Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2601.20985 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910004263518208 |
|---|---|
| author | Akgül, Abdullah Baykal, Gulcin Haußmann, Manuel Çelikok, Mustafa Mert Kandemir, Melih |
| author_facet | Akgül, Abdullah Baykal, Gulcin Haußmann, Manuel Çelikok, Mustafa Mert Kandemir, Melih |
| contents | Optimal control of complex environments with robotic systems faces two complementary and intertwined challenges: efficient organization of sensory state information and far-sighted action planning. Because the reinforcement learning framework addresses only the latter, it tends to deliver sample-inefficient solutions. Active inference is the state-of-the-art process theory that explains how biological brains handle this dual problem. However, its applications to artificial intelligence have thus far been limited to extensions of existing model-based approaches. We present a formal abstraction of reinforcement learning algorithms that spans model-based, distributional, and model-free approaches. This abstraction seamlessly integrates active inference into the distributional reinforcement learning framework, making its performance advantages accessible without transition dynamics modeling. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_20985 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Distributional Active Inference Akgül, Abdullah Baykal, Gulcin Haußmann, Manuel Çelikok, Mustafa Mert Kandemir, Melih Machine Learning Optimal control of complex environments with robotic systems faces two complementary and intertwined challenges: efficient organization of sensory state information and far-sighted action planning. Because the reinforcement learning framework addresses only the latter, it tends to deliver sample-inefficient solutions. Active inference is the state-of-the-art process theory that explains how biological brains handle this dual problem. However, its applications to artificial intelligence have thus far been limited to extensions of existing model-based approaches. We present a formal abstraction of reinforcement learning algorithms that spans model-based, distributional, and model-free approaches. This abstraction seamlessly integrates active inference into the distributional reinforcement learning framework, making its performance advantages accessible without transition dynamics modeling. |
| title | Distributional Active Inference |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2601.20985 |