ITA 2011: Favorite Talks
ITA was great, partly for the reasons visible at
right, and partly for getting to enjoy to gracious hospitality
of Doug White, but mostly
for the scientific exchange. So, some links to my favorite talks. (Note
"favorite" and not "best".) I will not attempt to explain any of these
adequately, or to list everyone's co-authors. It's good that so many of the
papers are on arxiv, but unfortunate that not all of them are.
- Todd Coleman, "A relationship between information theory and stochastic
control when beliefs are decision variables" (Abstract, arxiv:1102.0250)
- Ali Jadbabaie, "Consensus, Social Learning, and Distributed Estimation" (Abstract [in the evil Word format], SSRN/1550809)
- The talk focused on new results which are not included in the SSRN paper,
but in the same set-up --- if I can trust my notes, replacing sufficient
conditions from the old paper with weaker necessary-and-sufficient
- Aryeh Kontorovich, "Efficient classification for metric data" (Abstract, PDF of COLT 2010 paper)
- It would be nice to know how this relates to Laplacian
regularization methods, as in Belkin and Niyogi (journal version).
- Daniel Lee, "Learning metrics for nearest neighbor classification"
(Abstract, PDF of NIPS 23 (2010) paper)
- The results for nearest-neighbor classification are nice, but it's
the extension to estimating f-divergences, including Kullback-Leibler
divergences and total variation distance, that I find most interesting.
- Maxim Raginsky, "Shannon meets Blackwell and Le Cam: Coding Theorems of
Information Theory and Comparison of Experiments"
- Sasha Rakhlin, "From statistical learning to online learning and games"
(Abstract, arxiv:1011.3168, arxiv:1006.1138)
- Irina Rish, "A greedy
coordinate ascent method for learning sparse Gaussian MRFs" (PDF of paper)
- Is it possible to remove the Gaussian assumption by combining this with
- Aarti Singh, "Consistent recovery of high-dimensional graph-structured patterns" (Abstract, PDF of NIPS 23 (2010) paper)
- Ramon van Handel, "On a question of Blackwell concerning hidden
Markov chains" (Abstract, arxiv:0910.3603)
- Ramon had just nine slides, including title and references, and
they weren't over-packed either. I envy this skill.
- Frank Wood, "The
of ICML 2009 paper, website)
- Roughly speaking, the sequence memoizer is a
Markov chains, with prior distributions tuned to handle the
large-number-of-rare-events issue ubiquitous in text. The compression
performance of the sequence memoizer is really remarkable, as you
can see for yourself, and it would be nice to
understand it better. Likewise, Frank's work, with David Pfau and Nicholas
mixtures of probabilistic deterministic finite automata, is like a
of CSSR. It would be good to
understand the conditions under which either the memoizer or the PDFA mixture
are consistent estimators, but
I would think that.
- Serdar Yüksel, "Comparison and characterization of observation channels
for stochastic stabilization of noisy linear systems" (Abstract, arxiv.org:1009.3824)
- Brian Ziebart, "Behavior
forecasting with the principle of maximum causal entropy"
see webpage for related papers)
- Actually, I didn't go to Brian's talk, but he was kind enough to explain it
to me anyway. It doesn't make me reconsider my general skepticism about
entropy maximization as a principal of
inference, but does make me even more interested in what sorts
of large deviations principles
might apply to cabbies, and to games more generally.
In addition to the talks, and many enlightening conversations,
Anand introduced Maxim and me to
the Noble Experiment,
surely the best cocktail lounge in which the wall opposite the bar is
entirely covered in gilded skulls. At least one of the three of us should
probably have done some memento mori blogging.
"zombie-blogging" the workshop, which makes me fear for the future of ITA.
(He says he was only sick with the flu, but by this point we all know
rest of that story goes.)
Enigmas of Chance;
Posted at February 13, 2011 17:30 | permanent link