Training AI with Humans

This course explores the intersection of machine learning (ML) and human input through various methodologies and tools. Spanning five modules, you will gain a comprehensive understanding of machine learning techniques, the role of human annotation in ML performance, and the principles and practices of crowdsourcing. The course covers key aspects of designing and implementing crowdsourced studies, calculating inter-annotator agreements, and leveraging crowdsourcing to enhance ML performance. Practical skills will be developed through hands-on activities using platforms like Amazon Mechanical Turk (AMT) and analyzing the data collected from such platforms.

In this module, you will be introduced to the fundamentals of machine learning (ML). You will learn the definition and principles of ML, and gain practical skills in calculating and comparing ML performance metrics. You will get a chance to understand how to construct ML classifiers and analyze their effectiveness across different algorithms. This module prepares you to apply ML techniques effectively in various domains, enhancing your ability to solve complex problems using data-driven approaches.

In this module, you will explore the significance of IAA in Machine Learning (ML) performance. You will learn to calculate IAA manually and implement Krippendorf’s Alpha using the software. You will gain insights into how IAA impacts the reliability of annotated data and its implications for ML model training. This module equips you with essential skills to ensure consistency and reliability in data annotation processes, crucial for effective ML applications.

In this module, you will be introduced to the concept and practical applications of crowdsourcing. You will get a chance to learn how crowdsourcing enhances problem-solving through collective efforts and explore real-world use cases. You will be able to establish your first Amazon Mechanical Turk (AMT) account and understand the platform's capabilities for executing crowdsourced tasks. You will get a chance to delve into crowdsourcing design principles to optimize task efficiency and reliability. This module prepares you to leverage crowdsourcing effectively for diverse applications, from data annotation to research experiments.

This module focuses on leveraging Amazon Mechanical Turk (AMT) for crowdsourcing studies. You will learn to design effective experiments using AMT, ensuring optimal task design and participant engagement. You will be able to collect data through AMT and perform initial analyses to derive meaningful insights from crowdsourced data. You will also understand the implications of AMT addiction and ethical considerations in platform-based research. This module equips you with practical skills to conduct reliable and insightful crowdsourcing studies using AMT.

This module explores the intersection of crowdsourcing and ML performance enhancement. You will be able to evaluate how Inter-Annotator Agreement (IAA) affects ML model reliability and accuracy. You will explore case studies such as COVID test kit distribution and organ transplant matching to understand real-world applications. You will learn to optimize ML performance through effective crowdsourcing design, ensuring data quality and reliability in machine learning applications.