Please Note: This seminar will be held online.
Sequential Monte Carlo for Applications in Structural Biology and Financial Time Series
Sequential Monte Carlo (SMC) methods are a set of algorithms widely used to sample from multivariate distributions of interest. We propose a new variant of SMC by extending the framework of the existing SMC method for producing unbiased particle weights to continuous state spaces. Our method provides an efficient way of sampling particles with unbiased weights to approximate the target distributions and thus can be used for Monte Carlo integration. Moreover, we demonstrate the optimality of our method in terms of minimizing the expected squared error loss. We present two applications of our method: (i) estimating the Boltzmann averages of the proposed protein structural quantities over twenty practical proteins data and the spike protein receptor-binding domain (RBD) of SARS-CoV-2 viruses and (ii) smoothing and parameter learning for the stochastic volatility model in financial time series.