# A Closer Look at Deep Policy Gradients

arXiv: https://arxiv.org/abs/1811.02553 · HTML: https://mlsanity.com/paper/1811.02553/

## Key facts

- Cited by 12 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 6 Nov 2018.

## Built on by (most cited first, 12 of 12)

- [Uncovering Instabilities in Variational-Quantum Deep Q-Networks](https://mlsanity.com/paper/2202.05195/) (2022)
- [Differentiable Discrete Event Simulation for Queuing Network Control](https://mlsanity.com/paper/2409.03740/) (2024)
- [PPO in the Fisher-Rao geometry](https://mlsanity.com/paper/2506.03757/) (2025)
- [Learning Robust Policies for Generalized Debris Capture with an Automated Tether-Net System](https://mlsanity.com/paper/2201.04180/) (2022)
- [Sample Dropout: A Simple yet Effective Variance Reduction Technique in Deep Policy Optimization](https://mlsanity.com/paper/2302.02299/) (2023)
- [Preventing Learning Stagnation in PPO by Scaling to 1 Million Parallel Environments](https://mlsanity.com/paper/2603.06009/) (2026)
- [Comparative Analysis and Parametric Tuning of PPO, GRPO, and DAPO for LLM Reasoning Enhancement](https://mlsanity.com/paper/2512.07611/) (2025)
- [Processing Network Controls via Deep Reinforcement Learning](https://mlsanity.com/paper/2205.02119/) (2022)
- [Entropy Augmented Reinforcement Learning](https://mlsanity.com/paper/2208.09322/) (2022)
- [Decentralized Coordination of Distributed Energy Resources through Local Energy Markets and Deep Reinforcement Learning](https://mlsanity.com/paper/2404.13142/) (2024)
- [Graph-attention-based Casual Discovery with Trust Region-navigated Clipping Policy Optimization](https://mlsanity.com/paper/2412.19578/) (2024)
- [ANO: Robust Policy Optimization via Bounded, Redescending Gain Fields](https://mlsanity.com/paper/2605.02320/) (2026)

## Abstract (arXiv)

We study how the behavior of deep policy gradient algorithms reflects the conceptual framework motivating their development. To this end, we propose a fine-grained analysis of state-of-the-art methods based on key elements of this framework: gradient estimation, value prediction, and optimization landscapes. Our results show that the behavior of deep policy gradient algorithms often deviates from what their motivating framework would predict: the surrogate objective does not match the true reward landscape, learned value estimators fail to fit the true value function, and gradient estimates poorly correlate with the "true" gradient. The mismatch between predicted and empirical behavior we uncover highlights our poor understanding of current methods, and indicates the need to move beyond current benchmark-centric evaluation methods.