# Dissecting Deep RL with High Update Ratios: Combatting Value Divergence

arXiv: https://arxiv.org/abs/2403.05996 · HTML: https://mlsanity.com/paper/2403.05996/

## Key facts

- Cited by 17 papers in the map (data through 14 Oct 2026).
- Placed in Deep RL algorithms and representations, one of 15 directions in Deep reinforcement learning methods.
- Builds on 49 of its 77 references that are in the map.
- First posted on arXiv 9 Mar 2024.

## Placed in

- Deep reinforcement learning methods / [Deep RL algorithms and representations](https://mlsanity.com/deeprl/deep-rl-algorithms-and-representations/)

## Builds on (most cited first, 30 of 49)

- [Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor](https://mlsanity.com/paper/1801.01290/) (2018)
- [Continuous control with deep reinforcement learning](https://mlsanity.com/paper/1509.02971/) (2015)
- [Playing Atari with Deep Reinforcement Learning](https://mlsanity.com/paper/1312.5602/) (2013)
- [Addressing Function Approximation Error in Actor-Critic Methods](https://mlsanity.com/paper/1802.09477/) (2018)
- [Deep Reinforcement Learning with Double Q-learning](https://mlsanity.com/paper/1509.06461/) (2015)
- [Root Mean Square Layer Normalization](https://mlsanity.com/paper/1910.07467/) (2019)
- [Off-Policy Deep Reinforcement Learning without Exploration](https://mlsanity.com/paper/1812.02900/) (2018)
- [Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)](https://mlsanity.com/paper/1511.07289/) (2015)
- [A Minimalist Approach to Offline Reinforcement Learning](https://mlsanity.com/paper/2106.06860/) (2021)
- [Deep Reinforcement Learning at the Edge of the Statistical Precipice](https://mlsanity.com/paper/2108.13264/) (2021)
- [When to Trust Your Model: Model-Based Policy Optimization](https://mlsanity.com/paper/1906.08253/) (2019)
- [A Distributional Perspective on Reinforcement Learning](https://mlsanity.com/paper/1707.06887/) (2017)
- [TD-MPC2: Scalable, Robust World Models for Continuous Control](https://mlsanity.com/paper/2310.16828/) (2023)
- [dm_control: Software and Tasks for Continuous Control](https://mlsanity.com/paper/2006.12983/) (2020)
- [Efficient Online Reinforcement Learning with Offline Data](https://mlsanity.com/paper/2302.02948/) (2023)
- [Randomized Ensembled Double Q-Learning: Learning Fast Without a Model](https://mlsanity.com/paper/2101.05982/) (2021)
- [The Primacy Bias in Deep Reinforcement Learning](https://mlsanity.com/paper/2205.07802/) (2022)
- [Understanding and Improving Layer Normalization](https://mlsanity.com/paper/1911.07013/) (2019)
- [Revisiting Fundamentals of Experience Replay](https://mlsanity.com/paper/2007.06700/) (2020)
- [The Dormant Neuron Phenomenon in Deep Reinforcement Learning](https://mlsanity.com/paper/2302.12902/) (2023)
- [Understanding plasticity in neural networks](https://mlsanity.com/paper/2303.01486/) (2023)
- [Controlling Overestimation Bias with Truncated Mixture of Continuous Distributional Quantile Critics](https://mlsanity.com/paper/2005.04269/) (2020)
- [SUNRISE: A Simple Unified Framework for Ensemble Learning in Deep Reinforcement Learning](https://mlsanity.com/paper/2007.04938/) (2020)
- [Bigger, Better, Faster: Human-level Atari with human-level efficiency](https://mlsanity.com/paper/2305.19452/) (2023)
- [Implicit Under-Parameterization Inhibits Data-Efficient Deep Reinforcement Learning](https://mlsanity.com/paper/2010.14498/) (2020)
- [Taming the Noise in Reinforcement Learning via Soft Updates](https://mlsanity.com/paper/1512.08562/) (2015)
- [Revisiting the Minimalist Approach to Offline Reinforcement Learning](https://mlsanity.com/paper/2305.09836/) (2023)
- [Understanding and Preventing Capacity Loss in Reinforcement Learning](https://mlsanity.com/paper/2204.09560/) (2022)
- [Loss of Plasticity in Continual Deep Reinforcement Learning](https://mlsanity.com/paper/2303.07507/) (2023)
- [Generalization and Regularization in DQN](https://mlsanity.com/paper/1810.00123/) (2018)

