Symbolic Pregression: Discovering Physical Laws from Distorted Video
arXiv:2005.11212 · doi:10.1103/PhysRevE.103.043307
Abstract
We present a method for unsupervised learning of equations of motion for objects in raw and optionally distorted unlabeled video. We first train an autoencoder that maps each video frame into a low-dimensional latent space where the laws of motion are as simple as possible, by minimizing a combination of non-linearity, acceleration and prediction error. Differential equations describing the motion are then discovered using Pareto-optimal symbolic regression. We find that our pre-regression ("pregression") step is able to rediscover Cartesian coordinates of unlabeled moving objects even when the video is distorted by a generalized lens. Using intuition from multidimensional knot-theory, we find that the pregression step is facilitated by first adding extra latent space dimensions to avoid topological problems during training and then removing these extra dimensions via principal component analysis.
arXiv commentExpanded and improved physics discussion, additional method details. 9 pages, 7 figs
References in corpus (28)
- SciPy 1.0--Fundamental Algorithms for Scientific Computing in Python
- Auto-Encoding Variational Bayes
- Representation Learning with Contrastive Predictive Coding
- Discovering governing equations from data: Sparse identification of nonlinear dynamical systems
- Unsupervised Learning of Video Representations using LSTMs
- Machine learning phases of matter
- Learning phase transitions by confusion
- Discovering physical concepts with neural networks
- Hamiltonian Neural Networks
- Action-Conditional Video Prediction using Deep Networks in Atari Games
- Decomposing Motion and Content for Natural Video Sequence Prediction
- Deep Predictive Coding Networks for Video Prediction and Unsupervised Learning
- Video (language) modeling: a baseline for generative models of natural videos
- Accelerating Eulerian Fluid Simulation With Convolutional Networks
- Learning Thermodynamics with Boltzmann Machines
- Unsupervised Learning of Disentangled Representations from Video
- Mutual Information, Neural Networks and the Renormalization Group
- Sparse Identification of Nonlinear Dynamics for Rapid Model Recovery
- Toward an AI Physicist for Unsupervised Learning
- AI Feynman 2.0: Pareto-optimal symbolic regression exploiting graph modularity
- Stochastic Variational Video Prediction
- Deep Learning the Quantum Phase Transitions in Random Electron Systems: Applications to Three Dimensions
- A General Deep Learning Framework for Network Reconstruction and Dynamics Learning
- Learning phase transitions from dynamics
- Unsupervised Learning of Latent Physical Properties Using Perception-Prediction Networks
- Extracting Interpretable Physical Parameters from Spatiotemporal Systems using Unsupervised Learning
- Video Ladder Networks
- Comment on the article "Distilling free-form natural laws from experimental data"
Cited by in corpus (12)
- AI Feynman 2.0: Pareto-optimal symbolic regression exploiting graph modularity
- Machine-Learning Non-Conservative Dynamics for New-Physics Detection
- Learning phase transitions from regression uncertainty: A new regression-based machine learning approach for automated detection of phases of matter
- Learning Potentials of Quantum Systems using Deep Neural Networks
- Noether Networks: Meta-Learning Useful Conserved Quantities
- Uncovering Closed-form Governing Equations of Nonlinear Dynamics from Videos
- Reveal flocking phase transition of self-propelled active particles by machine learning regression uncertainty
- SyMetric: Measuring the Quality of Learnt Hamiltonian Dynamics Inferred from Vision
- Similarity-Based Equational Inference in Physics
- Physics-Informed Deep Learning Characterizes Morphodynamics of Asian Soybean Rust Disease
- Physical discovery in representation learning via conditioning on prior knowledge: applications for ferroelectric domain dynamics
- Deep learning of thermodynamic laws from microscopic dynamics