Research map / Probabilistic and scientific modeling / Molecules and proteins
Mixture of Concept Bottleneck Experts
Francesco De Santis, Gabriele Ciravegna, Giovanni De Felice, Arianna Casanova, Francesco Giannini, Michelangelo Diligenti, Johannes Schneider, Danilo Giordano, Mateo Espinosa Zarlenga, Pietro Barbiero
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 Symbolic M-CBE instead of Lin-M-CBE or MLP-M-CBE for recovering the true expression | M-CBE · Pendulum, MNIST-Arith., MAWPS | Tree Edit Distance: Pendulum 5.71 -> 0.00; MAWPS 15.00 -> 0.00 | Improved |
| use Boolean-expression predictors (CMR, DCR) when the concept set is large (ablation) | concept bottleneck predictor · CUB200 and CIFAR10 (more than 100 concepts) | error rate: near-100% error | Hurt |
| use linear or symbolic experts instead of Boolean experts in high-dimensional concept spaces | M-CBE · high-dimensional concept datasets | accuracy: up to +65% | Improved |
| increase the number of experts above one (ablation) | M-CBE · AWA2-Incomplete, CUB200-Incomplete, MAWPS | error rate / MAE | Improved |
| replace black-box MLP task predictor with structured Lin-M-CBE or Sym-M-CBE (ablation) | M-CBE · classification and regression benchmarks | accuracy vs complexity | Improved |
| expand symbolic operator vocabulary from {+,-} to the complete set (ablation) | Sym-M-CBE · Pendulum | MAE: 3.80 -> 0.46 | Improved |
| expand symbolic operator vocabulary from {+,-} to {+,-,x} (ablation) | Sym-M-CBE · MAWPS | MAE: 2.73 -> 1.29 | Improved |
| expand symbolic operator vocabulary from medium to complete (ablation) | Sym-M-CBE · MAWPS | MAE: 1.29 -> 1.36 | No clear effect |
| use a finite set of experts instead of per-sample expression trees (LICEM) (ablation) | Lin-M-CBE · CUB-200 | global interpretability | Improved |
Closest papers by meaning
- Interpretable but Fragile? Robustness of Concept Bottlenecks under Geometric-Semantic Perturbations (NeurIPS 2026 · similarity 0.89)
- An Analysis of Concept Bottleneck Models: Measuring, Understanding, and Mitigating the Impact of Noisy Annotations (NeurIPS 2025 · similarity 0.89)
- Measuring What Matters: Synthetic Benchmarks for Concept Bottleneck Models (NeurIPS 2026 · similarity 0.89)
- Stochastic Concept Bottleneck Models (NeurIPS 2024 · similarity 0.89)
- Editable Concept Bottleneck Models (ICML 2025 · similarity 0.88)
- Interpretable Concept-Based Memory Reasoning (NeurIPS 2024 · 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.
