Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Global inducing point variational posteriors for Bayesian neural networks and deep Gaussian processes
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 |
|---|---|---|---|
| Use global inducing points instead of factorised inference | Bayesian neural network · CIFAR-10 without tempering or data augmentation | accuracy (%): 80.27 -> 86.70 | Helped |
| Use global inducing points instead of local inducing points | Bayesian neural network · CIFAR-10 without tempering or data augmentation | ELBO: -0.84 -> -0.68 | Helped |
| Use global inducing points at all layers instead of factorised inference with global inducing points only at the last layer | Bayesian neural network · CIFAR-10 without tempering or data augmentation | accuracy (%): 83.33 -> 86.70 | Helped |
| Use fac → gi instead of factorised inference | ResNet-18 Bayesian neural network with tempering and data augmentation · CIFAR-10 | accuracy (%): 87.52 -> 92.41 | Helped |
| Use a global inducing posterior rather than a layer-factorised posterior | Bayesian neural network · 1D regression toy dataset | predictive uncertainty quality | Helped |
| Use global inducing points at the last layer rather than factorised or local inducing inference | Deep linear Bayesian neural network · Toy linear-model data across network depths and widths | ELBO | Helped |
| Use global inducing inference instead of factorised inference or DSVI | Bayesian neural network and deep Gaussian process · UCI regression datasets | ELBO | Helped |
| Use global inducing inference instead of DSVI | Deep Gaussian process · Some UCI regression datasets | test log likelihood | Mixed |
Closest papers by meaning
- Structured and Efficient Variational Deep Learning with Matrix Gaussian Posteriors (ICML 2016 · similarity 0.91)
- On the Expressiveness of Approximate Inference in Bayesian Neural Networks (NeurIPS 2020 · similarity 0.91)
- Sparse Inducing Points in Deep Gaussian Processes: Enhancing Modeling with Denoising Diffusion Variational Inference (ICML 2024 · similarity 0.90)
- VIKING: Deep variational inference with stochastic projections (NeurIPS 2025 · similarity 0.88)
- Bayesian Basis Function Approximation for Scalable Gaussian Process Priors in Deep Generative Models (ICML 2025 · similarity 0.87)
- Bayesian Optimization via Continual Variational Last Layer Training (ICLR 2025 · similarity 0.87)
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.