in Sydney 🇦🇺, December 2024. The CfP….

RT @kasperglarsen: Excited to announce two post doc positions on learning theory, under my supervision at Aarhus University. The positions….

Delighted to receive the Faculty Teaching Excellence Award 2024. Thanks to all the students and colleagues for their immense support

RT @diveshaggarwal: We have an amazing line of speakers for the CS research week from January 9 to 12 at the Department of Computer Science….

RT @nutanlimaye: You can take many road to reach Copenhagen. :) Check this announcement. Many brilliant minds unde….

RT @csfacultyjobs: Faculty Positions, National University of Singapore, Singapore

Our technique is very generic and can be applied to many other distribution testing problems as well.

Next, we argue that the output distributions of testers with general conditional sampling oracle and this new weaker query model are "close" in total variation distance, which enables us to lift the lower bound to the general conditional model.

We overcome this barrier by considering a weaker query access model, which enables us to "sparsify" the decision tree and break the \sqrt{log log n} barrier.

The current best upper bound is O(log log n) (the domain is [n]), and there was a quadratic gap between the best-known upper and lower bound. Previous lower bound techniques seem to possess a \sqrt{log log n} barrier.

Given two distributions (with sample access), how efficiently (in terms of sample complexity) can we test whether they are equal or "far" in total variation distance? A tester can get a sample after conditioning on any arbitrary subset of its choice.

We essentially close the gap between the upper and lower bound, resolving Open Problem 87 of

Excited to share our new lower bound result on the sample complexity of the equivalence testing problem, a fundamental distribution testing problem, in the conditional sampling model. (joint work with Gunjan Kumar and Sourav Chakraborty).

RT @divy93t: [Announcement] Thrilled to support faculty in India, South and Southeast Asia through dedicated programs for the region. All….