## Built on by (most cited first, 17 of 17)

- [Normalization and effective learning rates in reinforcement learning](https://mlsanity.com/paper/2407.01800/) (2024)
- [RL Token: Bootstrapping Online RL with Vision-Language-Action Models](https://mlsanity.com/paper/2604.23073/) (2026)
- [Plasticity Loss in Deep Reinforcement Learning: A Survey](https://mlsanity.com/paper/2411.04832/) (2024)
- [XQC: Well-conditioned Optimization Accelerates Deep Reinforcement Learning](https://mlsanity.com/paper/2509.25174/) (2025)
- [MAD-TD: Model-Augmented Data stabilizes High Update Ratio RL](https://mlsanity.com/paper/2410.08896/) (2024)
- [Relative Entropy Pathwise Policy Optimization](https://mlsanity.com/paper/2507.11019/) (2025)
- [Average-Reward Soft Actor-Critic](https://mlsanity.com/paper/2501.09080/) (2025)
- [FlashSAC: Fast and Stable Off-Policy Reinforcement Learning for High-Dimensional Robot Control](https://mlsanity.com/paper/2604.04539/) (2026)
- [Stealing That Free Lunch: Exposing the Limits of Dyna-Style Reinforcement Learning](https://mlsanity.com/paper/2412.14312/) (2024)
- [What Does Flow Matching Bring To TD Learning?](https://mlsanity.com/paper/2603.04333/) (2026)
- [Scaling DRL for Decision Making: A Survey on Data, Network, and Training Budget Strategies](https://mlsanity.com/paper/2508.03194/) (2025)
- [A Tutorial: An Intuitive Explanation of Offline Reinforcement Learning Theory](https://mlsanity.com/paper/2508.07746/) (2025)
- [Symmetry-Aware Steering of Equivariant Diffusion Policies: Benefits and Limits](https://mlsanity.com/paper/2512.11345/) (2025)
- [Behavior-Consistent Deep Reinforcement Learning](https://mlsanity.com/paper/2605.21214/) (2026)
- [Distributional Value Estimation Without Target Networks for Robust Quality-Diversity](https://mlsanity.com/paper/2604.20381/) (2026)
- [SingularClip: Preventing Spectral Collapse to Maintain Plasticity in Continual and Reinforcement Learning](https://mlsanity.com/paper/2608.18319/) (2026)
- [Rank Collapse Is Recoverable, Growing |Q| Is Not: Out-of-Sample Early Warning for Value Divergence in High-UTD Soft Actor-Critic](https://mlsanity.com/paper/2609.32819/) (2026)

## Abstract (arXiv)

We show that deep reinforcement learning algorithms can retain their ability to learn without resetting network parameters in settings where the number of gradient updates greatly exceeds the number of environment samples by combatting value function divergence. Under large update-to-data ratios, a recent study by Nikishin et al. (2022) suggested the emergence of a primacy bias, in which agents overfit early interactions and downplay later experience, impairing their ability to learn. In this work, we investigate the phenomena leading to the primacy bias. We inspect the early stages of training that were conjectured to cause the failure to learn and find that one fundamental challenge is a long-standing acquaintance: value function divergence. Overinflated Q-values are found not only on out-of-distribution but also in-distribution data and can be linked to overestimation on unseen action prediction propelled by optimizer momentum. We employ a simple unit-ball normalization that enables learning under large update ratios, show its efficacy on the widely used dm_control suite, and obtain strong performance on the challenging dog tasks, competitive with model-based approaches. Our results question, in parts, the prior explanation for sub-optimal learning due to overfitting early data.