Coin Sampling: Gradient-Based Bayesian Inference without Learning Rates

Authors: Louis Sharrock, Christopher Nemeth

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We illustrate the performance of our approach on a range of numerical examples, including several high-dimensional models and datasets, demonstrating comparable performance to other Par VI algorithms with no need to tune a learning rate.
Researcher Affiliation Academia 1Department of Mathematics, Lancaster University, UK.
Pseudocode Yes Algorithm 1 Coin Wasserstein Gradient Descent
Open Source Code Yes Code to reproduce our numerical results can be found at https://github.com/louissharrock/Coin-SVGD.
Open Datasets Yes We test our algorithm using the Covertype dataset, which consists of 581,012 data points and 54 features. (Gershman et al., 2012) [...] We test the performance of our algorithms on several UCI datasets. (Liu & Wang, 2016; Hernandez-Lobato & Adams, 2015) [...] We test our algorithm on the Movie Lens dataset (Harper & Konstan, 2015)...
Dataset Splits Yes We randomly partition the data into a training dataset (70%), validation dataset (10%), and testing dataset (20%).
Hardware Specification Yes We perform all experiments using a Mac Book Pro 16 (2021) laptop with Apple M1 Pro chip and 16GB of RAM.
Software Dependencies No The paper mentions 'Python 3, Py Torch, Theano, and Jax' but does not provide specific version numbers for these software components.
Experiment Setup Yes In all cases, we run both algorithms using N = 20 particles, and for T = 1000 iterations. We initialise the particles according to (θi 0)N i=1 i.i.d. N(0, 0.12).