Ç¥ÁØÈ­ Âü¿©¾È³»

TTAÀÇ Ç¥ÁØÇöȲ

Ȩ > Ç¥ÁØÈ­ °³¿ä > TTAÀÇ Ç¥ÁØÇöȲ

Ç¥ÁعøÈ£ TTAK.KO-11.0244 ±¸Ç¥ÁعøÈ£
Á¦°³Á¤ÀÏ 2018-12-19 ÃÑÆäÀÌÁö 16
ÇѱÛÇ¥Áظí ÀÓº£µðµå ½Ã½ºÅÛ¿¡¼­ OpenCL ±â¹Ý Äܺ¼·ç¼Ç ½Å°æ¸Á µ¥ÀÌÅÍ ÀÎÅÍÆäÀ̽º
¿µ¹®Ç¥Áظí Data Interface for Convolutional Neural Network based on OpenCL in Embedded Systems
Çѱ۳»¿ë¿ä¾à ÀÌ Ç¥ÁØÀº ÀÓº£µðµå ½Ã½ºÅÛ¿¡¼­ OpenCL ±â¹Ý Äܺ¼·ç¼Ç ½Å°æ¸Á µ¥ÀÌÅÍ ÀÎÅÍÆäÀ̽º Ç¥ÁØÀ¸·Î ÀÓº£µðµå ½Ã½ºÅÛ Æ¯Â¡À» °í·ÁÇÑ Äܺ¼·ç¼Ç ½Å°æ¸Á µ¥ÀÌÅÍ ±¸Á¶¿Í À̸¦ À§ÇÑ »ç¿ëÀÚ ÀÎÅÍÆäÀ̽º¸¦ Á¤ÀÇÇÑ´Ù. º» Ç¥ÁØÀº ´Ù¾çÇÑ ÀÓº£µðµå ½Ã½ºÅÛ¿¡¼­ µ¿ÀÛ °¡´ÉÇϵµ·Ï OpenCLÀ» ±â¹ÝÀ¸·Î ÇÑ´Ù. OpenCLÀº °³¹æÇü ¹ü¿ë º´·Ä ÄÄÇ»Æà ÇÁ·¹ÀÓ¿öÅ©·Î½á ´Ù¾çÇÑ Çϵå¿þ¾î Á¦Á¶¾÷ü¿¡¼­ Á¦Á¶ÇÑ °¢°¢ÀÇ Çϵå¿þ¾î¿¡¼­ µ¿ÀÛ °¡´ÉÇÏ´Ù.
ÀÌ Ç¥ÁØ¿¡¼­ ±â¼úÇÏ´Â µ¥ÀÌÅÍ ±¸Á¶ ¹× »ç¿ëÀÚ ÀÎÅÍÆäÀ̽º´Â Äܺ¼·ç¼Ç ½Å°æ¸ÁÀÇ °¢ °èÃþ¿¡¼­ »ç¿ëµÇ´Â µ¥ÀÌÅ͸¦ °øÅëÀûÀ¸·Î °ü¸®Çϱâ À§ÇÑ Ç¥ÁØÈ­µÈ ±¸Á¶¸¦ Á¦°øÇÑ´Ù. ¶ÇÇÑ ÅëÇÕ ¸Þ¸ð¸® ±¸Á¶¿Í °°Àº ÀÓº£µðµå ½Ã½ºÅÛÀÇ Çϵå¿þ¾î Ư¡À» È°¿ëÇÏ¿© µ¥ÀÌÅÍ º¹»ç ºñ¿ë°ú ¸Þ¸ð¸® »ç¿ë·®À» ÃÖ¼ÒÈ­ÇÑ´Ù.
¿µ¹®³»¿ë¿ä¾à The standard defines a data structure and an user interface which consider the characteristics of embedded system based on OpenCL. It is based on OpenCL for operation in various embedded systems. OpenCL is an open and general-purpose parallel computing framework and may run on a variety of hardware manufactured by various hardware vendors.
The data structure and user interface described in the standard provide a standardized structure for commonly managing the data used in each layer of the convolutional neural network. It also utilizes the hardware features of embedded systems such as an integrated memory architecture to minimize data copying costs and memory usage.
±¹Á¦Ç¥ÁØ
°ü·ÃÆÄÀÏ TTAK.KO-11.0244.pdf TTAK.KO-11.0244.pdf            

ÀÌÀü
ÀÓº£µðµå µð¹ÙÀ̽º¿¡¼­ÀÇ Àΰø½Å°æ¸Á ¸ðµ¨ÀÇ »óÈ£ ¿î¿ëÀ» À§ÇÑ ¸®Æ÷ÁöÅ丮ÀÇ ¿ä±¸»çÇ×
´ÙÀ½
°ø°³¼ÒÇÁÆ®¿þ¾î ¶óÀ̼±½º Á¤Ã¥ ¼ö¸³ ¸ðµ¨