Efficient Data Parallel Distributed Training with Flyte, Spark and Horovod

Explore efficient data parallel distributed training techniques using Flyte, Spark, and Horovod in this 41-minute conference talk presented by Ketan Umare and Katrina Rogan from Union.ai. Gain insights into the integration of these powerful tools for optimizing machine learning workflows. Learn about Flyte's architecture, concepts, and user journey, including workflow creation, registration, and execution. Discover how to leverage Spark for data processing and Horovod for distributed deep learning. The presentation covers key topics such as averages, code examples, stack traces, and an example scenario, providing a comprehensive overview of the subject matter. Enhance your understanding of distributed training methodologies and their practical applications in modern data science and machine learning projects.

Introduction
Agenda
Recap
Overview
Averages
Spark
What is Flyte
Workflows
User Journey
Code Example
Registration
Launching an execution
Graph of execution
Stack trace
Flyte concepts
Flyte architecture
Demo
Example Scenario