Independently Controllable Features

TL;DR

This paper explores how agents in interactive environments can identify controllable features by experimenting with actions, aiming to disentangle underlying causes of data variation.

Contribution

It introduces a novel idea that controllable latent factors can be discovered through an agent’s ability to influence data in interactive settings.

Findings

Preliminary experiments show the feasibility of identifying controllable features.

The naive method can find factors related to agent actions.

Results suggest potential for improved feature disentanglement in interactive environments.

Abstract

Finding features that disentangle the different causes of variation in real data is a difficult task, that has nonetheless received considerable attention in static domains like natural images. Interactive environments, in which an agent can deliberately take actions, offer an opportunity to tackle this task better, because the agent can experiment with different actions and observe their effects. We introduce the idea that in interactive environments, latent factors that control the variation in observed data can be identified by figuring out what the agent can control. We propose a naive method to find factors that explain or measure the effect of the actions of a learner, and test it in illustrative experiments.