Toward an AI Physicist for Unsupervised Learning

TL;DR

This paper introduces an AI Physicist framework that uses theories and specialized learning to improve unsupervised physics prediction, achieving faster learning and more accurate models than standard neural networks.

Contribution

It proposes a novel theory-based paradigm with a generalized-mean-loss and description length objective, enabling theories to specialize, unify, and be expressed as symbolic formulas in unsupervised learning.

Findings

Achieves billion-fold reduction in prediction error compared to standard neural nets.

Successfully learns and identifies physical laws in complex environments.

Recovers exact integer and rational parameters in physics simulations.

Abstract

We investigate opportunities and challenges for improving unsupervised machine learning using four common strategies with a long history in physics: divide-and-conquer, Occam's razor, unification and lifelong learning. Instead of using one model to learn everything, we propose a novel paradigm centered around the learning and manipulation of *theories*, which parsimoniously predict both aspects of the future (from past observations) and the domain in which these predictions are accurate. Specifically, we propose a novel generalized-mean-loss to encourage each theory to specialize in its comparatively advantageous domain, and a differentiable description length objective to downweight bad data and "snap" learned theories into simple symbolic formulas. Theories are stored in a "theory hub", which continuously unifies learned theories and can propose theories when encountering new environments. We test our implementation, the toy "AI Physicist" learning agent, on a suite of increasingly complex physics environments. From unsupervised observation of trajectories through worlds involving random combinations of gravity, electromagnetism, harmonic motion and elastic bounces, our agent typically learns faster and produces mean-squared prediction errors about a billion times smaller than a standard feedforward neural net of comparable complexity, typically recovering integer and rational theory parameters exactly. Our agent successfully identifies domains with different laws of motion also for a nonlinear chaotic double pendulum in a piecewise constant force field.