Posterior and Computational Uncertainty in Gaussian Processes

Authors: Jonathan Wenger, Geoff Pleiss, Marvin Pförtner, Philipp Hennig, John P. Cunningham

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

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we empirically demonstrate the consequences of ignoring computational uncertainty and show how implicitly modeling it improves generalization performance on benchmark datasets.
Researcher Affiliation Academia 1 University of Tübingen 2 Columbia University 3 Max Planck Institute for Intelligent Systems, Tübingen
Pseudocode Yes Algorithm 1: A Class of Computation-Aware Iterative Methods for GP Approximation
Open Source Code Yes An implementation of Algorithm 1, based on Ke Ops [48] and Prob Num [60], is available at: https://github.com/JonathanWenger/itergp
Open Datasets Yes as well as a range of UCI datasets [61] with training set sizes n = 5, 287 to 57, 247, dimensions d = 9 to 26 and standardized features.
Dataset Splits Yes All experiments were run 10 times with randomly sampled training and test splits of 90/10 and we report average metrics with 95% confidence intervals.
Hardware Specification Yes All experiments were run on an NVIDIA GeForce RTX 2080 Ti graphics card.
Software Dependencies No An implementation of Algorithm 1, based on Ke Ops [48] and Prob Num [60], is available at: https://github.com/JonathanWenger/itergp. While specific software names are mentioned, their version numbers are not provided.
Experiment Setup No The paper states that hyperparameters are selected using a specific training procedure and that a zero mean prior and Matérn(1/2) kernel are used. However, it does not provide specific numerical values for hyperparameters such as learning rate, batch size, or optimizer settings within the main text.