Egor Shulgin
@egor_shulg
On the job market | PhD student @KAUST_News | ex-@Apple, @samsungresearch
Many many many thanks to the lecturers of the first week for their dedication (could only get some of them on this photo...) #MlssSenegal2025 @adjiboussodieng @mblondel_ml @neu_rips @Ashia__Wilson @natschluter @SeydaNgom @dohmatobelvis @egor_shulg @joof @eugene_ndiaye
Gluon is an LMO-based optimizer that unifies Muon and Scion, addressing two key gaps πͺπ― π΅π©π¦πͺπ³ π΅π©π¦π°π³πΊ: 1. Layer-wise updates (not full-model) β aligns with how models are trained in practice 2. New smoothness model β uses (Lβ, Lβ)-smoothness per layer, notβ¦
I am at ICML 2024 (@icmlconf) in Vienna. Come to our poster #1209 (today, 1:30pm-3pm, Hall C 4-9) on theory for Independent Subnetwork Training (IST). Led by my PhD student Egor Shulgin (@egor_shulg); who is also here this week!