Unsupervised Learning of Latent Physical Properties Using Perception-Prediction Networks

Abstract

We propose a framework for the completely unsupervised learning of latent object properties from their interactions: the perception-prediction network (PPN). Consisting of a perception module that extracts representations of latent object properties and a prediction module that uses those extracted properties to simulate system dynamics, the PPN can be trained in an end-to-end fashion purely from samples of object dynamics. The representations of latent object properties learned by PPNs not only are sufficient to accurately simulate the dynamics of systems comprised of previously unseen objects, but also can be translated directly into human-interpretable properties (e.g., mass, coefficient of restitution) in an entirely unsupervised manner. Crucially, PPNs also generalize to novel scenarios: their gradient-based training can be applied to many dynamical systems and their graph-based structure functions over systems comprised of different numbers of objects. Our results demonstrate the efficacy of graph-based neural architectures in object-centric inference and prediction tasks, and our model has the potential to discover relevant object properties in systems that are not yet well understood.

UAI 2018 (oral)

Cited by in corpus (19)

• Discovering physical concepts with neural networks
• Integration of Neural Network-Based Symbolic Regression in Deep Learning for Scientific Discovery
• Interaction networks for the identification of boosted $H\to b\overline{b}$ decays
• Symbolic Pregression: Discovering Physical Laws from Distorted Video
• Visual Dynamics: Stochastic Future Generation via Layered Cross Convolutional Networks
• CoPhy: Counterfactual Learning of Physical Dynamics
• Operationally meaningful representations of physical systems in neural networks
• Physics-as-Inverse-Graphics: Unsupervised Physical Parameter Estimation from Video
• Estimating Mass Distribution of Articulated Objects using Non-prehensile Manipulation
• Object and Relation Centric Representations for Push Effect Prediction
• Graph Neural Reasoning May Fail in Certifying Boolean Unsatisfiability
• Neural Message Passing for Multi-Label Classification
• Intelligence, physics and information -- the tradeoff between accuracy and simplicity in machine learning
• Deep learning reveals hidden interactions in complex systems
• Uncovering Closed-form Governing Equations of Nonlinear Dynamics from Videos
• Physical Primitive Decomposition
• Adding Intuitive Physics to Neural-Symbolic Capsules Using Interaction Networks
• Predicting the Physical Dynamics of Unseen 3D Objects
• Toward Building Science Discovery Machines