Machine Learning with Comparisons

Together with graduate student Aadirupa Saha at IISc, Aditya Gopalan’s work on online learning with relative preferences has resulted in a new framework for optimizing over sets of alternatives with only relative, subset-wise observations. This is a general framework that is applicable to many automated recommendation systems that can sequentially elicit only  relative preferences from, say, human users, e.g., “Do you like X over Y (and Z)?” This study has yielded state-of-the-art learning algorithms that make optimal subset selection decisions in terms of regret and rank-order estimation error, along with new insights on how to efficiently make comparisons to elicit items’ utilities within a wide range of social choice models.

References:

Aadirupa Saha and Aditya Gopalan, From PAC to Instance-Optimal Sample Complexity in the Plackett-Luce Model. 37th International Conference on Machine Learning (ICML), 2020.

Aadirupa Saha and Aditya Gopalan, Best-item Learning in Random Utility Models with Subset Choices. 23rd International Conference on Artificial Intelligence and Statistics (AISTATS), 2020.