AI Paper Atlas

Research map / Vision-language and multimodal / Visual question answering

Compositional Neuro-Symbolic Concepts in Neural Activities

Yanchen Wang, Joy Hsu, Ehsan Adeli, Jiajun Wu

ICLR 2026 · Vision-language and multimodal › Visual question answering · approach: Visual reasoning diagnostics

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
use neuro-symbolic compositional grounding instead of the best prior decoderfMRI decoder · BOLD5000-QAaccuracy (overall): 0.4773 -> 0.7041Improved
use neuro-symbolic compositional grounding instead of the best prior decoderfMRI decoder · CNeuroMod-QAaccuracy (overall): 0.4642 -> 0.7046Improved
use NEURONA on unseen query compositionsfMRI decoder · BOLD5000-QA (unseen queries)accuracy (overall): 0.4754 -> 0.6840Improved
use NEURONA on unseen query compositionsfMRI decoder · CNeuroMod-QA (unseen queries)accuracy (overall): 0.4502 -> 0.6583Improved
add multi-region grounding without guidance (ablation)NEURONA single-region · BOLD5000-QAaccuracy (overall): 0.6451 -> 0.6476No clear effect
add multi-region grounding without guidance (ablation)NEURONA single-region · CNeuroMod-QAaccuracy (overall): 0.6429 -> 0.6162Hurt
add subject guidance to multi-region predicate grounding (ablation)NEURONA multi-region · BOLD5000-QAaccuracy (overall): 0.6476 -> 0.6678Improved
add object guidance to multi-region predicate grounding (ablation)NEURONA multi-region · CNeuroMod-QAaccuracy (overall): 0.6162 -> 0.6872Improved
add both subject and object guidance (ablation)NEURONA multi-region · BOLD5000-QAaccuracy (overall): 0.6476 -> 0.7102Improved
add both subject and object guidance (ablation)NEURONA multi-region · CNeuroMod-QAaccuracy (overall): 0.6162 -> 0.7189Improved
use Yeo-17 atlas instead of Yeo-7 (ablation)NEURONA · BOLD5000-QAaccuracy (overall): 0.6864 -> 0.7041Improved
use all Yeo-7 networks instead of one network subset (ablation)NEURONA · BOLD5000-QAaccuracy (overall): 0.6804 -> 0.7102Improved

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