Á¤º¸°úÇÐȸ ÄÄÇ»ÆÃÀÇ ½ÇÁ¦ ³í¹®Áö (KIISE Transactions on Computing Practices)
Current Result Document : 247 / 247
ÇѱÛÁ¦¸ñ(Korean Title) |
ºí·ÏüÀÎ ±â¹ÝÀÇ ºÐ»êÇü µö·¯´× ½Ã½ºÅÛ ¼³°è ¹× ±¸Çö |
¿µ¹®Á¦¸ñ(English Title) |
Design and Implementation of Blockchain-Based Distributed Deep Learning System |
ÀúÀÚ(Author) |
°Á¤È£
±è¹Î¼ö
Á¤±¤¼ö
Jeongho Kang
Minsu Kim
Kwangsue Chung
±èÀç¼®
½ÅÁø¸í
Ȳµµ¿¬
ÃÖÀ±È£
Jae-Seok Kim
Jinmyeong Shin
Do-Yeon Hwang
Yoon-Ho Choi
ÃßÇöÁö
±è¿¬Èñ
ÃÖº´ÁÖ
Hyun-Ji Chu
Yeonhee Kim
Byoungju Choi
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 27 NO. 08 PP. 0362 ~ 0371 (2021. 08) |
Çѱ۳»¿ë (Korean Abstract) |
´Ù¾çÇÑ ÇüÅÂÀÇ, ±×¸®°í ´Ù·®ÀÇ µ¥ÀÌÅÍ°¡ Á¡Â÷ ÃàÀûµÇ¾î°¡¸é¼ À̸¦ Çؼ®ÇÏ°í È°¿ëÇϱâ À§ÇÑ µö·¯´× ÇнÀÀÇ Á߿伺ÀÌ ´õ¿í Ä¿Á³´Ù. µö·¯´× ÇнÀ ½Ã ÇÊ¿äÇÑ ¿¬»êÀÌ Á¡Â÷ º¹ÀâÇØÁü¿¡ µû¶ó ÀÌ¿¡ ÇÊ¿äÇÑ ÄÄÇ»Æà ´É·ÂÀ» ºÐ»êÇÏ¿© ó¸®ÇÏ°íÀÚ ÇÏ´Â ºÐ»êÇü µö·¯´× ½Ã½ºÅÛÀÌ µîÀåÇÏ°Ô µÇ¾ú°í, À̶§ °¢ ÇнÀ ³ëµå °£¿¡ ÇнÀ °á°úÀÎ °¡ÁßÄ¡ µ¥ÀÌÅ͸¦ ±³È¯ÇÏ´Â ¹æ¹ý¿¡ ´ëÇÑ ³íÀÇ°¡ ´Ù¾çÇÏ°Ô µÇ¾î¿Ô´Ù. ±×·¯³ª ±âÁ¸¿¡ Á¦½ÃµÇ¾î¿Ô´ø °¡ÁßÄ¡ ±³È¯ ¹æ½ÄÀº Á¦3 °ü¸®ÀÚ¸¦ ÇÊ¿ä·Î ÇÏ´Â Áß¾Ó ÁýÁßÇü ¹æ½ÄÀ̰ųª, ȤÀº °¢ ³ëµå°£ µ¥ÀÌÅÍ ¹«°á¼ºÀÇ Ä§ÇØ À§ÇèÀÌ ÀÜÀçÇÏ´Â P2P ±â¹ÝÀÇ ¹æ½ÄÀ̾ú´Ù. ÀÌ¿¡ º» ¿¬±¸¿¡¼´Â P2P ȯ°æ¿¡¼µµ µ¥ÀÌÅÍÀÇ ¹«°á¼ºÀ» º¸ÀåÇÏ¸é¼ °¡ÁßÄ¡ µ¥ÀÌÅ͸¦ ±³È¯ÇÒ ¼ö ÀÖ´Â ºí·ÏüÀÎ ±â¹ÝÀÇ ºÐ»êÇü µö·¯´× ½Ã½ºÅÛ Chainlearn À» Á¦¾ÈÇϸç, ÀÌ¿¡ ´ëÇÑ È°¿ë »ç·Ê¸¦ º¸ÀÌ°í ±âÁ¸ ½Ã½ºÅÛ°ú ºñ±³ÇÑ Chainlearn¸¸ÀÇ Æ¯ÀåÁ¡À» ºÐ¼®ÇÑ´Ù. |
¿µ¹®³»¿ë (English Abstract) |
With gradual accumulation of various forms and large quantities of data, the importance of deep learning process to interpret and utilize them has increased. The increasing complexity of computations required for deep learning has led to the emergence of distributed deep learning systems to deal with the computing power in part, resulting in discussions on how to exchange weighted data, the result of training between learning nodes. Previously proposed weight exchange methods were either centralized, requiring third-party administrators, or P2P-based. They show a risk of infringing data integrity between nodes. Thus, in this work, we proposed a blockchain-based distributed deep learning system called Chainlearn. It allows for exchanging weighted data while ensuring data integrity even in P2P environments. This study also used cases and analyzed characteristics of Chainlearn compared to existing systems. |
Å°¿öµå(Keyword) |
¸ð¹ÙÀÏ ´Ü¸»
À¯ÇØ µ¿¿µ»ó
½ºÆ®¸®¹Ö Á¦¾î
À¯Çؼº º¯È
À¯»çµµ ºñ±³
mobile device
obscene video
streaming control
obscenity change
similarity comparison
µö·¯´×
StyleGAN
IndomainGAN
À̹ÌÁö ÇÕ¼º
½Ã¸Çƽ È®»ê
deep learning
StyleGAN
IndomainGAN
image synthesis
semantic diffusion
ºí·ÏüÀÎ
ºÐ»ê ½Ã½ºÅÛ
µö·¯´×
P2P
blockchain
distributed system
deep learning
P2P
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|