Isss Online

Introduction to Generative Models

This course first introduces likelihoods, mixture models, and the expectation-maximisation algorithm. Next, we delve into variational inference -- including diffusion models, VAEs and energy-based models. We would also delve into adversarial training, including GANs. Finally, we would understand the foundations of LLMs using autoregressive models. If time permits, we would also delve into recognition-based probabilistic models (RPMs).

4 weeks (30 hrs)

Duration of course

Online

Learning Mode

Available

Recorded Video

Lecture Overview

Introduction to Generative Models

Module 1: Likelihoods, Mixture Models, EM Module 2: Variational Inference, Latent Variable models Module 3: Variational Autoencoders Module 4: Diffusion Models, Energy-based Models, Flow Models Module 5: Adversarial Learning-based Generative Modelling Module 6: Autoregressive Models Model 7: Recognition Parametrised Probabilistic Models (RPMs) -- optional Recorded/Live Sessions will be uploaded/conducted. We will concentrate on the mathematics behind these state-of-the-art models. We will see demonstrations in Python, and 2 assignments and solutions will be uploaded for problem-solving.

Introduction to Generative Models

Students wanting to understand models like DALLE, Chatbots today. But students must have the following background (atleast some understanding). It will be useful for UG, PG, and early PhD students who want to get started with advanced representation learning.
Linear Algebra
Probability and Statistics
Basics of Machine and Deep Learning
Basics of Multivariate Calculus – like gradients
Basics of Optimisation: Gradient Descent
Basic Python coding

Download Flyer