Douglas Durian

I’m an experimentalist working on the statistical and nonlinear physics of widely-varied forms of soft matter. Until recently my focus was on particulate systems composed of densely packed colloids, grains, or bubbles. But my primary focus is shifting to a new class of systems we call “Contrastive Local Learning Networks (CLLNs).” Like dense particulate systems, CLLNs feature unusual dynamics in a complex high-dimensional landscape. Our CLLNs are metamaterials composed of many copies of a specially-designed repeat unit, arrayed in highly connected network, which are capable of learning complex AI-like functionalities autonomously using local learning rules – without the need for external memory or digital processor—similar to the brain. In effect, these are neural networks in which both training and inference are done automatically and in-memory by analog physical processes. As such, they potentially offer a tremendous advantage in terms of speed and energy efficiency over artificial neural networks at large scales. They also pose a host of fascinating new fundamental research questions at the intersection of physics, data science, and neuroscience. We are looking for motivated graduate students and postdocs to join our highly-collaborative interdisciplinary research team – please see the following papers/press and contact me for further information.

S. Dillavou, B. D. Beyer, M. Stern, A. J. Liu, M. Z. Miskin, D. J. Durian, Machine Learning Without a Processor: Emergent Learning in a Nonlinear Analog Network, Proc. Nat. Acad. Sci. 121, e2319718121 (2024). Press; press; commentary. S. Dillavou, B. Beyer, M. Stern, M. Miskin, A. Liu, D. Durian, Nonlinear classification without a processor, NeurIPS Workshop on Machine Learning with New Compute Paradigms (2023) M. Stern, S. Dillavou, D. Jayaraman, D. J. Durian, A. J. Liu, Training physical learning systems for power-efficient solutions, APL Machine Learning 2, 016114 (2024). L. E. Altman, M. Stern, A. J. Liu, D. J. Durian, Experimental Demonstration of Coupled Learning in Elastic Networks, Physical Review Applied 22, 024053 (2024).