Particle Semi-Implicit Variational Inference

Authors: Jen Ning Lim, Adam Johansen

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical results demonstrate that PVI performs favourably compared to other SIVI methods across various tasks. In this section, we compare PVI against other semi-implicit VI methods.
Researcher Affiliation Academia Jen Ning Lim University of Warwick Coventry, United Kingdom Jen-Ning.Lim@warwick.ac.uk Adam M. Johansen University of Warwick Coventry, United Kingdom a.m.johansen@warwick.ac.uk
Pseudocode Yes Algorithm 1 Particle Variational Inference (PVI)
Open Source Code Yes The code is available at https://github.com/jenninglim/pvi.
Open Datasets Yes We consider a Bayesian logistic regression problem on the waveform dataset (Breiman and Stone, 1984). Concrete (Yeh, 2007) Protein (Rana, 2013) Yacht (Gerritsma et al., 2013)
Dataset Splits No The paper does not explicitly state validation dataset splits for its experiments.
Hardware Specification Yes The code was written in JAX (Bradbury et al., 2018) and executed on a NVIDIA Ge Force RTX 4090.
Software Dependencies No The code was written in JAX (Bradbury et al., 2018). (No version number for JAX or other specific libraries).
Experiment Setup Yes PVI. We use M = 100, λθ = 0, λr = 10^-8, hx = 10^-2, hθ = 10^-4, Ψθ we use the RMSProp preconditoner, Ψr = Idz, and L = 250.