Research map / Vision-language and multimodal / Human and embodied modeling
The Neuro-Symbolic Concept Learner: Interpreting Scenes, Words, and Sentences From Natural Supervision
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 object-based visual representations with symbolic reasoning instead of convolutional or attention-based baselines | CLEVR count-based visual concept evaluation · CLEVR synthetic concept-count question set | accuracy: 8% | Helped |
| Use the learned Color operator to acquire a new color concept | NS-CL · CLEVR incremental concept learning Split B validation | QA accuracy: 4.6% | Helped |
| Transfer learned visual concepts to a new image-caption DSL without fine-tuning the visual perception module | NS-CL · Synthetic image-caption retrieval full dataset | retrieval accuracy: 68.9 -> 97.0 | Helped |
| Remove ImageNet pretraining from the visual perception module (ablation) | NS-CL · CLEVR | QA accuracy: 0.5% | Hurt |
| Remove ImageNet pretraining from the visual perception module (ablation) | NS-CL · CLEVR | concept learning accuracy: 0.2% | Hurt |
| Remove curriculum over the number of objects in scenes (ablation) | NS-CL · CLEVR | visual perception accuracy | Hurt |
| Remove curriculum over program complexity (ablation) | NS-CL · CLEVR | training convergence | Hurt |
| Use object masks to guide attention (ablation) | TbD-Mask and MAC-Mask · CLEVR VQA | training speed | Helped |
| Use object-mask variants instead of the original TbD and MAC models (ablation) | TbD-Mask and MAC-Mask · CLEVR VQA | QA accuracy | No clear effect |
Closest papers by meaning
- NeuronEye: Query-Guided Visual Concept Activation for Vision-Language Reasoning (NeurIPS 2026 · similarity 0.89)
- GENOME: Generative Neuro-Symbolic Visual Reasoning by Growing and Reusing Modules (ICLR 2024 · similarity 0.88)
- Language-Informed Visual Concept Learning (ICLR 2024 · similarity 0.88)
- Concept-RuleNet: Grounded Multi-Agent Neurosymbolic Reasoning in Vision Language Models (AAAI 2026 · similarity 0.87)
- NePTune: A Neuro-Pythonic Framework for Tunable Compositional Reasoning on Vision-Language (ICLR 2026 · similarity 0.87)
- Causal Graphical Models for Vision-Language Compositional Understanding (ICLR 2025 · similarity 0.86)
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.