Cost-aware Simulation-based Inference

2024-10-10

arXiv.org

AI Summary

Traditional simulation-based inference (SBI) methods—such as neural SBI and approximate Bayesian computation—suffer from prohibitive computational costs when applied to expensive, parameter-dependent simulators. Method: We propose the first cost-aware SBI framework, jointly designing rejection sampling and self-normalized importance sampling to explicitly model and optimize the trade-off between simulation budget and statistical efficiency; a neural density estimator is employed for posterior approximation. Contribution/Results: Evaluated on intractable real-world models from epidemiology and telecommunications engineering, our method reduces simulator calls by over 50% on average while maintaining high-fidelity posterior estimation. This work pioneers the integration of explicit computational cost modeling into the SBI paradigm, yielding a scalable and practical solution for Bayesian inference under high simulation costs.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Reasoning under Uncertainty: Probabilistic Inference
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Simulation-based inference (SBI) is the preferred framework for estimating parameters of intractable models in science and engineering. A significant challenge in this context is the large computational cost of simulating data from complex models, and the fact that this cost often depends on parameter values. We therefore propose extit{cost-aware SBI methods} which can significantly reduce the cost of existing sampling-based SBI methods, such as neural SBI and approximate Bayesian computation. This is achieved through a combination of rejection and self-normalised importance sampling, which significantly reduces the number of expensive simulations needed. Our approach is studied extensively on models from epidemiology to telecommunications engineering, where we obtain significant reductions in the overall cost of inference.

Problem

• Reduces computational cost of simulation-based inference
• Applies cost-aware methods to complex model simulations
• Combines rejection and importance sampling techniques

Innovation

• Cost-aware simulation-based inference
• Rejection and importance sampling
• Reduces expensive simulations needed