AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Learning When to Think: Dual-Reference Offline Optimization for Adaptive VLM Reasoning

Hongbin Lin, Sizhe Zou, Juangui Xu, Xinyue Xu, Zhuoli Ouyang, Stanislav Kolchin, Chengwei Qin, Zhongxiang Dai, Yao SHU

NeurIPS 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Reinforcement learning for visual reasoning · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Studies adaptive fast/slow thinking for vision-language models, where one model answers directly for simple inputs and reasons step by step only when needed, using offline dual-reference optimization.

Problem

Vision-language models forced to produce long reasoning traces for every input overthink and incur unnecessary inference cost; a single model should decide when to think.

Approach

Frames offline adaptive thinking as more than a length-compression problem and proposes dual-reference offline optimization to train a VLM to switch between direct answering and step-by-step reasoning. Only the abstract is available, so finer details are not stated.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add behavior-reference correction kappa (ablation)DR-BOLT · MathVision/DynaMath/MMMU-Proscore and slow ratio: slow ratio 0.13% -> 24.1%; MathVision 0.260 -> 0.606Improved
use exact instead of mode-wise correction (ablation)DR-BOLT · three transfer benchmarksscore: 0.606 -> 0.623 (MathVision)No clear effect
use dual-reference mixture instead of single-reference slow (ablation)BOLT · cross-benchmarkslow ratio: 97.4% -> 24.1%Improved
use DR-BOLT instead of mixture-reference DPO (ablation)DR-BOLT · MathVisionscore: 0.568 -> 0.606Improved
use DR-BOLT instead of mode-conditioned reward-weighted SFT (ablation)DR-BOLT · MathVisionscore: 0.513 -> 0.606Improved
use DR-BOLT instead of Reward SFT (ablation)DR-BOLT · MathVisionscore: 0.421 -> 0.606Improved
use offline DR-BOLT instead of online GRPODR-BOLT · MathVisionscore: 0.463 -> 0.606Improved
use offline DR-BOLT instead of online GRPODR-BOLT · training costtime (h): 202.4 -> 69.7Improved
change sampling fast-prob q from 0.7 to 0.3 (ablation)DR-BOLT · ScienceQAscore: 0.002 changeNo clear effect
increase length penalty lambda from 0 to 0.10 (ablation)DR-BOLT · ScienceQAresponse length (words): 28.9 -> 15.7Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy