CSE755: Generative Machine Learning

Fundamentals of Generative Models: probabilistic modeling, latent variable models, comparison with discriminative models; Autoregressive Models: PixelCNN, PixelRNN, sequential data generation; Variational Autoencoders (VAEs): encoder-decoder structure, latent space representation, reparameterization trick; Generative Adversarial Networks (GANs): architecture, loss functions, training challenges, improvements (WGAN, StyleGAN, BigGAN); Diffusion Models: denoising diffusion probabilistic models (DDPMs), score-based generative models, latent diffusion models; Energy-Based Models: Boltzmann machines, contrastive divergence, restricted Boltzmann machines; Normalizing Flows: invertible transformations, RealNVP, Glow, density estimation; Transformer-Based Generative Models: GPT, BERT for text generation, Vision Transformers (ViTs) for image synthesis; Multimodal Generation: text-to-image (DALL·E, Stable Diffusion), text-to-video, speech synthesis; Applications: data augmentation, drug discovery, creative AI, deepfakes; Challenges and Ethical Considerations: mode collapse, fairness, bias, responsible AI in generative models.

Course Objectives

The core objectives of this course are to:
Learn the principles behind generative algorithms, including probabilistic modeling, neural networks, and deep learning architectures
Gain hands-on experience with models like Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and diffusion models for creating images, text, and other data
Understand how to use generative AI for applications such as image synthesis, natural language generation etc.
Learn techniques for assessing model quality, mitigating biases, improving stability, and ensuring ethical ML use in generative applications

List of Books

1. To Be Added

Course Outcome