ANS: Adaptive Network Scaling for Deep Rectifier Reinforcement Learning Models
This addresses performance optimization for deep RL models using ReLU activations, but appears incremental as it focuses on scaling adjustments.
The paper tackled the problem of how reward scaling affects deep reinforcement learning agents with ReLU networks, proposing an Adaptive Network Scaling framework to find suitable reward scales and reporting empirical results.
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.