Loaded DiCE: Trading off Bias and Variance in Any-Order Score Function Estimators for Reinforcement Learning
arXiv:1909.10549
Abstract
Gradient-based methods for optimisation of objectives in stochastic settings with unknown or intractable dynamics require estimators of derivatives. We derive an objective that, under automatic differentiation, produces low-variance unbiased estimators of derivatives at any order. Our objective is compatible with arbitrary advantage estimators, which allows the control of the bias and variance of any-order derivatives when using function approximation. Furthermore, we propose a method to trade off bias and variance of higher order derivatives by discounting the impact of more distant causal dependencies. We demonstrate the correctness and utility of our objective in analytically tractable MDPs and in meta-reinforcement-learning for continuous control.
References in corpus (8)
- Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
- High-Dimensional Continuous Control Using Generalized Advantage Estimation
- An Introduction to Deep Reinforcement Learning
- Counterfactual Multi-Agent Policy Gradients
- Gradient Estimation Using Stochastic Computation Graphs
- The Optimal Reward Baseline for Gradient-Based Reinforcement Learning
- Monte Carlo Gradient Estimation in Machine Learning
- Some Considerations on Learning to Explore via Meta-Reinforcement Learning