- This event has passed.
Applied Statistics and Data Science Seminar on Monday March 10, 2025
March 10 @ 12:00 pm - 1:30 pm
Title: Variational Autoencoder Model for Exploring Latent Spaces in High-Dimensional Datasets
Venue, time and date: ISRT, 12:15 pm, March 10, 2025
Speaker: Mashfiqul Huq Chowdhury, PhD, Associate Professor at Mawlana Bhashani Science and Technology University
Abstract:
In this talk, I will focus on a probabilistic generative model known as the Variational Autoencoder (VAE). The VAE model uses variational Bayes to approximate the intractable posterior distribution over latent variables. I will begin by presenting the derivation of the evidence lower bound (ELBO) and then discuss the optimization procedure of the model. During training, the VAE learns smooth latent space representations through regularization. This learning paradigm can be applied to various tasks, including unsupervised clustering, regression, and the generation of new instances. To demonstrate its application, I will showcase how the VAE model can be applied to high-dimensional datasets and highlight the results in terms of clustering performance and sample generation.
