Generalized Variational Inference in Function Spaces: Gaussian Measures meet Bayesian Deep Learning

Authors: Veit David Wild, Robert Hu, Dino Sejdinovic

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The proposed method obtains state-of-the-art performance on several benchmark datasets.
Researcher Affiliation Collaboration Veit D. Wild Department of Statistics University of Oxford veit.wild@stats.ox.ac.uk Robert Hu Amazon robyhu@amazon.co.uk Dino Sejdinovic School of Computer and Mathematical Sciences University of Adelaide dino.sejdinovic@adelaide.edu.au
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes Codebase: https://github.com/Mr Huff/GWI
Open Datasets Yes UCI Regression... Fashion MNIST [Xiao et al., 2017] and CIFAR-10 [Krizhevsky et al., 2009]
Dataset Splits No The paper mentions 'We train on random 90% of the data and predict on 10%' for training and testing, but does not explicitly provide details for a separate validation split in the main text.
Hardware Specification Yes All experiments were performed on a single NVIDIA GeForce RTX 3090 GPU with 24GB of memory.
Software Dependencies No The paper mentions using 'deepobs library [Schneider et al., 2019]' but does not provide specific version numbers for software dependencies.
Experiment Setup Yes For the UCI experiments, we used a single hidden layer MLP with 50 units and ReLU activations. We used a batch size of 128 and trained for 200 epochs using the Adam optimizer with a learning rate of 0.001.