Suggested Papers
Learning with Square Loss: Localization through Offset Rademacher Complexity ---- Tengyuan Liang, Alexander Rakhlin and Karthik Sridharan
Unbounded losses and luckiness bounds
A Chaining Algorithm for Online Nonparametric Regression ---- Pierre Gaillard and Sébastien Gerchinovitz
I think this is cool because it transitions from using GD to EG updates.
Exp-Concavity of Proper Composite Losses ---- Parameswaran Kamalaruban, Robert Williamson and Xinhua Zhang
Kamal said there are strictly more mixable than exp-concave losses. That sounds intriguing. There might be hints of that in this paper.
Competitive Distribution Estimation: Why is Good-Turing Good ---- Alon Orlitsky and Ananda Theertha Suresh
I would love to understand the connection between Good-Turing and learning the alphabet size (i.e. paper by Hutter). Both are about estimating the probability of observing a yet unseen symbol.
On Elicitation Complexity ---- Rafael Frongillo and Ian Kash
This paper highlights connections between learning and finance.
Sparse Adaptive Dirichlet-Multinomial-like Processes ---- Marcus Hutter
This paper is about predicting when the alphabet size is not known. Key ingredient is estimating the "escape probability" of an as-of-yet unseen symbol appearing. This is especially interesting to compare with the Good-Turing work of Orlitsky et al.
No more pesky learning rates ---- Tom Schaul, Sixin Zhang, Yann LeCun
A chunk of research at CWI is devoted to learning the learning rate. So I am curious how this paper relates.
Regret Minimization in Games with Incomplete Information ---- M. Zinkevich, M. Bowling, M. Johanson, C. Piccione.
This paper discusses using online learning methods for solving incomplete information games (like poker where you do not see the other players' cards). The method used (Regret Matching) is not quite the standard algorithm (Hedge) and I'd like to understand its advantage.
Coin Betting and Parameter-Free Online Learning ---- Francesco Orabona, Dávid Pál
Connections between coin betting and online learning. There are links to Universal Coding and to Game Theoretic Probability (Shafer-Vovk)