Prof. Emo Welzl and Prof. Bernd Gärtner
|Mittagsseminar Talk Information|
Date and Time: Thursday, August 30, 2018, 12:15 pm
Duration: 30 minutes
Location: OAT S15/S16/S17
Speaker: Sebastian Stich (EPFL)
Nowadays machine learning applications require stochastic optimization algorithms that can be implemented on distributed systems. The communication overhead of the algorithms is a key bottleneck that hinders perfect scalability. Various recent works proposed to use quantization or sparsification techniques to reduce the amount of data that needs to be communicated, for instance by only sending the most significant entries of the stochastic gradient (top-k sparsification). Whilst this scheme shows good performance in practice it eluded theoretical analysis so far.
In this work we analyze a variant of Stochastic Gradient Descent (SGD) with k-sparsification (for instance top-k or random-k) and show that this scheme converges at the same rate as vanilla SGD. That is, the communication can be reduced by a factor of the dimension of the whilst still converging at the same rate.
Joint work with Jean-Baptiste Cordonnier and Martin Jaggi
Automatic MiSe System Software Version 1.4803M | admin login