Ȩ > Ç¥ÁØÈ °³¿ä > TTAÀÇ Ç¥ÁØÇöȲ
Ç¥ÁعøÈ£ | TTAK.KO-11.0290 | ±¸Ç¥ÁعøÈ£ | |
---|---|---|---|
Á¦°³Á¤ÀÏ | 2021-12-08 | ÃÑÆäÀÌÁö | 21 |
ÇѱÛÇ¥Áظí | ½Å°æ¸Á °¡¼Ó±â¸¦ žÀçÇÑ ÀÓº£µðµå ½Ã½ºÅÛ¿¡¼ È¿À²ÀûÀÎ ½Å°æ¸Á Ãß·ÐÀ» À§ÇÑ Áß°£ Ç¥Çö | ||
¿µ¹®Ç¥Áظí | Intermediate Representation for Efficient Neural Network Inference in an Embedded System Equipped with a Neural Network Accelerator | ||
Çѱ۳»¿ë¿ä¾à | ÀÌ Ç¥ÁØÀº ÀÚ¿ø »ç¿ëÀÌ Á¦ÇÑÀûÀÎ ÀÓº£µðµå ½Ã½ºÅÛ¿¡¼ ½Å°æ¸Á ¸ðµ¨À» È¿À²ÀûÀ¸·Î ½ÇÇàÇϱâ À§ÇÑ ÄÚµå »ý¼º¿¡ °üÇÑ ¹®Á¦¸¦ ÇØ°áÇϱâ À§ÇÑ ¹æ¹ýÀ» ´Ù·é´Ù. Á¦¾ÈÇÏ´Â ±â¼úÀº Ÿ±ê Çϵå¿þ¾î È¿À²ÀûÀÎ Äڵ带 »ý¼ºÇϱâ À§Çؼ Áß°£ Ç¥ÇöÀ» Á¤ÀÇÇÏ°í À̸¦ ÅëÇؼ °¢ ´Ü°è¿¡¼ ÃÖÀûÈ °úÁ¤À» ¼öÇàÇÑ´Ù. Á¤ÀÇÇÑ Áß°£ Ç¥ÇöÀº µÎ ´Ü°è·Î ±×·¡ÇÁ Áß°£ Ç¥Çö°ú ÅÙ¼ Áß°£ Ç¥ÇöÀÌ ÀÖ´Ù. | ||
¿µ¹®³»¿ë¿ä¾à | The standard discusses methods for solving the problem of code generation to efficiently execute neural network models in resource-constrained embedded systems. In order to generate the target hardware efficient code, the intermediate representation is defined and the optimization process is performed at each stage. The intermediate representation consists of two types: graph intermediate representation and tensor intermediate representation. | ||
±¹Á¦Ç¥ÁØ | |||
°ü·ÃÆÄÀÏ | TTAK.KO-11.0290.pdf |