Gradient dynamics in reinforcement learning
arXiv:2204.03973 · doi:10.1103/PhysRevE.106.025315
Abstract
Despite the success achieved by the analysis of supervised learning algorithms in the framework of statistical mechanics, reinforcement learning has remained largely untouched. Here we move towards closing the gap by analyzing the dynamics of the policy gradient algorithm. For a convex problem, we show that it obeys a drift-diffusion motion with coeffcients tuned by learning rate. Furthermore, we propose a mapping between a non-convex reinforcement learning problem and a disordered system. This mapping enables us to show how the learning rate acts as an effective temperature and thus is capable of smoothing rough landscapes, corroborating what is displayed by the drift-diffusive description and paving the way for physics-inspired algorithmic optimization based on annealing procedures in disordered systems.
arXiv comment15 pages, 6 figures. Submitted to Physical Review E
References in corpus (22)
- Machine learning and the physical sciences
- Searching for Exotic Particles in High-Energy Physics with Deep Learning
- Unsupervised learning of phase transitions: from principal component analysis to variational autoencoders
- Flow Navigation by Smart Microswimmers via Reinforcement Learning
- Boosting search by rare events
- Unreasonable Effectiveness of Learning Neural Networks: From Accessible States and Robust Ensembles to Basic Algorithmic Schemes
- Biological evolution through mutation, selection, and drift: An introductory review
- The Dynamic Phase Transition for Decoding Algorithms
- Learning physical properties of liquid crystals with deep convolutional neural networks
- Origin of the computational hardness for learning with binary synapses
- Hysteretic Optimization
- Matrix Games, Mixed Strategies, and Statistical Mechanics
- Optimization by thermal cycling
- Renormalization for Discrete Optimization
- Determining liquid crystal properties with ordinal networks and machine learning
- Replica Symmetry Breaking in the Random Replicant Model
- Statistical mechanics of the vertex-cover problem
- Statistical mechanics of random two-player games
- Entropy and typical properties of Nash equilibria in two-player games
- Jet grooming through reinforcement learning
- Two-population replicator dynamics and number of Nash equilibria in random matrix games
- Emergence of Lévy walks from second order stochastic optimization