DIAL：面向 LLM 智能体的方向感知自适应学习门控

View PDF HTML (experimental)

Abstract:Adaptive test-time compute for LLM agents aims to invoke extra computation only when it improves performance. Existing methods typically use confidence-, uncertainty-, or difficulty-based gates, assuming a fixed direction from the gating signal through compute need to the value of computation. This makes gating a utility-calibration problem: gating signals should align with whether extra computation improves the final outcome over the base policy. We show that this alignment is unstable: the same signal predicts rollout benefit in one setting and rollout harm in another, with reversals across environments and backbones even when the task is fixed. Wrong-direction gates can therefore worsen performance by precisely selecting harmful states. This reversal reflects a deeper distinction between compute need and compute suitability: a high uncertainty signal may indicate decision-difficult states where rollouts help compare alternatives, or intervention-unsuitable states where the current context does not support useful rollout-based improvement. Under this two-source model, fixed-direction gates are unreliable across heterogeneous settings. To address this, we propose DIAL (Direction-Informed Adaptive Learning), a sparse gate trained from signal-agnostic counterfactual exploration to learn the utility direction of state features per (environment, backbone). Across six environments and three backbones, DIAL yields a stronger overall success-cost trade-off than fixed-direction baselines. Our code is open-sourced at this https URL.

Comments:	Agent, Test time learing, Accpeted by NeuIPS 2026
Subjects:	Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
Cite as:	arXiv:2605.06908 [cs.LG]
	(or arXiv:2605.06908v2 [cs.LG] for this version)
	https://doi.org/10.48550/arXiv.2605.06908 arXiv-issued DOI via DataCite

Submission history

From: Ziming Li [view email]
[v1] Thu, 7 May 2026 20:15:51 UTC (522 KB)
[v2] Sun, 25 Oct 2026 16:16:38 UTC (539 KB)