# Fast Estimation of Bayesian State Space Models Using Amortized Simulation-Based Inference

arXiv: https://arxiv.org/abs/2210.07154 · HTML: https://mlsanity.com/paper/2210.07154/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 7 of its 85 references that are in the map.
- First posted on arXiv 13 Oct 2022.

## Builds on (most cited first, 7 of 7)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Variational Inference: A Review for Statisticians](https://mlsanity.com/paper/1601.00670/) (2016)
- [An Introduction to Variational Autoencoders](https://mlsanity.com/paper/1906.02691/) (2019)
- [Meta-Learning Priors for Efficient Online Bayesian Regression](https://mlsanity.com/paper/1807.08912/) (2018)
- [Gaussian variational approximation with sparse precision matrices](https://mlsanity.com/paper/1605.05622/) (2016)
- [An Adversarial Approach to Structural Estimation](https://mlsanity.com/paper/2007.06169/) (2020)
- [Probabilistic Surrogate Networks for Simulators with Unbounded Randomness](https://mlsanity.com/paper/1910.11950/) (2019)

## Abstract (arXiv)

This paper presents a fast algorithm for estimating hidden states of Bayesian state space models. The algorithm is a variation of amortized simulation-based inference algorithms, where a large number of artificial datasets are generated at the first stage, and then a flexible model is trained to predict the variables of interest. In contrast to those proposed earlier, the procedure described in this paper makes it possible to train estimators for hidden states by concentrating only on certain characteristics of the marginal posterior distributions and introducing inductive bias. Illustrations using the examples of the stochastic volatility model, nonlinear dynamic stochastic general equilibrium model, and seasonal adjustment procedure with breaks in seasonality show that the algorithm has sufficient accuracy for practical use. Moreover, after pretraining, which takes several hours, finding the posterior distribution for any dataset takes from hundredths to tenths of a second.