Michael I. Jordan

TL;DR: In this paper, a selective importance sampling estimator (sIS) is proposed for batch learning from extreme bandit feedback in the setting of extremely large action spaces, where the selected actions for the sIS estimator are the top-p actions of the logging policy, where p is adjusted from the data and is significantly smaller than the size of the action space.

TL;DR: In this paper, the authors provide theoretical complexity analysis for new algorithms to compute the optimal transport distance between two discrete probability distributions, and demonstrate their favorable practical performance over state-of-the-art primal-dual algorithms and their capability in solving other problems in large-scale, such as the Wasserstein barycenter problem for multiple probability distributions.

TL;DR: In this article, the authors present an algorithm that modifies any classifier to output a predictive set containing the true label with a user-specified probability, such as 90%, which provides a formal finite sample coverage guarantee for every model and dataset.

TL;DR: A general framework that unifies model-based and model-free RL, and an Admissible Bellman Characterization (ABC) class that subsumes nearly all Markov Decision Process (MDP) models in the literature for tractable RL is proposed, which provides a generic interface to design and analyze new RL models and algorithms.

TL;DR: A posterior inference algorithm is presented for the beta-Bernoulli process that exploits the stickbreaking representation, and experimental results for a discrete factor-analysis model are presented.