Effective fluctuating continuum models for stochastic gradient descen [In: Flows on measure spaces and applications in machine learning ; 22 March - 27 March 2026 ; report no. 13/2026]
Introduce thermal noise before discretizing, not afterwards! [In: Flows on measure spaces and applications in machine learning ; 22 March - 27 March 2026 ; report no. 13/2026]
Large spikes in SGD : a large-deviations view of catapults [In: Modern and emerging phenomena in machine learning ; 8 March - 13 March 2026 ; report no. 11/2026]