Mathematical ML and ML for Math - AltGDmin for Secure Federated Low Rank Matrix Learning and ML-Enabled K-12 Math Support

Explore mathematical machine learning and ML applications for mathematics education in this comprehensive lecture covering Byzantine-resilient federated learning algorithms and K-12 math support systems. Learn about the AltGDmin (Alternating Gradient Descent and Minimization) algorithm and its distributed extensions for secure federated low-rank matrix learning problems, including matrix completion, column-wise sensing, and phase retrieval with applications in recommender systems, multi-task representation learning, accelerated dynamic MRI, and Fourier ptychography. Discover how Byz-AltGDmin provides provably Byzantine-resilient solutions for vertically federated learning scenarios where adversarial nodes can poison model updates and collude against the system. Understand the advantages of AltGDmin over traditional Alternating Minimization approaches in terms of speed and communication efficiency for partly-decoupled optimization problems. Examine real-world applications in dynamic MRI and explore the CyMath program's ML-enabled K-12 mathematics tutoring approach that addresses the critical need for arithmetic fluency as a foundation for advanced STEM learning. Gain insights into how machine learning tools like ALEKS and Khan Academy can make mathematics education more scalable and equitable, while considering the long-term impact of current K-12 educational policies on STEM workforce preparation and mathematical skills equity.