Notes

Research notes and interactive explanations.

SoftServe: A scalable quasi-Newton method for deep learning

October 2026

How a soft secant constraint makes structured quasi-Newton updates possible. An interactive introduction to SoftServe, with a geometry demo and results from recurrent networks, autoencoders, and physics-informed models.