GAD-PVI: A General Accelerated Dynamic-Weight Particle-Based Variational Inference Framework

Authors: Fangyikang Wang, Huminhao Zhu, Chao Zhang, Hanbin Zhao, Hui Qian

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on both synthetic and real-world data demonstrate the faster convergence and reduced approximation error of GAD-PVI methods over the state-of-the-art. We evaluate our algorithms on various synthetic and real-world tasks. The empirical results demonstrate the superiority of our GAD-PVI methods.
Researcher Affiliation Academia 1College of Computer Science and Technology, Zhejiang University 2Advanced Technology Institute, Zhejiang University 3State Key Lab of CAD&CG, Zhejiang University
Pseudocode Yes Algorithm 1: General Accelerated Dynamic-weight Particlebased Variational Inference (GAD-PVI) framework
Open Source Code Yes Our appendix can be downloaded at https://github.com/zituitui/GAD-PVI paper
Open Datasets Yes We follow the experiment setting in (Chen et al. 2018b), and use the dataset LIDAR which consists of 221 observations. In this experiment, we study a Bayesian regression task with Bayesian neural network on 4 datasets from UCI and LIBSVM.
Dataset Splits Yes In the BNN task, we split 1/5 of the training set as our validation set to tune the parameters.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions 'POT library (Flamary et al. 2021)' but does not specify a version number. Other software mentions are generic without version details.
Experiment Setup Yes For all the algorithms, the particles weights are initialized to be equal. In the BNN task, we split 1/5 of the training set as our validation set to tune the parameters. The position step-size are tuned via grid search for the fixed-weight Par VI algorithms, then used in the corresponding dynamic-weight algorithms. The acceleration parameters and weight adjustment parameters are tuned via grid search for each specific algorithm. We use a one-hidden-layer neural network with 50 hidden units and maintain 128 particles. For all the datasets, we set the batchsize as 128.