Learning to Maximize Mutual Information for Dynamic Feature Selection

TL;DR

This paper introduces a simple, greedy, mutual information-based approach for dynamic feature selection, which outperforms existing methods and is trained via amortized optimization to approximate the optimal policy.

Contribution

It proposes a novel learning method for dynamic feature selection based on mutual information, bridging the gap between theoretical appeal and practical implementation.

Findings

Outperforms existing feature selection methods in experiments

Recovers the greedy policy when trained optimally

Validates the approach as simple yet powerful

Abstract

Feature selection helps reduce data acquisition costs in ML, but the standard approach is to train models with static feature subsets. Here, we consider the dynamic feature selection (DFS) problem where a model sequentially queries features based on the presently available information. DFS is often addressed with reinforcement learning, but we explore a simpler approach of greedily selecting features based on their conditional mutual information. This method is theoretically appealing but requires oracle access to the data distribution, so we develop a learning approach based on amortized optimization. The proposed method is shown to recover the greedy policy when trained to optimality, and it outperforms numerous existing feature selection methods in our experiments, thus validating it as a simple but powerful approach for this problem.