Operationally meaningful representations of physical systems in neural networks

TL;DR

This paper introduces a neural network architecture that learns operationally meaningful representations of physical systems, enabling efficient communication of relevant information and separation of parameters in classical and quantum physics.

Contribution

It proposes a novel neural network design that encodes explicit, operationally meaningful information about physical systems, including quantum correlations, and integrates with reinforcement learning for interactive exploration.

Findings

Learns compact representations separating local and correlation parameters.

Effectively applies to classical and quantum systems.

Enables exploration of experimental settings to identify relevant variables.

Abstract

To make progress in science, we often build abstract representations of physical systems that meaningfully encode information about the systems. The representations learnt by most current machine learning techniques reflect statistical structure present in the training data; however, these methods do not allow us to specify explicit and operationally meaningful requirements on the representation. Here, we present a neural network architecture based on the notion that agents dealing with different aspects of a physical system should be able to communicate relevant information as efficiently as possible to one another. This produces representations that separate different parameters which are useful for making statements about the physical system in different experimental settings. We present examples involving both classical and quantum physics. For instance, our architecture finds a compact representation of an arbitrary two-qubit system that separates local parameters from parameters describing quantum correlations. We further show that this method can be combined with reinforcement learning to enable representation learning within interactive scenarios where agents need to explore experimental settings to identify relevant variables.