Sparse Masked Attention Policies for Reliable Generalization
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866911463073906688 |
|---|---|
| author | Horsch, Caroline Engwegen, Laurens Weltevrede, Max Spaan, Matthijs T. J. Böhmer, Wendelin |
| author_facet | Horsch, Caroline Engwegen, Laurens Weltevrede, Max Spaan, Matthijs T. J. Böhmer, Wendelin |
| contents | In reinforcement learning, abstraction methods that remove unnecessary information from the observation are commonly used to learn policies which generalize better to unseen tasks. However, these methods often overlook a crucial weakness: the function which extracts the reduced-information representation has unknown generalization ability in unseen observations. In this paper, we address this problem by presenting an information removal method which more reliably generalizes to new states. We accomplish this by using a learned masking function which operates on, and is integrated with, the attention weights within an attention-based policy network. We demonstrate that our method significantly improves policy generalization to unseen tasks in the Procgen benchmark compared to standard PPO and masking approaches. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_19956 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Sparse Masked Attention Policies for Reliable Generalization Horsch, Caroline Engwegen, Laurens Weltevrede, Max Spaan, Matthijs T. J. Böhmer, Wendelin Machine Learning In reinforcement learning, abstraction methods that remove unnecessary information from the observation are commonly used to learn policies which generalize better to unseen tasks. However, these methods often overlook a crucial weakness: the function which extracts the reduced-information representation has unknown generalization ability in unseen observations. In this paper, we address this problem by presenting an information removal method which more reliably generalizes to new states. We accomplish this by using a learned masking function which operates on, and is integrated with, the attention weights within an attention-based policy network. We demonstrate that our method significantly improves policy generalization to unseen tasks in the Procgen benchmark compared to standard PPO and masking approaches. |
| title | Sparse Masked Attention Policies for Reliable Generalization |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2602.19956 |