Correctness-Optimized Residual Activation Lens (CORAL): Transferable and Calibration-Aware Inference-Time Steering

Miranda Muqing Miao, Young Min Cho, Lyle Ungar
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:88596-88614, 2026.

Abstract

Large language models (LLMs) exhibit persistent miscalibration, especially after instruction tuning and preference alignment. Modified training objectives can improve calibration, but retraining is expensive. Inference-time steering offers a lightweight alternative, yet most existing methods optimize proxies for correctness rather than correctness itself. We introduce CORAL (Correctness-Optimized Residual Activation Lens), a regularized inference-time steering method that captures distributed correctness signals from model internal activations using weight-decay MLP probes. We evaluate CORAL across three 7B-parameter models and find that it consistently improves accuracy by 10% and expected calibration error (ECE) by 50% on average. We additionally demonstrate that these gains transfer without retraining to the complete published test sets of four held-out benchmarks (ARC-Challenge, HellaSwag, Math-MC, OpenBookQA), averaging 14% accuracy improvements and 49% ECE improvements. Our results support the hypothesis that distributed information in model internals can be extracted using regularized probes when individual neurons are insufficient. CORAL thus provides a compute-efficient, transferable, and calibration-aware approach to improve MCQA performance during inference.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-miao26a, title = {Correctness-Optimized Residual Activation Lens ({CORAL}): Transferable and Calibration-Aware Inference-Time Steering}, author = {Miao, Miranda Muqing and Cho, Young Min and Ungar, Lyle}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {88596--88614}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/miao26a/miao26a.pdf}, url = {https://proceedings.mlr.press/v306/miao26a.html}, abstract = {Large language models (LLMs) exhibit persistent miscalibration, especially after instruction tuning and preference alignment. Modified training objectives can improve calibration, but retraining is expensive. Inference-time steering offers a lightweight alternative, yet most existing methods optimize proxies for correctness rather than correctness itself. We introduce CORAL (Correctness-Optimized Residual Activation Lens), a regularized inference-time steering method that captures distributed correctness signals from model internal activations using weight-decay MLP probes. We evaluate CORAL across three 7B-parameter models and find that it consistently improves accuracy by 10% and expected calibration error (ECE) by 50% on average. We additionally demonstrate that these gains transfer without retraining to the complete published test sets of four held-out benchmarks (ARC-Challenge, HellaSwag, Math-MC, OpenBookQA), averaging 14% accuracy improvements and 49% ECE improvements. Our results support the hypothesis that distributed information in model internals can be extracted using regularized probes when individual neurons are insufficient. CORAL thus provides a compute-efficient, transferable, and calibration-aware approach to improve MCQA performance during inference.} }
Endnote
%0 Conference Paper %T Correctness-Optimized Residual Activation Lens (CORAL): Transferable and Calibration-Aware Inference-Time Steering %A Miranda Muqing Miao %A Young Min Cho %A Lyle Ungar %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-miao26a %I PMLR %P 88596--88614 %U https://proceedings.mlr.press/v306/miao26a.html %V 306 %X Large language models (LLMs) exhibit persistent miscalibration, especially after instruction tuning and preference alignment. Modified training objectives can improve calibration, but retraining is expensive. Inference-time steering offers a lightweight alternative, yet most existing methods optimize proxies for correctness rather than correctness itself. We introduce CORAL (Correctness-Optimized Residual Activation Lens), a regularized inference-time steering method that captures distributed correctness signals from model internal activations using weight-decay MLP probes. We evaluate CORAL across three 7B-parameter models and find that it consistently improves accuracy by 10% and expected calibration error (ECE) by 50% on average. We additionally demonstrate that these gains transfer without retraining to the complete published test sets of four held-out benchmarks (ARC-Challenge, HellaSwag, Math-MC, OpenBookQA), averaging 14% accuracy improvements and 49% ECE improvements. Our results support the hypothesis that distributed information in model internals can be extracted using regularized probes when individual neurons are insufficient. CORAL thus provides a compute-efficient, transferable, and calibration-aware approach to improve MCQA performance during inference.
APA
Miao, M.M., Cho, Y.M. & Ungar, L.. (2026). Correctness-Optimized Residual Activation Lens (CORAL): Transferable and Calibration-Aware Inference-Time Steering. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:88596-88614 Available from https://proceedings.mlr.press/v306/miao26a.html.

Related Material