Learn More
Many architects believe that major improvements in cost-energy-performance must now come from domain-specific hardware. This paper evaluates a custom ASIC---called a Tensor Processing Unit (TPU) --- deployed in datacenters since 2015 that accelerates the inference phase of neural networks (NN). The heart of the TPU is a 65,536 8-bit MAC matrix multiply unit(More)
Interior proximal methods for variational inequalities are, in fact, designed to handle problems on polyhedral convex sets or balls, only. Using a slightly modified concept of Bregman functions, we suggest an interior proximal method for solving variational inequalities (with maximal monotone operators) on convex, in general non-polyhedral sets, including(More)
For variational inequalities characterizing saddle points of Lagragians associated with convex programming problems in Hilbert spaces, the convergence of an interior proximal method based on Bregman distance func-tionals is studied. The convergence results admit a successive approximation of the varia-tional inequality and an inexact treatment of the(More)
In this paper we clarify that the interior proximal method developed in [6] (vol. 27 of this journal) for solving variational inequalities with monotone operators converges under essentially weaker conditions concerning the functions describing the " feasible " set as well as the operator of the variational inequality. References [1] A. Auslender and M.(More)
On the basis of three different experiments: oddball task (visual, auditory, and audio-visual stimuli), modified Sternberg's, and multistage memory tasks, it was shown that: a) there was not a single typical spectral pattern type that would characterize the majority of the trials; b) the total number of the different spectral pattern types was limited; c)(More)
In the present experimental study, we examined the compositions of electroencephalographic (EEG) brain oscillations and their percent ratio in 12 subjects during resting conditions (closed and open eyes) and during the memory task (waiting, encoding and keeping-in-mind stages). The exact compositions of brain oscillations and their percent ratio were(More)
The short-term structure of electroencephalogram (EEG) spectral transformations during different brain functional states (closed/opened eyes and memory task) was studied. It was shown that approximately 50% of spectral pattern (SP) types occur not more than 2-3 times per 149 analysis epochs in a 1-min EEG. The remaining 50% of SP types were the same for the(More)