View Presentation
Overview
This is a non-technical, passion-driven presentation that explores the foundations of distributed deep learning. I explain key concepts such as data parallelism and model parallelism, and then dive into a research paper that caught my attention: ZeRO (Zero Redundancy Optimizer).
The presentation is conceptual in nature—no code was used—focusing instead on the ideas and motivations behind distributed approaches to training deep learning models.