Dive deep into the Sparsely-Gated Mixture-of-Experts (MoE) model in this comprehensive 23-minute video tutorial. Explore the challenges of training large models with limited resources and discover how MoE addresses these issues through conditional computation. Learn about dense and sparse models, pathway models, and the internal architecture of MoE. Understand how MoE processes text and image data, and gain insights into its components and functionality. Access research papers, code examples, and GitHub resources to further your understanding of this innovative approach to machine learning model design and implementation.
Research Paper Deep Dive - The Sparsely-Gated Mixture of Experts
Overview
Syllabus
- Paper Introduction
- Understanding the Problem
- Significant computation requirement
- Solution - Conditional Computation
- Dense and Sparse Models
- Pathway Models
- MoE Introduction
- MoE Internals
- MoE Components
- Data Processing in MoE
- Text Data Processing in MoE
- Image Data Processing in MoE
- Text and Image Data Processing in MoE
- Research Paper and Code
- Resources and GitHub Reference
Taught by
Prodramp
