Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
VICE: Variational Interpretable Concept Embeddings
Lukas Muttenthaler, Charles Yang Zheng, Patrick McClure, Robert A. Vandermeulen, Martin N Hebart, Francisco Pereira
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 VICE variational spike-and-slab embeddings instead of SPoSE | object embedding model · THINGS full training data | triplet accuracy: 0.637 -> 0.638 | No clear effect |
| use VICE instead of SPoSE | object embedding model · THINGS full training data | KL divergence: 0.103 -> 0.100 | No clear effect |
| use VICE instead of SPoSE | object embedding model · Adjectives test | triplet accuracy: 0.562 -> 0.559 | No clear effect |
| use VICE instead of SPoSE | object embedding model · Food test | triplet accuracy: 0.698 -> 0.693 | No clear effect |
| train on only 5% and 10% of THINGS data | VICE vs SPoSE · THINGS small subsets | accuracy and KL | Improved |
| use VICE instead of SPoSE on full THINGS | embedding model · THINGS | number of selected dimensions: 52 -> 44 | Improved |
| use VICE instead of SPoSE | embedding model · Food | number of selected dimensions: 16 -> 5 | Improved |
| use VICE instead of SPoSE | embedding model · Adjectives | number of selected dimensions: 22 -> 21 | No clear effect |
| use VICE instead of SPoSE | embedding model · THINGS | dimension reproducibility: 81.30 -> 87.01% | Improved |
| use VICE instead of SPoSE | embedding model · Food | dimension reproducibility: 62.88 -> 87.38% | Improved |
| use VICE instead of SPoSE | embedding model · Adjectives | dimension reproducibility: 71.64 -> 76.76% | Improved |
| use default hyperparameters instead of grid-searched best (ablation) | VICE · all three datasets | accuracy: within 0.015 | No clear effect |
Closest papers by meaning
- Projecting Assumptions: The Duality Between Sparse Autoencoders and Concept Geometry (NeurIPS 2025 · similarity 0.86)
- Nonparametric Identification of Latent Concepts (ICML 2025 · similarity 0.85)
- From Causal to Concept-Based Representation Learning (NeurIPS 2024 · similarity 0.85)
- Probabilistic Conceptual Explainers: Trustworthy Conceptual Explanations for Vision Foundation Models (ICML 2024 · similarity 0.85)
- Navigating Neural Space: Revisiting Concept Activation Vectors to Overcome Directional Divergence (ICLR 2025 · similarity 0.84)
- Learning Discrete Concepts in Latent Hierarchical Models (NeurIPS 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.
