Cyber2A Workshop

Foundation Models: The Cornerstones of Modern AI

Overview

Foundation models (FM) are deep learning models trained on massive raw unlabelled datasets usually through self-supervised learning. FMs enable today’s data scientists to use them as the base and fine-tune using domain specific data to obtain models that can handle a wide range of tasks [1, 6, 7]. In this talk, we provide an introduction to FMs, its history, evolution, and go through its key features and categories, and a few examples. We also briefly discuss how foundation models work. This talk will be a precursor to the hands-on session that follows on the same topic.

Image source: 2021 paper on foundation models by Stanford researchers [1].

In this session, we take a closer look at what constitutes a foundation model, a few examples, and some basic principles around how it works.

Outline

  1. Introduction to foundation models, its history and evolution
  2. Key features of foundation models
  3. Types of foundation models: Language, Vision, Generative, and Multimodal
  4. Examples of foundation models: BERT [3], GPT [4], YOLO [2], SAM [5], DALLE-2
  5. How do foundation models work?

Reference

  1. On the opportunities and risk of Foundation models
  2. You Only Look Once
  3. BERT
  4. GPT3
  5. Segment Anything Model
  6. NVIDIA blog post on foundation models
  7. What are Foundation Models? - Generative AI