Discovering physical concepts with neural networks

Abstract

Despite the success of neural networks at solving concrete physics problems, their use as a general-purpose tool for scientific discovery is still in its infancy. Here, we approach this problem by modelling a neural network architecture after the human physical reasoning process, which has similarities to representation learning. This allows us to make progress towards the long-term goal of machine-assisted scientific discovery from experimental data without making prior assumptions about the system. We apply this method to toy examples and show that the network finds the physically relevant parameters, exploits conservation laws to make predictions, and can help to gain conceptual insights, e.g. Copernicus' conclusion that the solar system is heliocentric.

arXiv comment4 pages main text + 11 pages appendix, changes since v2: improved references and presentation

References in corpus (41)

• Auto-Encoding Variational Bayes
• Discovering governing equations from data: Sparse identification of nonlinear dynamical systems
• The information bottleneck method
• Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)
• Solving the Quantum Many-Body Problem with Artificial Neural Networks
• Deep learning for universal linear embeddings of nonlinear dynamics
• Hidden Physics Models: Machine Learning of Nonlinear Partial Differential Equations
• Many-body quantum state tomography with neural networks
• Opening the Black Box of Deep Neural Networks via Information
• Disentangling by Factorising
• Interaction Networks for Learning about Objects, Relations and Physics
• Quantum Entanglement in Neural Network States
• Efficient Representation of Quantum Many-body States with Deep Neural Networks
• Restricted-Boltzmann-Machine Learning for Solving Strongly Correlated Quantum Systems
• Reconstructing quantum states with generative models
• Active learning machine learns to create new quantum experiments
• Machine Learning Topological States
• Learning Thermodynamics with Boltzmann Machines
• Automated Search for new Quantum Experiments
• Approximating quantum many-body wave-functions using artificial neural networks
• Neural-Network Quantum States, String-Bond States, and Chiral Topological States
• Learning New Physics from a Machine
• Mutual Information, Neural Networks and the Renormalization Group
• A Compositional Object-Based Approach to Learning Physical Dynamics
• Constructing exact representations of quantum many-body systems with deep neural networks
• Guiding New Physics Searches with Unsupervised Learning
• A Disentangled Recognition and Nonlinear Dynamics Model for Unsupervised Learning
• Toward an AI Physicist for Unsupervised Learning
• Learning Koopman Invariant Subspaces for Dynamic Mode Decomposition
• Learning hard quantum distributions with variational autoencoders
• Discovering objects and their relations from entangled scene representations
• Neural network representation of tensor network and chiral states
• Deep Learning the Quantum Phase Transitions in Random Electron Systems: Applications to Three Dimensions
• Learning phase transitions from dynamics
• Extracting dynamical equations from experimental data is NP-hard
• Quantum dynamics in transverse-field Ising models from classical networks
• Automated discovery of characteristic features of phase transitions in many-body localization
• Unsupervised Learning of Latent Physical Properties Using Perception-Prediction Networks
• QuCumber: wavefunction reconstruction with neural networks
• Comment on the article "Distilling free-form natural laws from experimental data"
• Autoencoding topology

Cited by in corpus (158)

