Online dynamical learning and sequence memory with neuromorphic nanowire networks

Nanowire Networks (NWNs) are neuromorphic systems whose nanowire–nanowire junctions exhibit synapse-like memristive switching. Prior work showed NWNs can generate neuromorphic dynamics and support temporal learning. This study demonstrates online learning from spatiotemporal dynamical features using an NWN device for MNIST digit classification and a novel sequence memory recall task. For MNIST, online dynamical learning achieves 93.4% accuracy and reveals a correlation between per-class accuracy and mutual information (MI). For sequence memory, spatiotemporal memory patterns embedded in device dynamics enable online learning and recall of a target digit from a semi-repetitive sequence. These results provide proof-of-concept of online learning from NWN dynamics and elucidate how memory enhances learning.

Ruomin Zhu, Sam Lilak, Alon Loeffler, Joseph Lizier, Adam Stieg, James Gimzewski, Zdenka Kuncic