Eliciting Simple and Transparent Algorithms for Statistical Learning via Majorization-Minimization

Explore how Majorization-Minimization (MM) algorithms can create simple and transparent approaches to statistical learning in this 31-minute conference talk. Discover the theoretical foundations of MM optimization and learn how this framework generates intuitive algorithms for complex statistical problems. Examine practical applications including power k-means clustering, generalized linear model regression with distance-to-set penalties, proximal distance algorithms for likelihood-based sparse covariance estimation, and advanced MM techniques involving path following and trust regions. Gain insights into how MM methods bridge the gap between computational efficiency and algorithmic interpretability in modern statistical learning, with examples drawn from clustering, regression, and covariance estimation problems.