Research map / Vision-language and multimodal / Human and embodied modeling
FALCON: Fast Visual Concept Learning by Integrating Images, Linguistic descriptions, and Conceptual Relations
Lingjie Mei, Jiayuan Mao, Ziqi Wang, Chuang Gan, Joshua B. Tenenbaum
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Use FALCON-G instead of NSCL+LSTM | CUB fast concept learning · CUB | QA accuracy: 79.53 -> 81.33 | Improved |
| Use box embeddings instead of hyperplane embeddings (ablation) | FALCON-G · CUB | QA accuracy: 74.11 -> 81.33 | Improved |
| Use box embeddings instead of hypercone embeddings (ablation) | FALCON-G · CUB | QA accuracy: 68.44 -> 81.33 | Improved |
| Add parsed concept graphs and GNN processing for supplemental sentences (ablation) | NSCL+LSTM on CUB · CUB | QA accuracy: 79.53 -> 78.50 | Improved |
| Replace averaged visual examples with an example-graph GNN (ablation) | NSCL+GNN on CUB · CUB | QA accuracy: 78.50 -> 81.33 | Improved |
| Train with fewer base concepts (ablation) | FALCON-G · CUB | QA accuracy: 81.33% -> 76.32% | Hurt |
| Add supplemental sentences containing related concepts (ablation) | FALCON-G · CUB | QA accuracy: 76.37% -> 81.33% | Improved |
| Use FALCON-G instead of NSCL+GNN | CLEVR with supplemental descriptions · CLEVR | QA accuracy: 84.11 -> 88.40 | Improved |
| Provide supplemental sentences rather than examples alone (ablation) | FALCON-G · CLEVR | QA accuracy: 68.47 -> 88.40 | Improved |
| Provide supplemental sentences rather than examples alone (ablation) | CNN+LSTM · CLEVR | QA accuracy: 41.92 -> 42.69 | No clear effect |
| Use FALCON-G instead of NSCL+GNN | CLEVR biased-data learning · CLEVR biased data | accuracy: 84.83% -> 88.29% | Improved |
| Use FALCON instead of MAC | GQA with supplemental sentences · GQA | accuracy: 54.99% -> 55.96% | Improved |
Closest papers by meaning
- Language-Informed Visual Concept Learning (ICLR 2024 · similarity 0.91)
- Bridging the gap to real-world language-grounded visual concept learning (NeurIPS 2025 · similarity 0.90)
- ConceptBed: Evaluating Concept Learning Abilities of Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.88)
- Context-Aware Meta-Learning (ICLR 2024 · similarity 0.87)
- An Image is Worth Multiple Words: Discovering Object Level Concepts using Multi-Concept Prompt Learning (ICML 2024 · similarity 0.87)
- Learning to Infer Generative Template Programs for Visual Concepts (ICML 2024 · similarity 0.87)
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.
