AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Seeing Symbols, Missing Structure: A Real-World Handwritten Mathematical Expression Recognition Benchmark for Large Models

Sheng Jiang, Lin Zhu, Runrui Li, Mei Wang, Qiannan Zhu, Yaoyao Zhong, Hua Huang

ICML 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Abstract visual reasoning benchmarks

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
apply full SASR (MSR + SIL) at inferenceQwen3-VL-8B-Instruct · HMER-BenchExpRate: 11.63 -> 24.52Improved
apply SASRQwen3-VL-8B-Instruct · HMER-BenchExpRate@CDM: 25.84 -> 29.98Improved
add multi-stage structure-aware reasoning (MSR) (ablation)Qwen3-VL-8B-Instruct · HMER-BenchExpRate: 11.63 -> 19.52Improved
add schema-anchored in-context learning (SIL) alone (ablation)Qwen3-VL-8B-Instruct · HMER-BenchExpRate: 11.63 -> 14.02Improved
add SIL on top of MSR (ablation)Qwen3-VL-8B-Instruct · HMER-BenchExpRate: 19.52 -> 24.52Improved
add Step1 schema inference to MSR (ablation)Qwen3-VL-8B-Instruct · HMER-BenchExpRate: 10.28 -> 19.52Improved
add Step3 symbol disambiguation to MSR (ablation)Qwen3-VL-8B-Instruct · HMER-Bench D3ExpRate: 5.27 -> 7.39Improved
add Step3 symbol disambiguation to MSR (ablation)Qwen3-VL-8B-Instruct · HMER-BenchExpRate: 18.84 -> 19.52Improved
add Simple CoT prompting (ablation)Qwen3-VL-8B-Instruct · HMER-BenchExpRate: 10.99 -> 11.36No clear effect
apply SASR on Ovis2.5-8BOvis2.5-8B · HMER-BenchExpRate: 7.72 -> 12.55Improved
apply SASR on Nemotron-Nano-12B-v2-VLNemotron-Nano-12B-v2-VL · HMER-BenchExpRate: 6.01 -> 8.91Improved
apply SASR on Kimi-K2.5Kimi-K2.5 · HMER-BenchExpRate: 28.06 -> 38.90Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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