ePubs

The open archive for STFC research publications

Full Record Details

Persistent URL http://purl.org/net/epubs/work/63574850
Record Status Checked
Record Id 63574850
Title Sequential Monte Carlo for Bayesian Optimisation
Contributors
Abstract We present the results of the PRACE Summer of High Performance Computing 2022 project exploring acceleration of Bayesian optimisation on arbitrary models converted from Stan using NumPyro. We demonstrate that starting from the basic ensembles of statistical physics and classical mechanics an efficient and scalable particle-based optimisation routine can be developed. We demonstrate the scalability of our implementation using NumPyro and Jax to scale the results across multiple CPU-only nodes as well as GPUs without requiring additional modifications of the code.
Organisation STFC , HC
Keywords Bayesian modelling , statistical physics , NumPyro , Stan , HPC , Bayesian inference
Funding Information
Related Research Object(s): 63563823
Licence Information: Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)
Language English (EN)
Type Details URI(s) Local file(s) Year
Presentation Presented at 23rd International Conference on Computer Science (ICCS 2023), Prague, Czech Republic, 3-5 Jul 2023. ICCS2023_AL_TW_ES_PhysicsSMC_V2.pdf 2023