Description: The goal of this research to develop a Bayesian approach using deep neural networks, specifically convolutional neural networks and recurrent neural networks. We train a Markov chain Monte Carlo (MCMC) parametrized by deep neural networks that converges and mixes quickly to target distribution. This is a generalization of Hamiltonian Monte Carlo, which is successful at sampling more challenging distributions such as strongly correlated Gaussian and mixture of Gaussian distributions. Based on data augmentation, a population estimation model is developed where MCMC is used to obtain the posterior distribution of population size.

Tools and skills : Python, PyTorch, R, capture-recapture population models.  

Supervision: Professor Nezam Kachouie.

This Research experience was in fullfilment of undergraduate research required for obtaining a BSc in Applied Mathematics at Florida Institute of Technology (Dec 2018).