TherML: Thermodynamics of Machine Learning
arXiv:1807.04162
Abstract
In this work we offer a framework for reasoning about a wide class of existing objectives in machine learning. We develop a formal correspondence between this work and thermodynamics and discuss its implications.
Presented at the ICML 2018 workshop on Theoretical Foundations and Applications of Deep Generative Models
References in corpus (7)
- Weight Uncertainty in Neural Networks
- Gradient-based Hyperparameter Optimization through Reversible Learning
- Deep Variational Information Bottleneck
- The thermodynamics of prediction
- A Complete Recipe for Stochastic Gradient MCMC
- Multivariate Information Bottleneck
- A Bayesian Perspective on Generalization and Stochastic Gradient Descent
Cited by in corpus (13)
- Recent Advances in Autoencoder-Based Representation Learning
- State Predictive Information Bottleneck
- Dynamics-Aware Unsupervised Discovery of Skills
- Action and Perception as Divergence Minimization
- Learning about learning by many-body systems
- Differentiable Physics: A Position Piece
- Pareto-optimal clustering with the primal deterministic information bottleneck
- The Physics of Learning
- Dueling Decoders: Regularizing Variational Autoencoder Latent Spaces
- Likelihood Ratio Exponential Families
- Understanding Learning Dynamics of Binary Neural Networks via Information Bottleneck
- A Free-Energy Principle for Representation Learning
- Quantifying many-body learning far from equilibrium with representation learning