Sparse Variational Inference: Bayesian Coresets from Scratch

Authors: Trevor Campbell, Boyan Beronov

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we compare the quality of coresets constructed via the proposed Sparse VI greedy coreset construction method, uniform random subsampling, and Hilbert coreset construction (GIGA [32]).
Researcher Affiliation Academia Trevor Campbell Department of Statistics University of British Columbia Vancouver, BC V6T 1Z4 trevor@stat.ubc.ca Boyan Beronov Department of Computer Science University of British Columbia Vancouver, BC V6T 1Z4 beronov@cs.ubc.ca
Pseudocode Yes Algorithm 1 Greedy sparse stochastic variational inference
Open Source Code Yes code is available at www.github.com/trevorcampbell/bayesian-coresets.
Open Datasets Yes This dataset was constructed by merging housing prices from the UK land registry data https://www.gov.uk/government/statistical-data-sets/price-paid-data-downloads with latitude & longitude coordinates from the Geonames postal code data http://download.geonames.org/export/zip/.
Dataset Splits No The paper describes the datasets used and the overall experiment settings (e.g., number of samples, iterations) but does not provide specific train, validation, or test dataset split percentages or counts, nor does it reference standard dataset splits for reproduction.
Hardware Specification Yes Experiments were performed on a machine with an Intel i7 8700K processor and 32GB memory
Software Dependencies No The paper describes the algorithms and computational environment (e.g., processor, memory) but does not specify software dependencies with version numbers (e.g., specific Python, PyTorch, or library versions).
Experiment Setup Yes We used a learning rate of γt = t^-1, T = 100 weight update optimization iterations, and M = 200 greedy iterations, although note that this is an upper bound on the size of the coreset as the same data point may be selected multiple times.