Return Capping: Sample-Efficient CVaR Policy Gradient Optimisation
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866908456547516416 |
|---|---|
| author | Mead, Harry Costen, Clarissa Lacerda, Bruno Hawes, Nick |
| author_facet | Mead, Harry Costen, Clarissa Lacerda, Bruno Hawes, Nick |
| contents | When optimising for conditional value at risk (CVaR) using policy gradients (PG), current methods rely on discarding a large proportion of trajectories, resulting in poor sample efficiency. We propose a reformulation of the CVaR optimisation problem by capping the total return of trajectories used in training, rather than simply discarding them, and show that this is equivalent to the original problem if the cap is set appropriately. We show, with empirical results in an number of environments, that this reformulation of the problem results in consistently improved performance compared to baselines. We have made all our code available here: https://github.com/HarryMJMead/cvar-return-capping. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_20887 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Return Capping: Sample-Efficient CVaR Policy Gradient Optimisation Mead, Harry Costen, Clarissa Lacerda, Bruno Hawes, Nick Machine Learning Artificial Intelligence When optimising for conditional value at risk (CVaR) using policy gradients (PG), current methods rely on discarding a large proportion of trajectories, resulting in poor sample efficiency. We propose a reformulation of the CVaR optimisation problem by capping the total return of trajectories used in training, rather than simply discarding them, and show that this is equivalent to the original problem if the cap is set appropriately. We show, with empirical results in an number of environments, that this reformulation of the problem results in consistently improved performance compared to baselines. We have made all our code available here: https://github.com/HarryMJMead/cvar-return-capping. |
| title | Return Capping: Sample-Efficient CVaR Policy Gradient Optimisation |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2504.20887 |