Research map / Vision-language and multimodal / Human and embodied modeling
Cognitive Deficit of Deep Learning in Numerosity
Xiaolin Wu, Xi Zhang, Xiao Shu
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 |
|---|---|---|---|
| Test on triangles rather than training-set circles | Subitizing DCNN · Subitizing, triangles | accuracy: 0.45 | Hurt |
| Test on squares rather than training-set circles | Subitizing DCNN · Subitizing, squares | accuracy: 0.76 | Hurt |
| Swap foreground and background colors | Subitizing DCNN · Subitizing, swapped colors | accuracy: 0.42 | Hurt |
| Use normalized boundary representations instead of raw image pixels | Subitizing DCNN · Subitizing across varied shapes, colors, and region-boundary representations | classification accuracy: 1; 0.790; 0.920 | Improved |
| Scale objects up 50% | Human-guided DCNN · Subitizing, scaled-up objects | probability of judging 3 as 3: 0.496 | Hurt |
| Scale objects down 50% | Human-guided DCNN · Subitizing, scaled-down objects | probability of judging 3 as 3: 0.021 | Hurt |
| Add residual blocks to the subitizing network (ablation) | Subitizing DCNN · Subitizing generalization | generalization | No clear effect |
| Use morphology-based connectivity-preserving pruning | Deterministic RCNN · Subitizing across object size and shape | generalization | Improved |
| Train the deterministic RCNN from random weight initialization (ablation) | Deterministic RCNN · Subitizing | training convergence | Hurt |
| Initialize weights slightly perturbed from the designed kernels (ablation) | Deterministic RCNN · Subitizing | training convergence | Hurt |
| Decompose subitizing into erosion, connected components, and counting subnetworks | Subitizing DCNN · Subitizing across object sizes and shapes | generalization | Improved |
Closest papers by meaning
- Alignment of CNN and Human Judgments of Geometric and Topological Concepts (AAAI 2025 · similarity 0.87)
- Unveiling the Visual Counting Bottleneck in Vision-Language Models (ICML 2026 · similarity 0.85)
- Core Knowledge Deficits in Multi-Modal Language Models (ICML 2025 · similarity 0.84)
- Geometry of naturalistic object representations in recurrent neural network models of working memory (NeurIPS 2024 · similarity 0.84)
- Caption This, Reason That: VLMs Caught in the Middle (NeurIPS 2025 · similarity 0.84)
- CogFlow: Bridging Perception and Reasoning through Knowledge Internalization for Visual Mathematical Problem Solving (ICLR 2026 · similarity 0.84)
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.