• Explainable Machine Learning for Scientific Insights and Discoveries
• Self-Referencing Embedded Strings (SELFIES): A 100% robust molecular string representation
• Social physics
• Deep learning in nano-photonics: inverse design and beyond
• Hamiltonian Neural Networks
• On scientific understanding with artificial intelligence
• SELFIES and the future of molecular string representations
• Towards high-throughput superconductor discovery via machine learning
• Improving aircraft performance using machine learning: a review
• The 2021 Quantum Materials Roadmap
• Optimizing Quantum Error Correction Codes with Reinforcement Learning
• A Review of Machine Learning Methods Applied to Structural Dynamics and Vibroacoustic
• Machine Learning for Quantum Matter
• Robust learning from noisy, incomplete, high-dimensional experimental data via physically constrained symbolic regression
• Deep Learning for Feynman's Path Integral in Strong-Field Time-Dependent Dynamics
• Unsupervised phase discovery with deep anomaly detection
• Review: Deep Learning in Electron Microscopy
• Data-driven recovery of hidden physics in reduced order modeling of fluid flows
• Unveiling phase transitions with machine learning
• Weakly-Supervised Disentanglement Without Compromises
• Deep learning to discover and predict dynamics on an inertial manifold
• Computer-inspired Quantum Experiments
• Autonomous inference of complex network dynamics from incomplete and noisy data
• Towards a Collective Agenda on AI for Earth Science Data Analysis
• Physics enhanced neural networks predict order and chaos
• Exploring QCD matter in extreme conditions with Machine Learning
• Unsupervised machine learning of topological phase transitions from experimental data
• Modeling System Dynamics with Physics-Informed Neural Networks Based on Lagrangian Mechanics
• Learning models of quantum systems from experiments
• A neural network oracle for quantum nonlocality problems in networks
• Discovering Symmetry Invariants and Conserved Quantities by Interpreting Siamese Neural Networks
• Data-Driven Reduced-Order Modeling of Spatiotemporal Chaos with Neural Ordinary Differential Equations
• Machine Learning meets Quantum Foundations: A Brief Survey
• The Ramanujan Machine: Automatically Generated Conjectures on Fundamental Constants
• Symbolic Pregression: Discovering Physical Laws from Distorted Video
• Conceptual understanding through efficient inverse-design of quantum optical experiments
• Breaking Adiabatic Quantum Control with Deep Learning
• Deep Molecular Dreaming: Inverse machine learning for de-novo molecular design and interpretability with surjective representations
• Manifold Learning for Knowledge Discovery and Intelligent Inverse Design of Photonic Nanostructures: Breaking the Geometric Complexity
• Adaptable Hamiltonian neural networks
• Mutual information disentangles interactions from changing environments
• Deep neural networks for the prediction of the optical properties and the free-form inverse design of metamaterials
• Learning Hamiltonian dynamics by reservoir computer
• Soft Metamaterials: Adaptation and Intelligence
• Variational Quantum Anomaly Detection: Unsupervised mapping of phase diagrams on a physical quantum computer
• Classifying global state preparation via deep reinforcement learning
• Extracting Interpretable Physical Parameters from Spatiotemporal Systems using Unsupervised Learning
• Unsupervised learning using topological data augmentation
• Estimating the degree of non-Markovianity using machine learning
• Screening and understanding Li adsorption on 2-dimensional metallic materials by learning physics
• Scientific intuition inspired by machine learning generated hypotheses
• Deep learning-based design of broadband GHz complex and random metasurfaces
• Flexible learning of quantum states with generative query neural networks
• Learning quantum dynamics with latent neural ODEs
• Incorporating Physical Knowledge into Machine Learning for Planetary Space Physics
• Measuring the anomalous quartic gauge couplings in the $W^{+}{W}^{-}\to W^{+}{W}^{-}$ process at muon collider using artificial neural networks
• Crowdsourcing human common sense for quantum control
• Solving time dependent Fokker-Planck equations via temporal normalizing flow
• An Interpretable Machine Learning Model for Deformation of Multi-Walled Carbon Nanotubes
• Probing non-Markovian quantum dynamics with data-driven analysis: Beyond "black-box" machine learning models
• Operationally meaningful representations of physical systems in neural networks
• Learning Interpretable Representations of Entanglement in Quantum Optics Experiments using Deep Generative Models
• mHealth hyperspectral learning for instantaneous spatiospectral imaging of hemodynamics
• Using a nested anomaly detection machine learning algorithm to study the neutral triple gauge couplings at an \texorpdfstring{$e^{+}{e}^{-}$}{e+e-} collider
• Detecting anomalous quartic gauge couplings using the isolation forest machine learning algorithm
• Convolutional-network models to predict wall-bounded turbulence from wall quantities
• Deep learning analysis of polaritonic waves images
• Searching for anomalous quartic gauge couplings at muon colliders using principle component analysis
• Learning and correcting non-Gaussian model errors
• Machines Learn to Infer Stellar Parameters Just by Looking at a Large Number of Spectra
• Physics-informed Information Field Theory for Modeling Physical Systems with Uncertainty Quantification
• Extract the energy scale of anomalous $\gamma \gamma \to W^{+}{W}^{-}$ scattering in the vector boson scattering process using artificial neural networks
• From observations to complexity of quantum states via unsupervised learning
• MLAnalysis: An open-source program for high energy physics analyses
• Explaining dark matter halo density profiles with neural networks
• Self-supervised Learning with Physics-aware Neural Networks I: Galaxy Model Fitting
• Is the Machine Smarter than the Theorist: Deriving Formulas for Particle Kinematics with Symbolic Regression
