Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Radial and Directional Posteriors for Bayesian Deep Learning
Changyong Oh, Kamil Adamczewski, Mijung Park
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 |
|---|---|---|---|
| add directional vMF component to radial-only posterior (ablation) | RDP · Wheel bandit | average cumulative regret: 3.6 -> 2.4 | Improved |
| use RDP posterior instead of VMG | VMG · UCI regression (10 datasets) | test log-likelihood: 6 out of 10 | Mixed |
| use RDP instead of VMG | VMG · UCI Power | test log-likelihood: -2.82±.01 -> -0.14±.01 | Improved |
| use RDP instead of VMG | VMG · UCI Yacht | test log-likelihood: -1.30±.02 -> -2.36±.04 | Hurt |
| use RDP instead of VMG | VMG · UCI Energy | test log-likelihood: -1.06±.03 -> -1.18±.03 | Hurt |
| use RDP instead of VMG | VMG · UCI Kin8nm | test log-likelihood: 1.10±.01 -> 2.17±.00 | Improved |
| prune VGG16 further with directional component after radial pruning (ablation) | RDP (R) · CIFAR10 VGG16 | test error: 8.7 -> 9.1 | Hurt |
| prune further with directional component (ablation) | RDP (R) · CIFAR10 VGG16 | FLOPs: 172M -> 170M | Improved |
| use RDP compression instead of BC-GHS | BC-GHS · MNIST LeNet-5 | FLOPs: 169K -> 125K | Improved |
| use RDP compression instead of BC-GHS | BC-GHS · MNIST LeNet-5 | parameters: 15K -> 20K | Hurt |
| use RDP instead of BC-GNJ | BC-GNJ · CIFAR10 VGG16 | test error: 8.6 -> 8.7 | No clear effect |
Closest papers by meaning
- Efficient Variational Inference for Sparse Deep Learning with Theoretical Guarantee (NeurIPS 2020 · similarity 0.91)
- Structured and Efficient Variational Deep Learning with Matrix Gaussian Posteriors (ICML 2016 · similarity 0.91)
- VIKING: Deep variational inference with stochastic projections (NeurIPS 2025 · similarity 0.90)
- Variational Learning of Fractional Posteriors (ICML 2025 · similarity 0.89)
- Bayesian Online Natural Gradient (BONG) (NeurIPS 2024 · similarity 0.88)
- Variational Bayesian Last Layers (ICLR 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.
