AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Enhancing Numerical Prediction of MLLMs with Soft Labeling

Pei Wang, Zhaowei Cai, Hao Yang, Davide Modolo, Ashwin Swaminathan

ICCV 2025 · Large language models › Language, safety and interpretability · approach: Statistical frameworks for LLM evaluation · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace one-hot hard labels with soft labels on digit tokensLLaVA-7B specialist · RefCOCOg valAcc@0.5: 79.0 -> 82.0Improved
replace one-hot hard labels with soft labels on digit tokensLLaVA-7B generalist · RefCOCOg valAcc@0.5: 72.1 -> 78.5Improved
soft labeling on stronger 13B modelLLaVA-13B specialist · RefCOCOg valAcc@0.5: 86.0 -> 86.7Improved
soft labeling for chart numeric answersLLaVA-13B generalist · ChartQA numerical subsetaccuracy: 69.7 -> 72.6Improved
soft labeling gain larger on numerical subset than full ChartQA setLLaVA-7B specialist · ChartQAaccuracy: +1.7 (all) vs +2.3 (numerical)Improved
soft labeling for counting on balanced testLLaVA-7B specialist · TallyQA simple balancedRMSE: 3.84 -> 3.20Improved
soft labeling on standard unbalanced TallyQA simpleLLaVA-7B specialist · TallyQA simpleAcc.: 78.5 -> 79.0Improved
use triangular/binomial instead of uniform soft distribution (ablation)soft labeling · ChartQA numericalaccuracy: 66.8 -> 68.6Improved
uniform distribution (label smoothing) vs hard labeling (ablation)LLaVA-7B specialist · ChartQA numericalaccuracy: 66.3 -> 66.8Improved
set eta too small or too large (ablation)soft labeling · RefCOCOg val, ChartQA, TallyQAaccuracy/RMSEHurt
increase lambda beyond 5 (ablation)soft labeling · RefCOCOg val, ChartQA numerical, TallyQAaccuracy/RMSEHurt
soft labeling effect on generic taskssoft labeling · generic tasks (Supp.)performanceNo clear effect

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