Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Repulsive Deep Ensembles are Bayesian
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| add weight-space repulsive term (WGD) to deep ensembles | deep ensemble · FashionMNIST | accuracy: 91.122 -> 91.312 | Helped |
| add weight-space repulsive term (kde-WGD) | deep ensemble · CIFAR10 | accuracy: 85.552 -> 85.904 | Helped |
| move repulsion from weight space to function space (fWGD) (ablation) | repulsive ensemble · FashionMNIST | AUROC(H): 0.960 -> 0.971 | Helped |
| move repulsion from weight space to function space (fWGD) (ablation) | repulsive ensemble · CIFAR10 | AUROC(H): 0.838 -> 0.791 | Hurt |
| move repulsion from weight space to function space (fWGD) (ablation) | repulsive ensemble · CIFAR10 | accuracy: 85.904 -> 84.888 | Hurt |
| function-space repulsion (fWGD) in place of weight-space repulsion (ablation) | repulsive ensemble · FashionMNIST | NLL: 0.128 -> 0.125 | Helped |
| use model disagreement instead of entropy for OOD detection | ensembles · FashionMNIST | AUROC: 0.960 -> 0.970 | Helped |
| use L1 (Laplace) kernel instead of L2 in KDE (ablation) | WGD · experiments | performance | No clear effect |
| add repulsion to ensembles on CIFAR10 for entropy OOD detection | deep ensemble · CIFAR10 | AUROC(H): 0.843 -> 0.838 | Hurt |
Closest papers by meaning
- A Rigorous Link between Deep Ensembles and (Variational) Bayesian Methods (NeurIPS 2023 · similarity 0.88)
- Input-gradient space particle inference for neural network ensembles (ICLR 2024 · similarity 0.88)
- Agree to Disagree: Demystifying Homogeneous Deep Ensembles through Distributional Equivalence (ICLR 2025 · similarity 0.87)
- Sharpness-diversity tradeoff: improving flat ensembles with SharpBalance (NeurIPS 2024 · similarity 0.84)
- Structure-informed Risk Minimization for Robust Ensemble Learning (ICML 2025 · similarity 0.84)
- Enhancing Diversity in Bayesian Deep Learning via Hyperspherical Energy Minimization of CKA (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.