Operationally meaningful representations of physical systems in neural networks

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.

24 pages, 13 figures

References in corpus (13)

• Reinforcement Learning with Unsupervised Auxiliary Tasks
• Entanglement by Path Identity
• Neural network quantum state tomography in a two-qubit experiment
• Quantum circuit optimization with deep reinforcement learning
• Visualizing operators of coupled spin systems
• Attention-based Quantum Tomography
• Independently Controllable Features
• Reinforcement Learning through Asynchronous Advantage Actor-Critic on a GPU
• Skill Learning by Autonomous Robotic Playing using Active Learning and Creativity
• How a minimal learning agent can infer the existence of unobserved variables in a complex environment
• Learning Generalizable Physical Dynamics of 3D Rigid Objects
• Learning the Arrow of Time
• Adding Intuitive Physics to Neural-Symbolic Capsules Using Interaction Networks

Cited by in corpus (14)

• Experimental quantum speed-up in reinforcement learning agents
• Computer-inspired Quantum Experiments
• Learning models of quantum systems from experiments
• Learning minimal representations of stochastic processes with variational autoencoders
• Hybrid discrete-continuous compilation of trapped-ion quantum circuits with deep reinforcement learning
• Deep learning insights into cosmological structure formation
• Explainable Representation Learning of Small Quantum States
• $\beta$-Variational Autoencoder as an Entanglement Classifier
• Observing Schrödinger's Cat with Artificial Intelligence: Emergent Classicality from Information Bottleneck
• Learning the dynamics of Markovian open quantum systems from experimental data
• Compressed Hierarchical Representations for Multi-Task Learning and Task Clustering
• Forming complex neurons by four-wave mixing in a Bose-Einstein condensate
• Interpretable representation learning of quantum data enabled by probabilistic variational autoencoders
• Learning Physical Concepts in Cyber-Physical Systems: A Case Study