Position: Lifetime tuning is incompatible with continual reinforcement learning

Golnaz Mesbahi, Parham Mohammad Panahi, Olya Mastikhina, Steven Tang, Martha White, Adam White
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:81786-81796, 2025.

Abstract

In continual RL we want agents capable of never-ending learning, and yet our evaluation methodologies do not reflect this. The standard practice in RL is to assume unfettered access to the deployment environment for the full lifetime of the agent. For example, agent designers select the best performing hyperparameters in Atari by testing each for 200 million frames and then reporting results on 200 million frames. In this position paper, we argue and demonstrate the pitfalls of this inappropriate empirical methodology: lifetime tuning. We provide empirical evidence to support our position by testing DQN and SAC across several of continuing and non-stationary environments with two main findings: (1) lifetime tuning does not allow us to identify algorithms that work well for continual learning—all algorithms equally succeed; (2) recently developed continual RL algorithms outperform standard non-continual algorithms when tuning is limited to a fraction of the agent’s lifetime. The goal of this paper is to provide an explanation for why recent progress in continual RL has been mixed and motivate the development of empirical practices that better match the goals of continual RL.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-mesbahi25a, title = {Position: Lifetime tuning is incompatible with continual reinforcement learning}, author = {Mesbahi, Golnaz and Mohammad Panahi, Parham and Mastikhina, Olya and Tang, Steven and White, Martha and White, Adam}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {81786--81796}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/mesbahi25a/mesbahi25a.pdf}, url = {https://proceedings.mlr.press/v267/mesbahi25a.html}, abstract = {In continual RL we want agents capable of never-ending learning, and yet our evaluation methodologies do not reflect this. The standard practice in RL is to assume unfettered access to the deployment environment for the full lifetime of the agent. For example, agent designers select the best performing hyperparameters in Atari by testing each for 200 million frames and then reporting results on 200 million frames. In this position paper, we argue and demonstrate the pitfalls of this inappropriate empirical methodology: lifetime tuning. We provide empirical evidence to support our position by testing DQN and SAC across several of continuing and non-stationary environments with two main findings: (1) lifetime tuning does not allow us to identify algorithms that work well for continual learning—all algorithms equally succeed; (2) recently developed continual RL algorithms outperform standard non-continual algorithms when tuning is limited to a fraction of the agent’s lifetime. The goal of this paper is to provide an explanation for why recent progress in continual RL has been mixed and motivate the development of empirical practices that better match the goals of continual RL.} }
Endnote
%0 Conference Paper %T Position: Lifetime tuning is incompatible with continual reinforcement learning %A Golnaz Mesbahi %A Parham Mohammad Panahi %A Olya Mastikhina %A Steven Tang %A Martha White %A Adam White %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-mesbahi25a %I PMLR %P 81786--81796 %U https://proceedings.mlr.press/v267/mesbahi25a.html %V 267 %X In continual RL we want agents capable of never-ending learning, and yet our evaluation methodologies do not reflect this. The standard practice in RL is to assume unfettered access to the deployment environment for the full lifetime of the agent. For example, agent designers select the best performing hyperparameters in Atari by testing each for 200 million frames and then reporting results on 200 million frames. In this position paper, we argue and demonstrate the pitfalls of this inappropriate empirical methodology: lifetime tuning. We provide empirical evidence to support our position by testing DQN and SAC across several of continuing and non-stationary environments with two main findings: (1) lifetime tuning does not allow us to identify algorithms that work well for continual learning—all algorithms equally succeed; (2) recently developed continual RL algorithms outperform standard non-continual algorithms when tuning is limited to a fraction of the agent’s lifetime. The goal of this paper is to provide an explanation for why recent progress in continual RL has been mixed and motivate the development of empirical practices that better match the goals of continual RL.
APA
Mesbahi, G., Mohammad Panahi, P., Mastikhina, O., Tang, S., White, M. & White, A.. (2025). Position: Lifetime tuning is incompatible with continual reinforcement learning. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:81786-81796 Available from https://proceedings.mlr.press/v267/mesbahi25a.html.

Related Material