Location: HRBRC HQ, 1070 University Blvd Suite 2105, 听Portsmouth, VA 23703, Room # 2111

Date: Thursday, April 10th, 2025,

Time: 12:00 p.m.

Teams Link:

Please join us at the HRBRC HQ, room 2111 for a fun and informal seminar! We will have ping-pong ready to go to de-compress after the seminar. In-person attendance is encouraged. :) This session will also be hosted virtually on Microsoft Teams.

In-person attendees can enjoy pizza, snacks, and beverages during the event.

Abstract:

Transformers have revolutionized artificial intelligence, powering groundbreaking advances behind state-of-the-art Large Language Models (LLMs) such as GPT-4. This seminar provides an in-depth exploration of the Transformer architecture, clearly illustrating tokenization, positional embeddings, and the mechanics of attention鈥攃overing detailed computations of queries, keys, and values through intuitive examples. We will also delve into advanced efficiency techniques, including KV caching, multi-head attention variants such as multi-query, grouped-query, and multi-head latent attention. Additionally, the discussion extends to Vision Transformers, highlighting key adaptations enabling transformers to excel in computer vision tasks and demonstrating their versatility across diverse fields. Whether you're new to Transformers or highly experienced, this seminar aims to deepen your understanding of the foundational technology shaping modern AI.

Speaker Bio:

Achyut Paudel is a Research Scientist at the Joint Institute of Advanced Computing for Environmental Studies (JI-ACES), a collaborative research center between 黑料不打烊 (黑料不打烊) and Jefferson Lab (JLab). His research emphasizes developing advanced data science methods aimed at addressing critical challenges in population and clinical health. Paudel earned his Ph.D. in Aerospace Engineering and Mechanics from the University of Alabama (2022), a Master's degree in Engineering from 脡cole Centrale Nantes, France (2017), and a Bachelor's degree in Engineering from Nanjing University of Aeronautics and Astronautics, China (2012). His expertise includes Uncertainty Quantification, Stochastic Optimization, Surrogate Modeling, Machine Learning, and Deep Learning. Outside of research, he enjoys outdoor activities, guitar playing, billiards, and ping pong.