Beyond Tech - Machine Learning in Science & Policy - Professor David Dunson, Duke

Explore the challenges and opportunities of applying machine learning beyond the tech industry in this 58-minute lecture by Professor David Dunson from Duke University. Delve into the fundamental differences between tech applications and those in science and policy-making, examining why popular algorithms like deep learning may fail in other domains. Learn about the potential pitfalls of naively applying off-the-shelf ML algorithms in criminal justice, neuroscience, policy-making, and healthcare. Discover the importance of developing targeted methods that address selection bias, uncertainty quantification, limited training data, and complex observations. Focus on two specific problems: removing sensitive variable influence for fair predictive algorithms and creating interpretable models of human traits based on brain connection structure. Gain insights into Bayesian statistical theory, dimensionality reduction, and nonparametric approaches for high-dimensional and complex data across various disciplines.

Intro
What is machine learning?
Some examples of labeled data
Mimicking automating humans?
Self-driving cars
What creates critical problems for deep learning?
Dealing with the data deluge in science
ML in policy & automated decision making
A geometric solution
Application to creativity