Vector-Valued Control Variates

Authors: Zhuo Sun, Alessandro Barp, Francois-Xavier Briol

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate our methodology on a range of problems including multifidelity modelling, Bayesian inference for dynamical systems, and model evidence computation through thermodynamic integration.
Researcher Affiliation Academia 1University College London, London, UK 2University of Cambridge, Cambridge, UK 3The Alan Turing Institute, London, UK.
Pseudocode Yes Algorithm 1 Block-coordinate descent for vv-CVs with unknown task relationship
Open Source Code Yes The code to reproduce our results is available at: https://github.com/jz-fun/ Vector-valued-Control-Variates-Code.
Open Datasets Yes We use the dataset of snowshoe hares (preys) and Canadian lynxes (predators) from Hewitt (1921), and implement Bayesian inference on model parameters x by using no U-turn sampler (NUTS) in Stan (Carpenter et al., 2017).
Dataset Splits No The paper focuses on Monte Carlo methods and discusses sample sizes for integration tasks (e.g., 'm = (m L, m H) = (40, 40)'), but it does not specify explicit train/validation/test dataset splits with percentages or absolute counts for the input data itself.
Hardware Specification No The paper does not provide specific hardware details such as GPU models (e.g., NVIDIA A100), CPU models (e.g., Intel Xeon), or cloud compute specifications used for running experiments.
Software Dependencies No The paper mentions using 'Stan (Carpenter et al., 2017)' and the 'Adam optimiser (Kingma & Ba, 2015)' but does not provide specific version numbers for these or any other software dependencies crucial for replication.
Experiment Setup Yes Hyper-parameter tuning: batch size 5; learning rate 0.05; total number of epochs 30. Base kernel: squared exponential kernel Optimisation: λ = 0.001; batch size is 5; learning rate is 0.001; total number of epochs 400.