Title :
Distributed stochastic optimization and learning
Author :
Shamir, Ohad ; Srebro, Nathan
Author_Institution :
Dept. of Comput. Sci. & Appl. Math., Weizmann Inst. of Sci., Rehovot, Israel
fDate :
Sept. 30 2014-Oct. 3 2014
Abstract :
We consider the problem of distributed stochastic optimization, where each of several machines has access to samples from the same source distribution, and the goal is to jointly optimize the expected objective w.r.t. the source distribution, minimizing: (1) overall runtime; (2) communication costs; (3) number of samples used. We study this problem systematically, highlighting fundamental limitations, and differences versus distributed consensus problems where each machine has a different, independent, objective. We show how the best known guarantees are obtained by an accelerated mini-batched SGD approach, and contrast the runtime and sample costs of the approach with those of other distributed optimization algorithms.
Keywords :
communication complexity; distributed algorithms; gradient methods; learning (artificial intelligence); optimisation; accelerated mini-batched SGD approach; communication cost minimization; distributed consensus problems; distributed stochastic optimization algorithms; learning; overall runtime minimization; source distribution; stochastic gradient descent; Approximation methods; Complexity theory; Optimization; Presses; Runtime; Stochastic processes; Vectors;
Conference_Titel :
Communication, Control, and Computing (Allerton), 2014 52nd Annual Allerton Conference on
Conference_Location :
Monticello, IL
DOI :
10.1109/ALLERTON.2014.7028543