Á¤º¸°úÇÐȸ ³í¹®Áö B : ¼ÒÇÁÆ®¿þ¾î ¹× ÀÀ¿ë
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
¿Â¶óÀÎ ÇʱâÀνÄÀ» À§ÇÑ Áõ°¡ÇÏ´Â µ¥ÀÌÅ͸¦ ÀÌ¿ëÇÑ ¾Ó»óºí ±â¹ý |
¿µ¹®Á¦¸ñ(English Title) |
Ensemble Methods with Increasing Data for Online Handwriting Recognition |
ÀúÀÚ(Author) |
±èÅÂÁØ
ÀåÇÏ¿µ
¹ÚÁ¤¿Ï
Ȳ¼ºÅÃ
À庴Ź
Tae-Jun Kim
Ha-Young Jang
Jeongwan Park
Seongtack Hwang
Byoung-Tak Zhang
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 41 NO. 02 PP. 0164 ~ 0170 (2014. 02) |
Çѱ۳»¿ë (Korean Abstract) |
¸ð¹ÙÀÏ ±â±âÀÇ ´ëÁßÈ¿Í ÇÔ²² Çʱâü ÀνÄÀÇ Á߿伺Àº ´õ¿í Ä¿Áö°í ÀÖ´Ù. Çʱ⠵¥ÀÌÅÍ´Â µ¥ÀÌÅÍ¿¡ Á¸ÀçÇÏ´Â ºÐ»ê(variance)ÀÌ ¸Å¿ì Å©±â ¶§¹®¿¡ µ¥ÀÌÅÍ°¡ Áõ°¡ÇÔ¿¡ µû¶ó ¹®Á¦ÀÇ º¹Àâµµ°¡ ±Þ°ÝÈ÷ Áõ°¡Çϴ Ư¼ºÀÌ ÀÖ´Ù. ÀÌ·¯ÇÑ Æ¯¼ºÀ¸·Î ÀÎÇÏ¿© ´ë¿ë·®ÀÇ µ¥ÀÌÅ͸¦ ÀÌ¿ëÇÏ¿© Àνı⸦ ÇнÀ½ÃÅ°±âµµ ¾î·Æ°í ÇнÀ½Ã°£µµ ±æ¾îÁø´Ù´Â ¹®Á¦Á¡ÀÌ ÀÖ´Ù. º» ³í¹®¿¡¼´Â ÀÌ·¯ÇÑ ¹®Á¦Á¡µéÀ» ÇØ°áÇϱâ À§ÇÑ ¾Ó»óºí ±â¹ýÀ» Á¦½ÃÇÏ¿´´Ù. Á¦¾ÈÇÑ ¹æ¹ý·ÐÀº ¸ð¹ÙÀÏ ±â±â¸¦ ÅëÇؼ ÃàÀûµÇ´Â Çʱ⠵¥ÀÌÅ͸¦ È¿À²ÀûÀ¸·Î ÀÌ¿ëÇϱâ À§ÇÏ¿© ÀÏÁ¤·®ÀÇ µ¥ÀÌÅÍ°¡ ¸ðÀÏ ¶§¸¶´Ù »õ·Î¿î ¾àºÐ·ù±â(weak learner)¸¦ Ãß°¡ÇÔÀ¸·Î½á ¾Ó»óºí ¸ðµ¨À» ±¸ÃàÇÑ´Ù. Çʱâü ÀνÄÀ» À§Çؼ ¸¹ÀÌ »ç¿ëµÇ´Â Àΰø½Å°æ¸ÁÀº Çʱ⠵¥ÀÌÅÍÀÇ Å©±â°¡ Ä¿Áü¿¡ µû¶ó¼ µ¥ÀÌÅÍ ³»ÀÇ ºÐ»êµµ °°ÀÌ Ä¿Áö´Â ¹®Á¦·Î ÀÎÇÏ¿© ÇнÀ ½Ã°£ÀÌ ±Þ°ÝÈ÷ Áõ°¡ÇÏ°Ô µÇ´Âµ¥ ¾Ó»óºí ±â¹ýÀ» ÀÌ¿ëÇÑ Á¡ÁøÀû ÇнÀÀ» ÅëÇؼ ºü¸¥ ½Ã°£ ¾È¿¡ º¸´Ù È¿À²ÀûÀÎ ÇнÀÀÌ °¡´ÉÇÏ°Ô µÈ´Ù.
|
¿µ¹®³»¿ë (English Abstract) |
Today¡¯s mobile devices offer more intuitive touch interfaces and support handwriting recognition. The within class variance of handwritten characters causes the growth of complexity in handwriting recognition as data grows. Because of it, more time and efforts are required to train the recognizer as data grows. We propose the ensemble method with batch incremental learning. The proposed method adds the new weak learner to ensemble model, when incremental data reaches certain amount. The ensemble method with batch incremental learning reduces the training time of artificial neural network with large data set. It also tends to cancel out overfitting problem caused by high variance.
|
Å°¿öµå(Keyword) |
¾Ó»óºí ¸ðµ¨
¹è±ë
¿Â¶óÀÎ Çʱâü ÀνÄ
¿Â¶óÀÎ ÇнÀ
Á¡ÁøÀû ÇнÀ
ensemble model
bagging
online handwriting recognition
online learning
incremental learning
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|