• Speaker: Dr. Mohammmad Ariful Haque, Professor, EEE, BUET
• Date: 11 December 2024
• Time: 3 PM
• Venue: EEE 920, ECE Building, BUET
• Title of the Talk: Transformers, LLMs, and RAG: Introduction to Building a Personal Research Assistant with Generative AI

Faculty Seminar - Dr. Mohammmad Ariful Haque

Transformers, LLMs, and RAG: Introduction to Building a Personal Research Assistant with Generative AI

The Dept. of EEE, BUET is introducing regular research seminar for fostering knowledge sharing environment and encouraging collaborative research effort. This will be a platform for EEE, BUET faculty member, members of academia from home and abroad as well as experts in the field of electrical engineering to share their experience in research and innovation. The details of the first talk are shared below.

Abstract:

Generative AI is transforming the way we work, learn, and solve problems. In electrical and electronic engineering, it has applications in areas such as automated system design, power system optimization, smart grids, predictive maintenance, fault detection and diagnosis, signal processing, robotics, creating virtual labs and research assistants. At the core of generative AI is the transformer architecture, a powerful tool that enables artificial neural networks to process and generate human-like text and other data. This talk introduces the foundational concepts of transformers and explains how they are used to build large language models (LLMs) like GPT, ChatGPT, BERT, T5, and LLAMA. We will discuss how these models learn from data using self-supervised learning, how they can be customized through fine-tuning, and how reinforcement learning with human feedback (RLHF) improves their performance. To make AI systems more reliable, we will explore retrieval-augmented generation (RAG), a method that connects LLMs to external information sources to reduce errors and ground their responses in factual data. The talk includes a live demonstration of building a research assistant using the RAG pipeline. We will also discuss the future of AI, including multimodal models that handle text, images, and more, and introduce NotebookLM, an advanced AI-powered research assistant developed by Google. This session aims to inspire students and faculty members to explore the potential of generative AI in their academic and professional endeavors.