AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Grounding Multimodal Reasoning with Evidence-Ablated Negatives

chenxi liu, Tianyi Xiong, Yanshuo Chen, Heng Huang

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.

Proposes evidence-ablated negatives to ground RLVR-based multimodal reasoning in visual evidence, addressing credit assignment ambiguity in outcome supervision.

Problem

Outcome-based RLVR for multimodal reasoning can reward responses that exploit language priors or dataset shortcuts instead of grounding in visual evidence, creating credit assignment ambiguity.

Approach

Constructs high-quality negative responses by ablating the requisite visual evidence, so that reasoning that ignores the evidence is penalized during RL training. Details beyond the abstract are not available.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add evidence-ablated negatives to DAPODAPO · 9 multimodal reasoning benchmarks averageaverage accuracy: 50.3 -> 51.6 (3B); 56.5 -> 57.9 (7B); 71.9 -> 73.0 (Qwen3-VL-8B)Improved
use RankMix selection instead of global-only scaled-IIG selection (ablation)EVAN, Qwen2.5-VL-3B · 9 benchmarks averageaverage accuracy: 50.3 -> 51.6Improved
use RankMix instead of random selection (ablation)EVAN, Qwen2.5-VL-3B · 9 benchmarks averageaverage accuracy: 51.1 -> 51.6Improved
build negatives from evidence-ablated image instead of clean image (ablation)EVAN, Qwen2.5-VL-3B · 9 benchmarks averageaverage accuracy: 50.7 -> 51.6Improved
use scaled-IIG-guided patch masking instead of FGSM pixel perturbation (ablation)EVAN, Qwen2.5-VL-3B · 9 benchmarks averageaverage accuracy: 50.8 -> 51.6Improved
use scaled-IIG-guided masking instead of random patch masking at 50% (ablation)EVAN, Qwen2.5-VL-3B · 9 benchmarks averageaverage accuracy: 50.8 -> 51.6Improved
mask 50% of patches instead of 25%, 75% or 100% (ablation)EVAN, Qwen2.5-VL-3B · 9 benchmarks averageaverage accuracy: 51.1 / 51.4 / 50.9 -> 51.6Improved
select 20% of data instead of 30% (ablation)EVAN, Qwen2.5-VL-3B · 9 benchmarks averageaverage accuracy: 50.8 -> 51.6Improved
select 20% of data instead of 10% (ablation)EVAN, Qwen2.5-VL-3B · 9 benchmarks averageaverage accuracy: 51.5 -> 51.6No clear effect
add bounded negative advantage to evidence-ablated negatives (ablation)EVAN, Qwen2.5-VL-3B · training dynamicsentropy and gradient norm stabilityImproved
add clean-image likelihood and length filtering of negatives (ablation)EVAN, Qwen2.5-VL-7B · training dynamicsPPO KL and response lengthImproved
add evidence-ablated negatives to DAPOQwen2.5-VL-3B · ChartQA, GQA, Visual CounterFactaccuracy: Visual CounterFact pooled 81.61 -> 85.01; ChartQA 81.74 -> 82.55; GQA 54.46 -> 55.83Improved

Experimental setup

Builds on: RLVR

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