Batch-of-Thought: Cross-Instance Learning for Enhanced LLM Reasoning
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866914546982060032 |
|---|---|
| author | Yang, Xuan Jia, Furong Xie, Roy Xi, Xiong Bian, Hengwei Li, Jian Agrawal, Monica |
| author_facet | Yang, Xuan Jia, Furong Xie, Roy Xi, Xiong Bian, Hengwei Li, Jian Agrawal, Monica |
| contents | Current Large Language Model reasoning systems process queries independently, discarding valuable cross-instance signals such as shared reasoning patterns and consistency constraints. We introduce Batch-of-Thought (BoT), a training-free method that processes related queries jointly to enable cross-instance learning. By performing comparative analysis across batches, BoT identifies high-quality reasoning templates, detects errors through consistency checks, and amortizes computational costs. We instantiate BoT within a multi-agent reflection architecture (BoT-R), where a Reflector performs joint evaluation to unlock mutual information gain unavailable in isolated processing. Experiments across three model families and six benchmarks demonstrate that BoT-R consistently improves accuracy and confidence calibration while reducing inference costs by up to 61%. Our theoretical and experimental analysis reveals when and why batch-aware reasoning benefits LLM systems. Our code is available at https://github.com/xuanyang19/BoT |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_02950 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Batch-of-Thought: Cross-Instance Learning for Enhanced LLM Reasoning Yang, Xuan Jia, Furong Xie, Roy Xi, Xiong Bian, Hengwei Li, Jian Agrawal, Monica Artificial Intelligence Current Large Language Model reasoning systems process queries independently, discarding valuable cross-instance signals such as shared reasoning patterns and consistency constraints. We introduce Batch-of-Thought (BoT), a training-free method that processes related queries jointly to enable cross-instance learning. By performing comparative analysis across batches, BoT identifies high-quality reasoning templates, detects errors through consistency checks, and amortizes computational costs. We instantiate BoT within a multi-agent reflection architecture (BoT-R), where a Reflector performs joint evaluation to unlock mutual information gain unavailable in isolated processing. Experiments across three model families and six benchmarks demonstrate that BoT-R consistently improves accuracy and confidence calibration while reducing inference costs by up to 61%. Our theoretical and experimental analysis reveals when and why batch-aware reasoning benefits LLM systems. Our code is available at https://github.com/xuanyang19/BoT |
| title | Batch-of-Thought: Cross-Instance Learning for Enhanced LLM Reasoning |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2601.02950 |