Overview
Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
Learn how to build effective data labeling pipelines for supervised machine learning projects through crowdsourcing in this 45-minute webinar. Explore real-life examples and best practices for obtaining high-quality labeled data that aligns with your specific problem. Discover the scalable approach of crowdsourcing across various domains, and gain insights into setting up instructions, interfaces, and quality control measures. Understand how to manage performers, implement behavior checks, and utilize pricing strategies for optimal results. Dive into topics such as aggregation techniques and integration with other machine learning tools to enhance your data labeling process.
Syllabus
Intro
Agenda
Labeled data: the missing pillar of Al
ML production pipeline
Data labelling requirements
Crowdsourcing - ML
Toloka platform
Crowdsourcing for ML data labelling
Instructions
Interface
Tolokers around the world
Filters Toloka example
Train your performers
Behavior checks
Fast responses example
Quality checks
Tips for control tasks
Control tasks example
Overlap and majority vote example
Pricing - Performance-based payment
Aggregation
Easy integration with other ML tools
Taught by
Open Data Science