Learn · Fundamentals of Generative AI and Large Language Models: Theory and Practice
Lecture 19 of 74Week 3 Variational Autoencoders (VAEs): Foundations
Intuition behind KL Divergence - Part A
Prof. Sriram Ganapathy, Prof. Ashwini Kodipalli and Prof. Baishali Garai · IISc Bangalore PES University RV University
Credits and licence
- From
- Fundamentals of Generative AI and Large Language Models: Theory and Practice, IISc Bangalore PES University RV University (at NPTEL)
- Licence
- CC BY-NC-SA, as NPTEL states it. Non-commercial use only.
- Attribution
- NPTEL states: "Distributed under Creative Commons Attribution-ShareAlike / CC BY - NC - SA", beside a BY-NC-SA badge. Coordinated by IISc Bangalore.
- Who can watch
- Tweens profiles and older. How courses are placed
All the lectures
- Introduction to Generative AI
- Activation Functions & Loss Functions in Deep Learning
- Optimizers - Part A
- Optimizers - Part B
- Convolutional Neural Network - Part A
- Convolutional Neural Network - Part B
- Introduction to Google Colab
- Practical Exercise - 1
- Practical Exercise - 2
- Introduction to Autoencoder
- Training objective: Reconstruction Loss (MSE, Binary Cross-Entropy)
- Types of Autoencoders: Architecture and Code-Level Understanding
- Regularization Techniques in AE: Denoising
- Regularization Techniques in AE: Sparse
- Regularization Techniques in AE: Contractive
- Numerical Example, Limitations of AE
- Practical Exercise - 1
- Practical Exercise - 2
- Intuition behind KL Divergence - Part A
- Intuition behind KL Divergence - Part B
- Introduction to VAE and the working of the Encoder
- Working of Probabilistic Decoder, ELBO, Loss Function of VAE
- Reparameterization Trick
- Numerical Example
- Practical Exercise - 1
- Entanglement, Disentanglement Latent Space, and Beta-VAE
- Conditional VAE
- Latent Space Interpolation
- Practical Exercise - 2
- Practical Exercise - 3
- Motivation for GANs
- GAN Architecture
- GAN Objective and Loss Functions
- GAN Convergence and Nash equilibrium
- Introduction to DCGAN
- Practical Exercise - 1
- Practical Exercise - 2
- Conditional GAN
- Pix2Pix GAN
- Cycle GAN
- Style GAN
- Style GAN 2
- Practical Exercise - 1
- Introduction to Diffusion Model
- Mathematics of Diffusion Models
- "Denoising Diffusion Probabilistic Model- Forward Process"
- "Denoising Diffusion Probabilistic Model- Reverse Process"
- Hand on on forward Diffusion process
- U-Net for Denoising
- Classifier Guided Diffusion Model
- Classifier free Diffusion Models
- Stable Diffusion Model
- Hands-on on reverse diffusion process
- Foundations of Natural Language Processing
- Sequential Modeling with RNNs and LSTM
- Evolution From LSTMs to Transformers
- Transformer Architecture: Encoder
- Practical Exercise - 1
- Transformer Architecture: Decoder
- Bidirectional Encoder Representations from Transformer (BERT)
- Generative Pre-trained Transformer (GPT) Decoder-only Large Language Model
- Prompt Engineering Basics
- Hands-on on LLM
- LLMs – Recap
- In-context Learning
- Low-Rank Adaptation (LoRA)
- RAG Principle
- RAG Recent Advances
- Tutorial on In-context Learning, RAG and LoRA
- LLMs for Text Generation
- LLMs for Audio, Video and Multimodal Tasks
- Size vs Performance and Benchmarking
- Bias, Fairness and Safety
- Tutorial on LLMs, Multimodal LLMs, Bias, Fairness, Evaluation and Safety