Neural Posterior Estimation for Intractable Likelihoods

Project description

This project is concerned with the development and application of Neural Posterior Estimation (NPE), a cutting-edge technique in simulation-based inference (SBI) that uses neural networks to approximate posterior distributions in statistical models where traditional methods (e.g. Markov Chain Monte Carlo) are computationally expensive/inefficient. NPE is particularly useful in scenarios involving complex likelihoods, high-dimensional data, and stochastic models commonly encountered in many application areas, leading to likelihood functions which are not tractable analytically.

The project provides an opportunity to bridge modern machine learning tools with classical statistical inference techniques. The student will gain expertise in simulation-based inference, neural networks, and Bayesian computation, contributing to the development of new methodologies at the intersection of statistics and AI.