Implementation Matters in Deep Policy Gradients: A Case Study on PPO and TRPO
Logan Engstrom, Andrew Ilyas, Shibani Santurkar, Dimitris Tsipras, Firdaus Janoos + 2 more
TLDR
This paper reveals that implementation-specific optimizations significantly influence the performance differences between PPO and TRPO in deep reinforcement learning.
Key contributions
- Identifies that code-level optimizations, often considered secondary, substantially impact agent performance.
- Demonstrates that these optimizations account for most of PPO's cumulative reward improvements over TRPO.
- Highlights how such implementation details fundamentally alter the behavior and effectiveness of policy gradient methods.
Why it matters
Understanding the true sources of performance gains in deep reinforcement learning is crucial for fair algorithm comparison and progress. This paper shows that subtle implementation choices, rather than core algorithmic innovations alone, can drive major improvements, emphasizing the need for careful evaluation and transparency in RL research.
Original Abstract
We study the roots of algorithmic progress in deep policy gradient algorithms through a case study on two popular algorithms: Proximal Policy Optimization (PPO) and Trust Region Policy Optimization (TRPO). Specifically, we investigate the consequences of "code-level optimizations:" algorithm augmentations found only in implementations or described as auxiliary details to the core algorithm. Seemingly of secondary importance, such optimizations turn out to have a major impact on agent behavior. Our results show that they (a) are responsible for most of PPO's gain in cumulative reward over TRPO, and (b) fundamentally change how RL methods function. These insights show the difficulty and importance of attributing performance gains in deep reinforcement learning. Code for reproducing our results is available at https://github.com/MadryLab/implementation-matters .
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.