Not All Queries Need Deep Thought: CoFiCot for Adaptive Coarse-to-fine Stateful Refinement

Dongxu Zhang, Hongqiang Lin, Yiding Sun, Pengyu Wang, Qirui Wang, Ning Yang, Jihua Zhu
Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence, PMLR 337:8115-8129, 2026.

Abstract

Scaling test-time computation enhances {LLM} reasoning ability but faces a uniform computation paradox. Allocating identical resources leads to over-correction on simple tasks and insufficient refinement on complex ones. To address this, we propose {CoFiCot}, a coarse-to-fine adaptive framework that dynamically tailors inference strategies to problem difficulty. Specifically, we implement a multi-metric classifier that triages queries by synthesizing semantic entropy, consensus reliability, and predicted reasoning depth . This enables a differentiated refinement stage that applies efficient aggregation for simple queries while routing complex ones to a context-aware correction loop . We formalize correction as a stateful sequential propagation process , where each repair is strictly conditioned on the verified history of prior rectifications. By integrating Process Reward Models (PRMs) within this state-dependent trajectory, {CoFiCot} effectively bridges the gap between granular error localization and global logical coherence, preventing the context fragmentation typical of stateless refinement methods.

Cite this Paper

BibTeX
@InProceedings{pmlr-v337-zhang26e, title = {Not All Queries Need Deep Thought: {CoFiCot} for Adaptive Coarse-to-fine Stateful Refinement}, author = {Zhang, Dongxu and Lin, Hongqiang and Sun, Yiding and Wang, Pengyu and Wang, Qirui and Yang, Ning and Zhu, Jihua}, booktitle = {Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence}, pages = {8115--8129}, year = {2026}, editor = {Perković, Emilija and Malinsky, Daniel}, volume = {337}, series = {Proceedings of Machine Learning Research}, month = {17--21 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v337/main/assets/zhang26e/zhang26e.pdf}, url = {https://proceedings.mlr.press/v337/zhang26e.html}, abstract = {Scaling test-time computation enhances {LLM} reasoning ability but faces a uniform computation paradox. Allocating identical resources leads to over-correction on simple tasks and insufficient refinement on complex ones. To address this, we propose {CoFiCot}, a coarse-to-fine adaptive framework that dynamically tailors inference strategies to problem difficulty. Specifically, we implement a multi-metric classifier that triages queries by synthesizing semantic entropy, consensus reliability, and predicted reasoning depth . This enables a differentiated refinement stage that applies efficient aggregation for simple queries while routing complex ones to a context-aware correction loop . We formalize correction as a stateful sequential propagation process , where each repair is strictly conditioned on the verified history of prior rectifications. By integrating Process Reward Models (PRMs) within this state-dependent trajectory, {CoFiCot} effectively bridges the gap between granular error localization and global logical coherence, preventing the context fragmentation typical of stateless refinement methods.} }
Endnote
%0 Conference Paper %T Not All Queries Need Deep Thought: CoFiCot for Adaptive Coarse-to-fine Stateful Refinement %A Dongxu Zhang %A Hongqiang Lin %A Yiding Sun %A Pengyu Wang %A Qirui Wang %A Ning Yang %A Jihua Zhu %B Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2026 %E Emilija Perković %E Daniel Malinsky %F pmlr-v337-zhang26e %I PMLR %P 8115--8129 %U https://proceedings.mlr.press/v337/zhang26e.html %V 337 %X Scaling test-time computation enhances {LLM} reasoning ability but faces a uniform computation paradox. Allocating identical resources leads to over-correction on simple tasks and insufficient refinement on complex ones. To address this, we propose {CoFiCot}, a coarse-to-fine adaptive framework that dynamically tailors inference strategies to problem difficulty. Specifically, we implement a multi-metric classifier that triages queries by synthesizing semantic entropy, consensus reliability, and predicted reasoning depth . This enables a differentiated refinement stage that applies efficient aggregation for simple queries while routing complex ones to a context-aware correction loop . We formalize correction as a stateful sequential propagation process , where each repair is strictly conditioned on the verified history of prior rectifications. By integrating Process Reward Models (PRMs) within this state-dependent trajectory, {CoFiCot} effectively bridges the gap between granular error localization and global logical coherence, preventing the context fragmentation typical of stateless refinement methods.
APA
Zhang, D., Lin, H., Sun, Y., Wang, P., Wang, Q., Yang, N. & Zhu, J.. (2026). Not All Queries Need Deep Thought: CoFiCot for Adaptive Coarse-to-fine Stateful Refinement. Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research 337:8115-8129 Available from https://proceedings.mlr.press/v337/zhang26e.html.

Related Material