Generative AI – A Practical Approach

Name: Generative AI – A Practical Approach
Rating: 5.0 (1 reviews)

Generative AI with Auto Encoder, Generative Adverserial Network, Large Language Models etc...

Created bySoubraylu Sivakumar

Last updated 6/2025

English

English [Auto],

What you'll learn

Understand the architecture and applications of autoencoders for data compression and reconstruction.
Explore GANs for generating realistic synthetic data across various domains.
Apply transformer models to solve sequence-based generative tasks effectively.
Integrate hybrid deep learning models for enhanced generative performance and flexibility.

Course content

2 sections • 9 lectures • 2h 22m total length

Introduction to GenAI-19:52
This lecture introduces the fundamentals of Artificial Intelligence and dives into the specialized domain of Generative AI (GenAI). It covers how GenAI models work, including their core principles and mechanisms. The session also highlights the growing importance of GenAI across industries. Finally, key challenges such as ethical concerns, data bias, and model limitations will be explored.
Introduction to GenAI-215:38
This lecture explores the diverse use cases and real-world applications of Generative AI across industries like healthcare, media, and education. It examines current challenges including data privacy, model reliability, and ethical concerns. A comparison between traditional AI and Generative AI will be provided to highlight key differences. The session also contrasts GenAI with Deep Learning, clarifying their relationship and unique capabilities. Overall, it offers a practical understanding of where and how GenAI is making an impact.

Introduction to Auto Encoder23:28
This lecture provides a comprehensive introduction to autoencoders, starting with their origin and evolution in the field of deep learning. It explains their core purpose, different types, and the underlying mathematical concepts. Practical examples and inference techniques are discussed to demonstrate their functionality. The session highlights key applications, benefits, and limitations of autoencoders. It concludes with future trends and comparisons with related models.
Vanilla Auto Encoder - Handson11:14
In this hands-on session, students will implement a vanilla autoencoder using Python and a deep learning framework. They will learn to train and evaluate it on real datasets.
Vanilla Autoencoder
Regularized and Non-Regularized Autoencoders25:47
This lecture explores both regularized and non-regularized autoencoders, beginning with their fundamental concepts and motivations. It covers various types of regularized AEs like Sparse, Denoising, Contractive, and Variational, along with their specific advantages and limitations. Non-regularized variants such as Vanilla, Undercomplete, Overcomplete, Symmetric, Shallow, and Deep autoencoders are also examined. The session includes comparisons between the two categories to highlight their use cases. Future directions and advancements in autoencoder research conclude the discussion.
Variational Autoencoder22:27
This lecture introduces Variational Autoencoders (VAEs), explaining why they are needed beyond traditional autoencoders. It covers real-world applications and the mathematical foundation behind VAEs, including latent space representation. An example is provided to illustrate how VAEs generate new data. The session also discusses the advantages, disadvantages, and how VAEs compare with other generative models. Finally, it explores the future potential and research directions of VAEs.
Variational Autoencoder - Handson-117:33
Contains core VAE model architecture (encoder/decoder classes), helper functions (loss calculations, metrics), and the main training loop. Handles data loading, model initialization, training process with loss computation, and metric tracking. Saves training progress and model checkpoints.
Variational Autoencoder - Handson-211:28
Focuses on evaluation and visualization - includes functions for reconstruction quality assessment, latent space visualization (t-SNE, distributions), and training progress plotting. Main execution block trains multiple VAEs with different latent dimensions and generates comparative results. Handles all visual outputs and final metric comparisons.
Variational Autoencoder - Handson-35:23
Training Progress Charts
Visualizes loss components (reconstruction, KL divergence, Dice) and quality metrics (PSNR, SSIM) across epochs. Includes comparative plots of train/test performance. Saved as high-resolution PNG files with latent dimension in titles.
Excel Metrics Sheets
Contains detailed epoch-by-epoch metrics (losses, MSE, PSNR) and final model performance summaries. Organized with separate tabs for different latent dimensions. Enables quantitative comparison of compression efficiency and reconstruction quality.
Latent Space Visualizations
Generates t-SNE plots, dimension distributions, and activation statistics for the learned latent space. Saved as PNG files showing cluster separation and dimension utilization patterns for each model configuration.
Variational Autoencoder

Requirements

Python Programming, Machine Learning and Deep Learning

Description

This course offers a practical and in-depth exploration into the world of Generative AI, focusing on widely used and impactful models such as Autoencoders, Generative Adversarial Networks (GANs), and Large Language Models (LLMs). Designed for learners with a basic understanding of machine learning and Python, the course begins by introducing the fundamentals of generative modeling—how machines learn to create data that mimics real-world patterns.

Students will first explore Autoencoders, including their vanilla and variational variants, and learn how to use them for tasks such as dimensionality reduction, anomaly detection, and data reconstruction. The course then transitions into GANs, diving into their unique adversarial training structure, generator-discriminator dynamics, and how they are used to create realistic images, audio, and other content.

Next, learners will engage with transformers and LLMs, understanding how these models power modern tools like ChatGPT, enabling natural language generation, summarization, and creative writing. Each module includes hands-on coding exercises using popular deep learning frameworks to solidify theoretical concepts through real-world application.

The course also addresses challenges such as training stability, ethical considerations, and model evaluation. Comparisons between generative approaches help students choose the right tool for specific tasks. By the end of the course, learners will be equipped to design, build, and apply generative AI models across various domains.

Who this course is for:

This course is designed for data scientists, ML engineers, AI enthusiasts, and researchers who want hands-on experience with generative models like Autoencoders, GANs, and Transformers. A basic understanding of Python and deep learning is recommended.

Generative AI – A Practical Approach

What you'll learn

Explore related topics

Course content

Introduction2 lectures • 26min

Auto Encoders7 lectures • 1hr 57min

Requirements

Description

Who this course is for: