AI Paper Atlas

Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes

On Divergence Measures for Bayesian Pseudocoresets

Balhae Kim, Jungwon Choi, Seanie Lee, Yoonho Lee, Jung-Woo Ha, Juho Lee

NeurIPS 2022 · Probabilistic and scientific modeling › Uncertainty and Gaussian processes · approach: Approximate inference in Bayesian neural networks · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
Use forward KL (BPC-fKL) instead of Wasserstein (BPC-W) pseudocoreset divergenceBayesian pseudocoreset · CIFAR10 10 ipc, HMCaccuracy: 0.3535 -> 0.4294Improved
Use forward KL instead of Wasserstein at 1 ipcBayesian pseudocoreset · CIFAR10 1 ipc, HMCaccuracy: 0.3435 -> 0.3354No clear effect
Use forward KL instead of reverse KLBayesian pseudocoreset · CIFAR10 20 ipc, HMCaccuracy: 0.3686 -> 0.4910Improved
Use pseudocoreset (BPC-fKL) instead of random coresetBayesian pseudocoreset · CIFAR10 10 ipc, HMCaccuracy: 0.2807 -> 0.4294Improved
Add test-time data augmentation with A-SGHMCBPC-fKL vs BPC-W · CIFAR10 20 ipcaccuracy: 0.4910 -> 0.5153 (fKL), 0.4546 -> 0.5691 (W)Mixed
Use BPC-fKL instead of BPC-W for memory and timeBayesian pseudocoreset · CIFAR10 computational costGPU memory and iteration timeImproved
Use BPC-fKL instead of BPC-W on corrupted inputsBayesian pseudocoreset · CIFAR10-Caccuracy and NLLImproved
Diagonal covariance Gaussian in BPC-W (ablation)BPC-W · CIFAR10 1 ipc, A-SGHMCaccuracy: 0.2934 -> 0.2959No clear effect
Diagonal covariance Gaussian in BPC-W (ablation)BPC-W · CIFAR10 10 ipc, A-SGHMCaccuracy: 0.4890 -> 0.4848No clear effect
Increase number of Gaussian samples S (ablation)BPC-fKL · CIFAR10 10 ipc, HMCaccuracyNo clear effect
Vary Gaussian variances (too small, 0.01, larger) (ablation)BPC-fKL · CIFAR10 10 ipc, HMCaccuracyMixed
Scale to larger pseudocoresets (50 and 100 ipc)BPC-fKL · CIFAR10, HMCaccuracy: 0.3922 -> 0.5557Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy