ANS: Adaptive Network Scaling for Deep Rectifier Reinforcement Learning Models
arXiv:1809.02112
Abstract
This work provides a thorough study on how reward scaling can affect performance of deep reinforcement learning agents. In particular, we would like to answer the question that how does reward scaling affect non-saturating ReLU networks in RL? This question matters because ReLU is one of the most effective activation functions for deep learning models. We also propose an Adaptive Network Scaling framework to find a suitable scale of the rewards during learning for better performance. We conducted empirical studies to justify the solution.
References in corpus (8)
- Adam: A Method for Stochastic Optimization
- Continuous control with deep reinforcement learning
- Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)
- Dueling Network Architectures for Deep Reinforcement Learning
- Benchmarking Deep Reinforcement Learning for Continuous Control
- Q-Prop: Sample-Efficient Policy Gradient with An Off-Policy Critic
- Learning values across many orders of magnitude
- On Characterizing the Capacity of Neural Networks using Algebraic Topology