arXiv:2405.01737v1 Announce Kind: new
Summary: Probability-free inference strategies primarily based on neural conditional density estimation have been proven to drastically cut back the simulation burden compared to classical strategies corresponding to ABC. When utilized within the context of any latent variable mannequin, corresponding to a Hidden Markov mannequin (HMM), these strategies are designed to solely estimate the parameters, relatively than the joint distribution of the parameters and the hidden states. Naive software of those strategies to a HMM, ignoring the inference of this joint posterior distribution, will thus produce an inaccurate estimate of the posterior predictive distribution, in flip hampering the evaluation of goodness-of-fit. To rectify this drawback, we suggest a novel, sample-efficient likelihood-free technique for estimating the high-dimensional hidden states of an implicit HMM. Our strategy depends on studying immediately the intractable posterior distribution of the hidden states, utilizing an autoregressive-flow, by exploiting the Markov property. Upon evaluating our strategy on some implicit HMMs, we discovered that the standard of the estimates retrieved utilizing our technique is similar to what may be achieved utilizing a way more computationally costly SMC algorithm.
Supply hyperlink