• Learning about learning by many-body systems
• Deep neural networks for inverse problems in mesoscopic physics: Characterization of the disorder configuration from quantum transport properties
• Tomography of time-dependent quantum spin networks with machine learning
• Neural-network-based parameter estimation for quantum detection
• Data-driven geophysics: from dictionary learning to deep learning
• Deep learning for nano-photonic materials -- The solution to everything!?
• Estimating the Euclidean quantum propagator with deep generative modeling of Feynman paths
• Learning minimal representations of stochastic processes with variational autoencoders
• Learning Order Parameters from Videos of Dynamical Phases for Skyrmions with Neural Networks
• Exploring super-additivity of coherent information of noisy quantum channels through Genetic algorithms
• Classification and reconstruction of optical quantum states with deep neural networks
• Deep learning insights into cosmological structure formation
• Neural networks in quantum many-body physics: a hands-on tutorial
• Graph Laplacians, Riemannian Manifolds and their Machine-Learning
• Machine learning in interpretation of electronic core-level spectra
• Physically interpretable machine learning algorithm on multidimensional non-linear fields
• Explainable Representation Learning of Small Quantum States
• Using k-means assistant event selection strategy to study anomalous quartic gauge couplings at muon colliders
• Learning Generalized Quasi-Geostrophic Models Using Deep Neural Numerical Models
• Sparse Symplectically Integrated Neural Networks
• Quantum Similarity Testing with Convolutional Neural Networks
• Noise-Robust End-to-End Quantum Control using Deep Autoregressive Policy Networks
• Physical Neural Networks with Self-Learning Capabilities
• Discover the GellMann-Okubo formula with machine learning
• Metalearning generalizable dynamics from trajectories
• Discovering Sparse Representations of Lie Groups with Machine Learning
• Evolving Quantum Circuits
• Classification of Equation of State in Relativistic Heavy-Ion Collisions Using Deep Learning
• Unsupervised Learning of Rydberg Atom Array Phase Diagram with Siamese Neural Networks
• Chaotic semi-express buses in a loop
• Optimize the event selection strategy to study the anomalous quartic gauge couplings at muon colliders using the support vector machine and quantum support vector machine
• Renormalized Mutual Information for Artificial Scientific Discovery
• Sparsely constrained neural networks for model discovery of PDEs
• Deep learning insights into non-universality in the halo mass function
• Learning Potentials of Quantum Systems using Deep Neural Networks
• Universes as Big Data
• Rapidly Encoding Generalizable Dynamics in a Euclidean Symmetric Neural Network
• Observing Schrödinger's Cat with Artificial Intelligence: Emergent Classicality from Information Bottleneck
• Using machine learning to find exact analytic solutions to analytically posed physics problems
• Robustness Assessment of Complex Networks using the Idle Network
• High-dimentional Multipartite Entanglement Structure Detection with Low Cost
• Thermodynamics of the Ising model encoded in restricted Boltzmann machines
• $\beta$-Variational Autoencoder as an Entanglement Classifier
• Visual Physics: Discovering Physical Laws from Videos
• A Parametric Level Set Method for Topology Optimization based on Deep Neural Network (DNN)
• Data-Space Validation of High-Dimensional Models by Comparing Sample Quantiles
• MC-LSTM: Mass-Conserving LSTM
• Exploring fundamental laws of classical mechanics via predicting the orbits of planets based on neural networks
• Observing how deep neural networks understand physics through the energy spectrum of one-dimensional quantum mechanics
• Joint Parameter Discovery and Generative Modeling of Dynamic Systems
• Deep learning reveals hidden interactions in complex systems
• Automated Discovery of Coupled Mode Setups
• Digital Discovery of a Scientific Concept at the Core of Experimental Quantum Optics
• Intelligence, physics and information -- the tradeoff between accuracy and simplicity in machine learning
• Time-Reversal Symmetric ODE Network
• Causal Learning by a Robot with Semantic-Episodic Memory in an Aesop's Fable Experiment
• Boosting on the shoulders of giants in quantum device calibration
• Neural Network Statistical Mechanics
• Deep Quantum Graph Dreaming: Deciphering Neural Network Insights into Quantum Experiments
• Deep-Learned Event Variables for Collider Phenomenology
• A quantum inspired approach to learning dynamical laws from data -- block-sparsity and gauge-mediated weight sharing
• A Use of Even Activation Functions in Neural Networks
• Rodent: Relevance determination in differential equations
• Topological Learning in Multi-Class Data Sets
• Artificial intelligence for representing and characterizing quantum systems
• Model-Independent Quantum Phases Classifier
• Adding Intuitive Physics to Neural-Symbolic Capsules Using Interaction Networks
• Similarity-Based Equational Inference in Physics
• Variational Integrator Graph Networks for Learning Energy Conserving Dynamical Systems
• Learning Continuous Face Representation with Explicit Functions
• Off-the-shelf deep learning is not enough: parsimony, Bayes and causality
• Toward Building Science Discovery Machines
• Deep learning of thermodynamic laws from microscopic dynamics
• AI Research Associate for Early-Stage Scientific Discovery
• Principle of learning sign rules by neural networks in qubit lattice models
• Discovering the building blocks of dark matter halo density profiles with neural networks
• Learning Physical Concepts in Cyber-Physical Systems: A Case Study
• Data-Driven Continuum Dynamics via Transport-Teleport Duality
• A general framework for knowledge integration in machine learning for electromagnetic scattering using quasinormal modes
• Learning relevant features for statistical inference
• Interpretable representation learning of quantum data enabled by probabilistic variational autoencoders