On Divergence Measures for Bayesian Pseudocoresets

Authors: Balhae Kim, Jungwon Choi, Seanie Lee, Yoonho Lee, Jung-Woo Ha, Juho Lee

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical results demonstrate that the pseudocoresets constructed from these methods reflect the true posterior even in high-dimensional Bayesian inference problems.
Researcher Affiliation Collaboration KAIST1, Stanford University2, NAVER AI Lab3, AITRICS4 {balhaekim, jungwon.choi, lsnfamily02}@kaist.ac.kr, yoonho@stanford.edu, jungwoo.ha@navercorp.com, juholee@kaist.ac.kr
Pseudocode Yes Algorithm 1 Bayesian Pseudocoresets with Forward KL
Open Source Code No The paper does not provide any explicit statement about releasing source code or a link to a code repository for the described methodology.
Open Datasets Yes We use the CIFAR10 dataset [15] to generate Bayesian pseudocoresets, and evaluate on the test split of CIFAR10 in addition to the CIFAR10-C dataset [12]
Dataset Splits No The paper mentions evaluating on the "test split" of CIFAR10 but does not explicitly detail the training/validation splits, their percentages, or refer to a specific citation for how these splits were obtained (e.g., "standard train/validation split from X").
Hardware Specification Yes We use 32 cores of Intel Xeon CPU Gold 5120 and 4 Tesla V100s.
Software Dependencies No The paper does not provide specific version numbers for any ancillary software dependencies (e.g., Python, PyTorch, TensorFlow, etc.).
Experiment Setup Yes As our results were not sensitive to the choice of hyperparameters, we used a single set of hyperparameters that performed best in initial experiments. Please refer to Appendix B for detailed evaluation settings including hyperparameters.