Generative AI ( Expert Lvl )

• Upon completion of this training program, participants will be able to:

1. Understand the fundamental concepts and principles of Generative AI.

2. Explain the architecture and functionality of various generative models. (GANs, VAEs, Diffusion Models, Transformers)

3. Describe the key components and architectures of Large Language Models (LLMs), Large Vision Models (LVMs), and Small Language Models (SLMs).

4. Apply best practices for fine-tuning, prompting, and evaluating generative models.

5. Use MLOps principles and tools to manage the lifecycle of generative AI models.

6. Design and implement AI agents that leverage generative models for various tasks.

7. Understand the ethical considerations and responsible development of generative AI.

8. Stay current with the latest advancements in generative AI research.

60 hr

The program is structured into 20 sessions, each lasting 3 hours. Each session will blend theoretical explanations with practical hands-on labs. The following outline highlights the key topics covered:

1. Session 1: Foundations of Generative AI

• • What is Generative AI? (Definition, History, Applications)

  • Types of Generative Models: GANs, VAEs, Diffusion Models, Autoregressive Models

  • Review of GANs and VAEs from Deep Learning Course (Brief Recap)

  • Evaluation Metrics for Generative Models: Inception Score, FID Score

  • Hands-on Project: Experimenting with pre-trained GANs or VAEs for image generation.

            2. Session 2: Autoregressive Models and Transformers

• • Introduction to Autoregressive Models: Markov Chains, N-gram Models

  • Transformers: A Deep Dive (Review from DL course, focusing on the decoder)

  • Self-Attention, Multi-Head Attention

  • Applications of Autoregressive Models: Text Generation, Music Generation

  • Hands-on Project: Building a simple autoregressive model for text generation.

           3. Session 3: Introduction to Large Language Models

• • What are LLMs? (Scale, Capabilities, Limitations)

  • Architecture of LLMs: Transformer-based Architectures (BERT, GPT, T5, LLaMA)

  • Pre-training and Fine-tuning of LLMs

  • Prompt Engineering: Designing effective prompts for LLMs

  • Hands-on Project: Exploring pre-trained LLMs using Hugging Face Transformers.

           4. Session 4: Prompt Engineering Techniques

• • Zero-Shot Learning, Few-Shot Learning

  • Chain-of-Thought Prompting

  • Retrieval-Augmented Generation (RAG)

  • Prompt Optimization

  • Hands-on Project: Applying different prompt engineering techniques to improve the performance of an LLM on a specific task.

           5. Session 5: Fine-tuning LLMs

• • Why Fine-tune LLMs?

  • Full Fine-tuning vs. Parameter-Efficient Fine-tuning (PEFT): LoRA, Adapter Layers

  • Preparing Data for Fine-tuning

  • Training and Evaluating Fine-tuned LLMs

  • Hands-on Project: Fine-tuning a pre-trained LLM for a specific task using PEFT techniques.

            6. Session 6: Evaluation and Analysis of LLMs

• • Metrics for Evaluating LLMs: Perplexity, BLEU Score, ROUGE Score, Human Evaluation

  • Analyzing the Outputs of LLMs: Identifying Biases, Hallucinations, and Errors

  • Techniques for Mitigating Biases and Errors

  • Hands-on Project: Evaluating the performance of a fine-tuned LLM and analyzing its outputs.

           7. Session 7: LLM Safety and Responsible AI

• • Risks associated with LLMs: Misinformation, Bias, Privacy

  • Techniques for mitigating risks: Content filtering, safety layers, adversarial training

  • Responsible AI principles and guidelines

  • Discussion: Ethical considerations of using LLMs in real-world applications.

          8. Session 8: Introduction to Large Vision Models

• • What are LVMs? (Scale, Capabilities, Limitations)

  • Architecture of LVMs: Vision Transformers (ViT), CLIP

  • Image Generation with LVMs: DALL-E, Stable Diffusion, Midjourney

  • Hands-on Project: Exploring image generation with pre-trained LVMs.

         9. Session 9: Text-to-Image Generation

• • Understanding Diffusion Models

  • Prompt Engineering for Image Generation

  • Controlling the Style and Content of Generated Images

  • Hands-on Project: Generating images from text prompts using Stable Diffusion or similar models.

        10. Session 10: LVM Applications

• Image Editing

• Video Generation

• 3D Modeling

• Hands-on Project: Exploring different applications of LVMs.

        11. Session 11: Introduction to Small Language Models

• What are SLMs? (Motivation, Advantages, and Disadvantages)

• Techniques for Compressing and Optimizing LLMs: Knowledge Distillation, Pruning, Quantization

• Hands-on Project: Implementing knowledge distillation to train a smaller model from a larger model.

        12. Session 12: Deployment of SLMs

• Edge Computing

• Mobile Devices

• Resource-Constrained Environments

• Hands-on Project: Deploying an SLM to a resource-constrained environment (e.g., Raspberry Pi).

        13. Session 13: Combining LLMs and LVMs

• • Multimodal Embeddings

  • Visual Question Answering

  • Image Captioning

  • Hands-on Project: Building a visual question answering system.

     14. Session 14: Building Multimodal Applications

• Image Editing with Text

• Video Summarization

• Creating Interactive Experiences

• Hands-on Project: Building a multimodal application of your choice.

( Check out the link for the other topics covered in Session 15...Session 20 )

• Total Credit Hours: 60
• Session Length: 3 hours per session
• Number of Sessions: 20
• Delivery Method: Remotely

• Upon completion of this course, trainees will be able to:

1. Differentiate between various generative modeling techniques.

2. Fine-tune pre-trained LLMs for specific tasks using techniques like parameter-efficient fine-tuning (PEFT).

3. Develop and evaluate prompts for LLMs to elicit desired outputs.

4. Integrate LLMs and LVMs to build multimodal AI systems.

5. Design and implement MLOps pipelines for generative AI models, including model versioning, deployment, and monitoring.

6. Build AI agents that can interact with users and perform tasks using generative models.

7. Critically evaluate the outputs of generative models and identify potential biases or limitations.

8. Contribute to generative AI projects and research.

9. Apply ethical principles to the development and deployment of generative AI systems.