µ¥ÀÌÅͺ£À̽º ¿¬±¸È¸Áö(SIGDB)
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
·©Å· µ¥ÀÌŸ¸¦ È°¿ëÇÑ È¸±ÍºÐ¼® ¼º´É °³¼± |
¿µ¹®Á¦¸ñ(English Title) |
Can Ranking Data Help Regression? |
ÀúÀÚ(Author) |
³ª½ÂÈÆ
±è¼ºÃ¶
À¯È¯Á¶
Seunghoon Na
Sungchul Kim
Hwanjo Yu
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 26 NO. 01 PP. 0061 ~ 0070 (2010. 04) |
Çѱ۳»¿ë (Korean Abstract) |
º» ³í¹®¿¡¼´Â ±âÁ¸ÀÇ È¸±ÍºÐ¼® ¹®Á¦¿¡ ÀÖ¾î¼ È¸±ÍºÐ¼® µ¥ÀÌÅÍ ÀÌ¿Ü¿¡ Ãß°¡ÀûÀÎ ·©Å· µ¥ÀÌÅÍ°¡ Á¸ÀçÇÒ °æ¿ì¿¡ µÎ µ¥ÀÌÅ͸¦ ¸ðµÎ È°¿ëÇÏ¿© ȸ±ÍºÐ¼®ÀÇ ¼º´ÉÀ» °³¼±½ÃÅ°´Â »õ·Î¿î ÇüÅÂÀÇ ÇнÀ¹ýÀ» Á¦¾ÈÇÑ´Ù. RankSVRÀº ÀÌ ³í¹®¿¡¼ Á¦¾ÈÇÏ´Â »õ·Î¿î ÇüÅÂÀÇ È¸±ÍºÐ¼® ±â¹ýÀ¸·Î, ȸ±ÍºÐ¼® ¹®Á¦¿¡¼ °¡Àå ³Î¸® »ç¿ëµÇ´Â SVR(Support Vector Regression)ÀÇ È¸±ÍºÐ¼® ÇÔ¼ö¸¦ ȸ±ÍºÐ¼® µ¥ÀÌÅÍ¿Í ·©Å· µ¥ÀÌÅ͸¦ ¸ðµÎ »ç¿ëÇÒ ¼ö ÀÖµµ·Ï º¯ÇüÇÑ »õ·Î¿î ÇüÅÂÀÇ margin maximization ¾Ë°í¸®ÁòÀÌ´Ù. º» ³í¹®ÀÇ ½ÇÇè°á°ú´Â ¸î °³ÀÇ toy¿¹Á¦¿Í ½ÇÁ¦ µ¥ÀÌÅ͸¦ »ç¿ëÇÏ¿© ÀûÀýÇÑ ÆĶó¹ÌÅ͸¦ »ç¿ëÇÑ RankSVRÀÌ ±âÁ¸ÀÇ SVRº¸´Ù ´õ ÁÁÀº ¼º´ÉÀ» ³½´Ù´Â °ÍÀ» º¸¿©ÁØ´Ù. ·©Å· µ¥ÀÌÅÍ¿Í È¸±ÍºÐ¼® µ¥ÀÌÅ͸¦ ¸ðµÎ °í·ÁÇÏ´Â ¹®Á¦´Â ±× µ¿¾È °ÅÀÇ ´Ù·ç¾îÁöÁö ¾ÊÀº »õ·Î¿î ¹®Á¦ÀÌ°í, ÀÌ ¹®Á¦¸¦ max-margin ±â¹ýÀ» È°¿ëÇÏ¿© ÇØ°áÇÏ¿´´Ù´Â °ÍÀÌ º» ³í¹®ÀÇ °¡Àå Å« ¼º°ú¶ó°í ÇÒ ¼ö ÀÖ´Ù. |
¿µ¹®³»¿ë (English Abstract) |
In this paper, we consider a new type of supervised learning problem for regression when a large number of ranking data is available in addition to the original regression training data. We propose a new type of margin maximization algorithm, RankSVR, which is a generalization of SVR(Support Vector Regression), that learns a regression function from both regression and ranking data. Experimental results on toy examples and several real datasets show that RankSVR significantly improves SVR in the range of reasonable parameter settings. |
Å°¿öµå(Keyword) |
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|