A0485
Title: Simulation-based inference for Markov time series via diffusion models
Authors: Kenji Fukumizu - The Institute of Statistical Mathematics (Japan) [presenting]
Abstract: Amortized simulation-based inference (SBI) methods train neural networks on simulated data to perform Bayesian inference. While this strategy avoids the need for tractable likelihoods, it often requires a large number of simulations and has been challenging to scale to time series data. Scientific simulators frequently emulate real-world dynamics through thousands of single-state transitions over time. An SBI approach is proposed that can exploit such Markovian simulators by locally identifying parameters consistent with individual state transitions. These local results are then composed to obtain a posterior over parameters that align with the entire time series observation. The focus is on applying this approach to neural posterior score estimation with diffusion models as a posterior sampler. It is demonstrated that the approach is more simulation-efficient than directly estimating the global posterior on several synthetic benchmark tasks and simulators used in ecology and epidemiology. Finally, the scalability and simulation efficiency of the approach are validated by applying it to a high-dimensional Kolmogorov flow simulator with around one million data dimensions.
