Learning Reasoning Rewards from Expert Demonstrations with Inverse Reinforcement Learning
Fuente:
arXiv
Guardado en:
| Autores principales: | , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866911691282841600 |
|---|---|
| author | Fanconi, Claudio Astorga, Nicolás van der Schaar, Mihaela |
| author_facet | Fanconi, Claudio Astorga, Nicolás van der Schaar, Mihaela |
| contents | Teaching large language models (LLMs) to reason during post-training typically relies on reinforcement learning with explicit outcome- or process-based reward functions. However, in many real-world settings, obtaining or defining such reward functions is difficult, especially for complex tasks, making learning from expert demonstrations an attractive alternative. The dominant approach, supervised fine-tuning (SFT), trains models to imitate expert reasoning traces directly, but suffers from the general limitations of off-policy learning: performance can be fragile to inference-time deviations from states explicitly covered by the demonstrations. To address this, we propose Reasoning Adversarial Inverse Reinforcement Learning (R-AIRL). Rather than imitating the expert's reasoning, R-AIRL infers the underlying process-level reward from the expert Chain-of-Thoughts. Through experiments on GSM8K, MMLU-Pro and MedReason we show that the reasoning reward function learned with R-AIRL can be effectively used throughout the training and inference pipeline: (1) to provide a training signal for post-training, outperforming SFT in most of the considered settings, (2) for inference-time reranking, improving pass@1 by up to 17.4 points, and (3) for process-level evaluation, localising reasoning failures with up to 86.1% accuracy. Overall, R-AIRL bridges imitation learning and reward-based optimisation, enabling the extraction of meaningful reasoning signals from expert thinking traces. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_01857 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Learning Reasoning Rewards from Expert Demonstrations with Inverse Reinforcement Learning Fanconi, Claudio Astorga, Nicolás van der Schaar, Mihaela Artificial Intelligence Teaching large language models (LLMs) to reason during post-training typically relies on reinforcement learning with explicit outcome- or process-based reward functions. However, in many real-world settings, obtaining or defining such reward functions is difficult, especially for complex tasks, making learning from expert demonstrations an attractive alternative. The dominant approach, supervised fine-tuning (SFT), trains models to imitate expert reasoning traces directly, but suffers from the general limitations of off-policy learning: performance can be fragile to inference-time deviations from states explicitly covered by the demonstrations. To address this, we propose Reasoning Adversarial Inverse Reinforcement Learning (R-AIRL). Rather than imitating the expert's reasoning, R-AIRL infers the underlying process-level reward from the expert Chain-of-Thoughts. Through experiments on GSM8K, MMLU-Pro and MedReason we show that the reasoning reward function learned with R-AIRL can be effectively used throughout the training and inference pipeline: (1) to provide a training signal for post-training, outperforming SFT in most of the considered settings, (2) for inference-time reranking, improving pass@1 by up to 17.4 points, and (3) for process-level evaluation, localising reasoning failures with up to 86.1% accuracy. Overall, R-AIRL bridges imitation learning and reward-based optimisation, enabling the extraction of meaningful reasoning signals from expert thinking traces. |
| title | Learning Reasoning Rewards from Expert Demonstrations with Inverse Reinforcement Learning |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2510.01857 |