Research map / Label-efficient and robust learning / Recognition and long-tail
GlanceNets: Interpretable, Leak-proof Concept-based Models
Emanuele Marconato, Andrea Passerini, Stefano Teso
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 GlanceNet (disentangled aligned concepts) instead of CBNM | CBNM · MNIST even vs odd | leakage: 28% -> 2% | Improved |
| remove open-set recognition from GlanceNet (CG-VAE) (ablation) | GlanceNet · MNIST even vs odd | leakage: 2% -> 30% | Hurt |
| use GlanceNet instead of CBNM | CBNM · dSprites position leakage | leakage: 80% -> negligible | Improved |
| remove open-set recognition from GlanceNet (CG-VAE) (ablation) | GlanceNet · dSprites position leakage | leakage: negligible -> 98% | Hurt |
| use GlanceNet instead of CBNM | CBNM · dSprites | alignment (DCI): up to 15% | Improved |
| use GlanceNet instead of CBNM | CBNM · MPI3D | alignment (DCI): minimum of 8% | Improved |
| use GlanceNet instead of CBNM at full supervision | CBNM · CelebA | alignment (DCI): almost 8% | Improved |
| use GlanceNet instead of CBNM at 1% or 10% supervision | CBNM · CelebA | alignment (DCI) | No clear effect |
| use GlanceNet instead of CBNM | CBNM · dSprites, MPI3D, CelebA | accuracy | No clear effect |
| use GlanceNet instead of CBNM | CBNM · CelebA | concept accuracy | No clear effect |
| extend latent space to 20 (beta-VAE) or 40 (TCVAE) dimensions (ablation) | GlanceNet (10 latent dims) · CelebA | alignment and explicitness | Improved |
| replace mixture prior with Gaussian prior (supervised VAE) (ablation) | GlanceNet · MNIST leakage | leakage | Hurt |
Closest papers by meaning
- Enhancing Performance of Explainable AI Models with Constrained Concept Refinement (ICML 2025 · similarity 0.86)
- Restyling Unsupervised Concept Based Interpretable Networks with Generative Models (ICLR 2025 · similarity 0.86)
- Walking the Web of Concept-Class Relationships in Incrementally Trained Interpretable Models (AAAI 2025 · similarity 0.85)
- FaCT: Faithful Concept Traces for Explaining Neural Network Decisions (NeurIPS 2025 · similarity 0.85)
- Concept Bottleneck Large Language Models (ICLR 2025 · similarity 0.84)
- Understanding Inter-Concept Relationships in Concept-Based Models (ICML 2024 · 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.
