Abstract:  SMachine learning methods typically use gradient descent – a centralized, top-down algorithm – to optimally modify every parameter. In this talk I'll discuss our recently realized electronic learning metamaterials that perform machine learning differently, in an entirely bottom-up manner (without help from a processor). Each element of our system follows local update rules, and global learning emerges from these dynamics. This is a feature shared with the brain, albeit with different rules and dynamics. I'll discuss the construction and operation of these systems, the breadth of tasks they can accomplish even within a single architecture choice, and similarities to biological systems including the brain. Further, I'll show that the system learns complex, nonlinear tasks in a predictable sequence, by lowering polynomial modes of the error. This ordering persists regardless of the relative sizes of these modes. Our system trains in seconds and performs learned tasks in microseconds, dissipating picojoules of power across each element. Future versions have enormous potential to be faster and more efficient than state-of-the-art machine learning solutions, while providing additional benefits like robustness to manufacturing defects, similar to how biological systems can endure damage but retain functionality.