Research map / Vision-language and multimodal / Visual question answering
Neural-Symbolic VQA: Disentangling Reasoning from Vision and Language Understanding
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| use symbolic program execution with a structural scene representation | NS-VQA · CLEVR validation | Overall accuracy: 99.1 -> 99.8 | Helped |
| pretrain with 90 rather than 270 annotated programs (ablation) | NS-VQA · CLEVR validation | Overall accuracy: 99.8 -> 74.4 | Hurt |
| pretrain with 180 rather than 270 annotated programs (ablation) | NS-VQA · CLEVR validation | Overall accuracy: 99.8 -> 89.5 | Hurt |
| fine-tune the attribute recognition network on split B images (ablation) | NS-VQA · CLEVR-CoGenT split B | Accuracy: 63.9 -> 98.9 | Helped |
| use grayscale input for shape recognition (ablation) | NS-VQA · CLEVR-CoGenT split B | Accuracy: 63.9 -> 98.4 | Helped |
| train the image parser on original CLEVR images (ablation) | NS-VQA · CLEVR-CoGenT split B | Accuracy: 63.9 -> 99.7 | Helped |
| use NS-VQA rather than IEP with 100 annotated programs | CLEVR-Humans · CLEVR-Humans | Accuracy: 38.7 -> 60.2 | Helped |
| use NS-VQA rather than IEP with 200 annotated programs | CLEVR-Humans · CLEVR-Humans | Accuracy: 40.1 -> 65.2 | Helped |
| use NS-VQA rather than IEP with 500 annotated programs | CLEVR-Humans · CLEVR-Humans | Accuracy: 49.2 -> 67.8 | Helped |
| use NS-VQA rather than IEP with 1K annotated programs | CLEVR-Humans · CLEVR-Humans | Accuracy: 63.4 -> 67.8 | Helped |
| use NS-VQA rather than IEP with 18K annotated programs | CLEVR-Humans · CLEVR-Humans | Accuracy: 66.6 -> 67.0 | No clear effect |
| use NS-VQA pretrained with 500 programs without adapting the scene parser (ablation) | NS-VQA · CLEVR-CoGenT split B | Accuracy: 99.8 -> 63.9 | Hurt |
Closest papers by meaning
- Neuro-Symbolic Visual Reasoning: Disentangling "Visual" from "Reasoning" (ICML 2020 · similarity 0.94)
- Probabilistic Neural Symbolic Models for Interpretable Visual Question Answering (ICML 2019 · similarity 0.92)
- GENOME: Generative Neuro-Symbolic Visual Reasoning by Growing and Reusing Modules (ICLR 2024 · similarity 0.90)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.89)
- VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning (CVPR 2024 · similarity 0.88)
- Synthesizing Visual Concepts as Vision-Language Programs (CVPR 2026 · similarity 0.88)
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.