Explore a comprehensive lecture on setting up machine learning projects, focusing on best practices for planning and implementation. Learn about a framework for understanding ML projects, using pose estimation for robotic grasping as a running case study. Discover key points for prioritizing projects, choosing appropriate metrics, and creating effective human baselines. Gain insights into the lifecycle of ML projects, product design considerations, and strategies for combining metrics. Understand the importance of accuracy requirements and how to assess project feasibility. Delve into practical examples and engage with questions to deepen your understanding of ML project setup and management.
Setting Up Machine Learning Projects - Full Stack Deep Learning - March 2019
Overview
Syllabus
Intro
Goals for the lecture
Running case study - pose estimation
Hypothetical Co. Full Stack Robotics (FSR) wants to use pose estimation to enable grasping
Lifecycle of a ML project
Outline of the rest of the lecture
Key points for prioritizing projects
A (general) framework for prioritizing projects
Why are accuracy requirements so important?
Product design can reduce need for accuracy
Another heuristic for assessing feasibility
Key points for choosing a metric
Review of accuracy, precision, and recall
Why choose a single metric?
How to combine metrics
Combining precision and recall
Thresholding metrics
Example: choosing a metric for pose estimation
How to create good human baselines Quality of baseline Low
Key points for choosing baselines
Questions?
Taught by
The Full Stack
