Efficient State-space Exploration in Massively Parallel Simulation Based Inference

TL;DR

This paper introduces a novel parallel SBI algorithm that replaces a fixed step-size with a Beta distribution, enabling more efficient parameter exploration and significantly reducing the number of simulations needed.

Contribution

It presents a new ABC-SMC algorithm leveraging Beta-distributed step-sizes to improve parameter space exploration and efficiency in massively parallel settings.

Findings

Achieves similar parameter learning quality with 100x fewer simulations.

Reduces run-to-run variance by approximately 80x across trials.

Demonstrates effectiveness on an epidemiology model using GPU.

Abstract

Simulation-based Inference (SBI) is a widely used set of algorithms to learn the parameters of complex scientific simulation models. While primarily run on CPUs in HPC clusters, these algorithms have been shown to scale in performance when developed to be run on massively parallel architectures such as GPUs. While parallelizing existing SBI algorithms provides us with performance gains, this might not be the most efficient way to utilize the achieved parallelism. This work proposes a new algorithm, that builds on an existing SBI method - Approximate Bayesian Computation with Sequential Monte Carlo(ABC-SMC). This new algorithm is designed to utilize the parallelism not only for performance gain, but also toward qualitative benefits in the learnt parameters. The key idea is to replace the notion of a single 'step-size' hyperparameter, which governs how the state space of parameters is explored during learning, with step-sizes sampled from a tuned Beta distribution. This allows this new ABC-SMC algorithm to more efficiently explore the state-space of the parameters being learnt. We test the effectiveness of the proposed algorithm to learn parameters for an epidemiology model running on a Tesla T4 GPU. Compared to the parallelized state-of-the-art SBI algorithm, we get similar quality results in $\sim 100$x fewer simulations and observe ~80x lower run-to-run variance across 10 independent trials.