Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2603.18886 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910060027838464 |
|---|---|
| author | Aggarwal, Pranjal Ghazvininejad, Marjan Kim, Seungone Kulikov, Ilia Lanchantin, Jack Li, Xian Li, Tianjian Liu, Bo Neubig, Graham Ovalle, Anaelia Saha, Swarnadeep Sukhbaatar, Sainbayar Welleck, Sean Weston, Jason Whitehouse, Chenxi Williams, Adina Xu, Jing Yu, Ping Yuan, Weizhe Zhang, Jingyu Zhao, Wenting |
| author_facet | Aggarwal, Pranjal Ghazvininejad, Marjan Kim, Seungone Kulikov, Ilia Lanchantin, Jack Li, Xian Li, Tianjian Liu, Bo Neubig, Graham Ovalle, Anaelia Saha, Swarnadeep Sukhbaatar, Sainbayar Welleck, Sean Weston, Jason Whitehouse, Chenxi Williams, Adina Xu, Jing Yu, Ping Yuan, Weizhe Zhang, Jingyu Zhao, Wenting |
| contents | The ability to precisely derive mathematical objects is a core requirement for downstream STEM applications, including mathematics, physics, and chemistry, where reasoning must culminate in formally structured expressions. Yet, current LM evaluations of mathematical and scientific reasoning rely heavily on simplified answer formats such as numerical values or multiple choice options due to the convenience of automated assessment. In this paper we provide three contributions for improving reasoning over mathematical objects: (i) we build and release training data and benchmarks for deriving mathematical objects, the Principia suite; (ii) we provide training recipes with strong LLM-judges and verifiers, where we show that on-policy judge training boosts performance; (iii) we show how on-policy training can also be used to scale test-time compute via aggregation. We find that strong LMs such as Qwen3-235B and o3 struggle on Principia, while our training recipes can bring significant improvements over different LLM backbones, while simultaneously improving results on existing numerical and MCQA tasks, demonstrating cross-format generalization of reasoning abilities. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_18886 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Reasoning over mathematical objects: on-policy reward modeling and test time aggregation Aggarwal, Pranjal Ghazvininejad, Marjan Kim, Seungone Kulikov, Ilia Lanchantin, Jack Li, Xian Li, Tianjian Liu, Bo Neubig, Graham Ovalle, Anaelia Saha, Swarnadeep Sukhbaatar, Sainbayar Welleck, Sean Weston, Jason Whitehouse, Chenxi Williams, Adina Xu, Jing Yu, Ping Yuan, Weizhe Zhang, Jingyu Zhao, Wenting Artificial Intelligence Computation and Language The ability to precisely derive mathematical objects is a core requirement for downstream STEM applications, including mathematics, physics, and chemistry, where reasoning must culminate in formally structured expressions. Yet, current LM evaluations of mathematical and scientific reasoning rely heavily on simplified answer formats such as numerical values or multiple choice options due to the convenience of automated assessment. In this paper we provide three contributions for improving reasoning over mathematical objects: (i) we build and release training data and benchmarks for deriving mathematical objects, the Principia suite; (ii) we provide training recipes with strong LLM-judges and verifiers, where we show that on-policy judge training boosts performance; (iii) we show how on-policy training can also be used to scale test-time compute via aggregation. We find that strong LMs such as Qwen3-235B and o3 struggle on Principia, while our training recipes can bring significant improvements over different LLM backbones, while simultaneously improving results on existing numerical and MCQA tasks, demonstrating cross-format generalization of reasoning abilities. |
| title | Reasoning over mathematical objects: on-policy reward modeling and test time aggregation |
| topic | Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2603.18886 |