Surya Ganguli
@SuryaGanguli

Associate Professor of Applied Physics at Stanford, with courtesy appointments in Computer Science, Electrical Engineering and Neurobiology.




Surya Ganguli    @SuryaGanguli
Great thread by @KuninDaniel on our new preprint "Rethinking the limiting dynamics of SGD: modified loss, phase space oscillations, and anomalous diffusion." https://t.co/JffAMLVhsv on how the loss landscape, batch size, learning rate and momentum yield anomalous diffusion

Surya Ganguli    @SuryaGanguli
1/ Our new work: "How many degrees of freedom do we need to train deep networks: a loss landscape perspective." https://t.co/O7UKkSOO63 We present a geometric theory that connects to lottery tickets and a new method: lottery subspaces. w/ @_BrettLarsen @caenopy @stanislavfort

Surya Ganguli    @SuryaGanguli
A great series #SciML series with lots of deep learning and statistical mechanics!

Surya Ganguli    @SuryaGanguli
1/ Our work w/ Schnitzer lab in @CellCellPress reveals and models a synchronization phase transition in movement initiation, that strengthens with skill learning, yielding millisecond cerebellar synch over a millimeter spatial scale! Lead by @Mark_J_Wagner https://t.co/UjgGqVIdot











 








Surya Ganguli

Simone Scardapane

Aran Komatsuzaki

Sergey Levine

Dan Roberts

Nils Reimers