Mi­crosoft’s Project Brain­wave of­fers real-time AI

OpenSource For You - - Fossbytes -

Ex­pand­ing its foot­print in the ar­ti­fi­cial in­tel­li­gence (AI) world, Mi­crosoft has un­veiled a new deep learn­ing ac­cel­er­a­tion plat­form called Project Brain­wave. The new project uses a set of field-programmable gate ar­rays (FPGA) de­ployed in the Azure cloud to en­able a real-time AI ex­pe­ri­ence at a faster pace.

The sys­tem un­der Mi­crosoft’s Project Brain­wave is built on three main lay­ers: a high-per­for­mance dis­trib­uted sys­tem ar­chi­tec­ture, a DNN engine syn­the­sised on FPGAs, and a com­piler and run­time for low-fric­tion de­ploy­ments of trained mod­els. The ex­ten­sive work on FPGAs by the Redmond gi­ant en­ables high per­for­mance through Project Brain­wave. Ad­di­tion­ally, the sys­tem ar­chi­tec­ture as­sures low la­tency and high through­put.

One of the big­gest ad­van­tages of the new Mi­crosoft project is the speed. FPGAs on the sys­tem are at­tached di­rectly with the net­work fab­ric to en­sure the high­est pos­si­ble speed. The high through­put de­sign makes it eas­ier to cre­ate deep learn­ing ap­pli­ca­tions that can run in real-time.

“Our sys­tem, de­signed for real-time AI, can han­dle com­plex, mem­ory-in­ten­sive mod­els such as LSTMs, with­out us­ing batch­ing to juice through­put,” Mi­crosoft’s dis­tin­guished en­gi­neer Doug Burger wrote in a blog post.

It is worth not­ing that Project Brain­wave is quite sim­i­lar to Google’s Ten­sor Pro­cess­ing Unit. How­ever, Mi­crosoft’s hard­ware sup­ports all the ma­jor deep learn­ing sys­tems. There is na­tive sup­port for Mi­crosoft’s Cog­ni­tive Toolkit as well as Google’s Ten­sorFlow. Brain­wave can speed up the pre­dic­tions from ma­chine learn­ing mod­els.

Newspapers in English

Newspapers from India

© PressReader. All rights reserved.