💥 NeuroSYNC Webinar Series — Learning Nonlinear Heterogeneity in Physical Kolmogorov-Arnold Networks

📌 Physical neural networks usually keep device nonlinearities fixed and train only the linear weights. Jack's talk turns that on its head: by training the synaptic nonlinearity itself within a Kolmogorov-Arnold framework, his group achieves markedly higher task performance per physical resource — with network sizes up to 40–500x smaller than conventional linear-weight approaches.

🔎 The work is demonstrated on silicon-on-insulator chips running at room temperature, and trained without backpropagation or surrogate device models, using a forward-only, noise-based perturbative learning method that can match or exceed backpropagation's performance.

📅 Thursday 9 July 2026

🕐 1:00–2:00pm (UK time)

📍 Online — https://lnkd.in/eFQA9NYV

Register here: https://lnkd.in/eFQA9NYV

@PhysicsStation