Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Deterministic Variational Inference for Robust Bayesian Neural Networks
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 |
|---|---|---|---|
| replace Monte Carlo gradient estimates with deterministic moment propagation | Bayesian neural networks · UCI regression datasets | test log-likelihood: Boston: -2.46±0.02 -> -2.41±0.02; Concrete: -3.07±0.01 -> -3.06±0.01; Energy: -1.03±0.04 -> -1.01±0.06; Kin8nm: 1.14±0.0 | Mixed |
| restrict propagated activation covariance to diagonal entries (ablation) | DVI · UCI regression datasets | test log-likelihood: Boston: -2.41±0.02 -> -2.42±0.02; Concrete: -3.06±0.01 -> -3.07±0.02; Energy: -1.01±0.06 -> -1.06±0.06; Kin8nm: 1.13±0.0 | Mixed |
| use a heteroscedastic rather than homoscedastic regression model (ablation) | DVI · UCI regression datasets | test log-likelihood: Boston: -2.58±0.04 -> -2.41±0.02; Concrete: -3.23±0.01 -> -3.06±0.01; Energy: -2.09±0.06 -> -1.01±0.06; Kin8nm: 1.01±0.0 | Helped |
| use data-adaptive empirical Bayes prior variances rather than one manually tuned shared variance (ablation) | DVI · UCI regression datasets | test log-likelihood | Helped |
Closest papers by meaning
- Variational Bayesian Last Layers (ICLR 2024 · similarity 0.91)
- Bayesian Online Natural Gradient (BONG) (NeurIPS 2024 · similarity 0.90)
- Training Bayesian Neural Networks with Sparse Subspace Variational Inference (ICLR 2024 · similarity 0.90)
- Reparameterized Importance Sampling for Robust Variational Bayesian Neural Networks (ICML 2024 · similarity 0.90)
- VIKING: Deep variational inference with stochastic projections (NeurIPS 2025 · similarity 0.89)
- Maxitive Donsker-Varadhan Formulation for Possibilistic Variational Inference (NeurIPS 2026 · similarity 0.89)
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.