Bayesian Coreset Construction via Greedy Iterative Geodesic Ascent

Authors: Trevor Campbell, Tamara Broderick

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The paper concludes with validation of GIGA on both synthetic and real datasets, demonstrating that it reduces posterior approximation error by orders of magnitude compared with previous coreset constructions.
Researcher Affiliation Academia 1Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology, Cambridge, MA, United States.
Pseudocode Yes Algorithm 1 GIGA: Greedy Iterative Geodesic Ascent
Open Source Code Yes Code for these experiments is available at https://github.com/trevorcampbell/bayesian-coresets.
Open Datasets No The paper mentions datasets like 'Phishing', 'DS1', 'Bike Trips', and 'Airport Delays' and refers to Appendix D for references, but Appendix D only lists paper citations, not direct links or explicit statements of public availability for the datasets themselves. Synthetic datasets are generated.
Dataset Splits No The paper mentions 'posterior sampling steps' and using 'coreset' vs 'full dataset' but does not specify explicit training, validation, and test dataset splits with percentages or sample counts.
Hardware Specification No The paper does not provide specific hardware details such as exact GPU/CPU models or processor types used for running experiments.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers (e.g., library or solver names with version numbers).
Experiment Setup Yes For posterior inference, we used Hamiltonian Monte Carlo (Neal, 2011) with 15 leapfrog steps per sample. We simulated a total of 6,000 steps, with 1,000 warmup steps for step size adaptation with a target acceptance rate of 0.8, and 5,000 posterior sampling steps.