Papers for

ai product teams deploying multimodal models

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

Improving multimodal reasoning accuracy with token contrast and confidence

Enhancing Multimodal RLVR with Pairwise Token Contrast and Adaptive Trajectory Confidence

Abstract: Reinforcement learning with verifiable rewards (RLVR) improves multimodal reasoning, but group-relative policy optimization broadcasts a single sequence-level advantage to every token of a response, although errors usually localize to a few decisive steps. Recent self-distilled variants refine credit with a teacher--student likelihood gap. However, this signal cannot tell whether it relies on the image or on a language shortcut, and is applied uniformly even when the teacher is uninformative. To address this, we propose Pairwise token contrast and Adaptive trajectory ConfidencE (PACE). Pairwise token contrast scores each response under four teacher views, crossing a correct or incorrect answer hint with the image present or removed, and compares the two hints for every token, which cancels shifts shared by both hints and reveals whether the token's support relies on the image. Adaptive trajectory confidence measures how well the teacher predicts each response given the correct hint and the image, normalizes this score within the batch, and uses it to set how far the token contrast redistributes credit within that response. PACE achieves the best average accuracy on three backbones and ranks first on all evaluated multimodal reasoning benchmarks with Qwen3-VL-8B-Instruct. Experiments show that both components matter: replacing pairwise token contrast with a single-view signal underperforms our method, while adaptive trajectory confidence adds further improvement.

Thu 15 OctComputation and Language
The gist
Reinforcement learning methods that guide multimodal models (those understanding images and text) often give credit to entire responses even when mistakes happen only in small parts. The authors propose a new technique called PACE, which looks at each word in a response from different perspectives—whether the answer is right or wrong, and whether the image is shown or not—to better understand how image information supports the answer. They also adjust how much the teacher model's confidence affects learning. This approach helps models learn more precisely where errors occur and improves their accuracy on multimodal reasoning tasks.
Open → 2610.21671v1