AI Paper Atlas

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

All You Need is a Good Functional Prior for Bayesian Deep Learning

Ba-Hien Tran, Simone Rossi, Dimitrios Milios, Maurizio Filippone

NeurIPS 2022 · Probabilistic and scientific modeling › Uncertainty and Gaussian processes · approach: Variational Bayesian deep learning

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 a GP-induced hierarchical weight prior instead of a fixed hierarchical prior (ablation)Bayesian CNN · CIFAR10, LeNet5accuracy: 75.22 -> 76.51Improved
Use a GP-induced hierarchical weight prior instead of a fixed hierarchical prior (ablation)Bayesian CNN · CIFAR10, PreResNet20accuracy: 87.26 -> 88.20Improved
Use a GP-induced hierarchical weight prior instead of a fixed hierarchical prior (ablation)Bayesian CNN · CIFAR10, VGG16accuracy: 86.03 -> 87.03Improved
Use a GP-induced Gaussian weight prior instead of a fixed Gaussian prior (ablation)Bayesian CNN · CIFAR10, LeNet5accuracy: 74.65 -> 75.15Improved
Use a GP-induced Gaussian weight prior instead of a fixed Gaussian prior (ablation)Bayesian CNN · CIFAR10, PreResNet20accuracy: 85.34 -> 86.86Improved
Use a GP-induced Gaussian weight prior instead of a fixed Gaussian prior (ablation)Bayesian CNN · CIFAR10, VGG16accuracy: 81.47 -> 83.34Improved
Apply temperature scaling to the fixed Gaussian posteriorBayesian CNN · CIFAR10, LeNet5accuracy: 74.65 -> 74.08Hurt
Apply temperature scaling to the fixed Gaussian posteriorBayesian CNN · CIFAR10, PreResNet20accuracy: 85.34 -> 87.70Improved
Apply temperature scaling to the fixed Gaussian posteriorBayesian CNN · CIFAR10, VGG16accuracy: 81.47 -> 82.25Improved
Match the neural-network functional prior to a target GP prior rather than use standard parameter priorsBayesian neural networks · UCI regressiontest NLLImproved
Use a GP-induced hierarchical prior rather than a fixed or GP-induced Gaussian priorBayesian neural networks · UCI classificationtest NLLImproved
Use Wasserstein optimization rather than KL-based prior matchingBNN functional-prior matching · UCI datasetssquared MMDImproved

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