AI Paper Atlas

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

Toward Structural Multimodal Representations: Specialization, Selection, and Sparsification via Mixture-of-Experts

Hahyeon Choi, Nojun Kwak

ICML 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Language-grounded visual concept learning · 0 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
add router-based Selection after Specialization (ablation)S3 (granularity 8) · MOSEIlinear probing accuracy: 75.78 -> 77.36Improved
add router-based Selection after Specialization (ablation)S3 (granularity 2) · MOSEIlinear probing accuracy: 74.94 -> 74.48Hurt
add Sparsification (pruning) after Selection (ablation)S3 · MOSEIlinear probing accuracy: 77.36 -> 77.95Improved
add Sparsification after Selection (ablation)S3 · MOSIlinear probing accuracy: 64.29 -> 66.13Improved
add Sparsification after Selection (ablation)S3 · UR-FUNNYlinear probing accuracy: 63.52 -> 64.87Improved
add Sparsification after Selection (ablation)S3 · MUStARDlinear probing accuracy: 61.59 -> 62.56Improved
add Selection after Specialization (ablation)S3 · UR-FUNNYlinear probing accuracy: 63.61 -> 63.52No clear effect
add Selection after Specialization (ablation)S3 · MUStARDlinear probing accuracy: 58.70 -> 61.59Improved
increase expert granularity from 2 to 8 (Specialization) (ablation)S3 · MOSEIlinear probing accuracy: 74.94 -> 75.78Improved
increase expert granularity from 2 to 8 (best sparsification) (ablation)S3 · MOSIlinear probing accuracy: 64.82 -> 66.13Improved
use S3 instead of DisentangledSSLS3 full · MOSEIlinear probing accuracy: 77.45 -> 77.95Improved
use S3 instead of DisentangledSSLS3 full · MUStARDlinear probing accuracy: 61.60 -> 62.56Improved

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