Getting settled in at my new academic home, with all the attendant administrivia, has sucked up almost two months of missed blogging opportunities. So, here goes.
This year’s ITA was held at a new location, Catamaran Resort on Mission Bay. Anand has already blogged about some talks; this is my selection, which is disjoint from his. I will follow Cosma’s example and just give very telegraphic summaries of some of my favorites with links to abstracts and/or papers (when available).
- Chandra Nair, “Upper concave envelopes and broadcast channels” (abstract)
— whenever we use auxiliary random variables in multiterminal information theory, convex analysis is lurking somewhere in the background; Chandra showed how to bring it to the foreground; is there a connection to comparison of experiments in the sense of Blackwell?
- Four talks on various information-theoretic problems in the finite-blocklength regime (via either the dispersion approach of Polyanskiy, Poor and Verdú based on the Berry-Esseen theorem, or various sharp asymptotic results like the Bahadur-Rao theorem):
- Tsachy Weissman, “New estimators of directed information,” joint work with Jiantao Jiao, Haim Permuter, Lei Zhao, Young-Han Kim (abstract, arXiv:1201.2334)
— how to use universal probability assignment (with or without smoothing) to estimate directed information between two random processes
- Michelle Effros, “Towards a computational information theory,” (abstract, related paper)
— exact computation of capacities (and achievable rate regions) in large networks is largely intractable; let’s forget exact results and develop approximation and bounding techniques: when can we replace one or more channels in a network with some others and still stay in the achievable rate region?
- Stefano Soatto, “Information forests,” joint work with Zhao Yi, Maneesh Dewan, and Yiqiang Zhan (paper)
— Stefano described an information-theoretic approach to designing decision trees (or forests, to be more precise) for classification
- Urbashi Mitra, “State-dependent active communication,” joint work with Chiranjib Choudhuri (abstract, related paper)
— how can we simultaneously transmit information over a channel with state and estimate the state sequence? the state sequence is modulated at the encoder by means of actions, which may or may not depend on previous channel states; the proof of converse used a neat data processing inequality for statistical estimation in the spirit of Blackwell’s comparison of experiments
- Two talks by Dayu Huang and Sean Meyn (Eminent Scholar) on statistical estimation in the sparse-sample regime:
“Error exponent for goodness of fit test with sparse samples” (abstract)
“Optimality of coincidence-based goodness of fit test for sparse sample problems” (paper)
- Mohammad Naghshvar, “Active sequential hypothesis testing,” joint work with Tara Javidi (abstract)
— Mohammad presented some new results on the topic; Jensen-Shannon divergence (i.e., the gap in Jensen’s inequality for entropy) made a surprising appearance
- Sirin Nitinawarat, “Controlled sensing for hypothesis testing,” joint work with George Atia and Venu Veeravalli (abstract)
— this talk was thematically related to Mohammad’s, but Sirin focused on fixed sample size problems