Á¤º¸°úÇÐȸ³í¹®Áö (Journal of KIISE)
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
´õ Å« ±¸Á¶ÀÇ ±³»ç ¸ðµ¨À» È°¿ëÇÑ °øÁ¤ÇÑ Æ¯Â¡ º¤ÅÍ Áõ·ù ±â¹ý |
¿µ¹®Á¦¸ñ(English Title) |
Fair Feature Distillation Using Teacher Models of Larger Architecture |
ÀúÀÚ(Author) |
Á¤»ó¿ø
¹®Å¼·
Sangwon Jung
Taesup Moon
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 48 NO. 11 PP. 1176 ~ 1183 (2021. 11) |
Çѱ۳»¿ë (Korean Abstract) |
´Ù¾çÇÑ ºñÀü ¾îÇø®ÄÉÀ̼ǿ¡¼ ¾Ë°í¸®ÁòÀÇ °øÁ¤¼ºÀ» ´Þ¼ºÇÏ´Â °ÍÀº Áß¿äÇØÁö°í ÀÖ´Ù. MMD ±â¹Ý °øÁ¤ÇÑ Æ¯Â¡ Áõ·ù(MFD)¶ó´Â ÃֽŠ°øÁ¤¼º ±â¹ýÀº Maximum Mean Discrepancy (MMD) ¸¦ »ç¿ëÇÑ Æ¯Â¡ Áõ·ù ¹æ¹ýÀ» ÅëÇØ ±âÁ¸ ¹æ¹ýµé°ú ºñ±³ÇßÀ» ¶§ Á¤È®µµ¿Í °øÁ¤¼ºÀ» »ó´çÈ÷ °³¼±½ÃÄ×Áö¸¸, ±×µéÀº ±³»ç ¸ðµ¨ÀÇ ±¸Á¶°¡ Çлý ¸ðµ¨°ú °°À» ¶§¸¸ Àû¿ëµÉ ¼ö ÀÖ¾ú´Ù. º» ³í¹®¿¡¼´Â, MFD¸¦ ±â¹ÝÀ¸·Î, ´õ Å« ±¸Á¶¸¦ °¡Áø ±³»ç ¸ðµ¨¿¡¼ÀÇ Æ¯Â¡ Áõ·ù¸¦ ÅëÇØ ºÒ°øÁ¤ÇÑ ÆíÇ⼺À» ¿ÏÈÇϴ ü°èÀû Á¢±Ù¹ýÀÎ MFD-RÀ» Á¦¾ÈÇÑ´Ù. ±¤¹üÀ§ÇÑ ½ÇÇèÀ» ÅëÇØ ¿ì¸®´Â MFD-RÀÌ ´Ù¸¥ ±âÁØ ¹æ¹ýÀ̳ª MFD¿Í ºñ±³ÇßÀ» ¶§, ´õ Å« ±³»ç ¸ðµ¨À» »ç¿ëÇÏ´Â ÀÌÁ¡ÀÌ ÀÖ´Ù´Â °ÍÀ» º¸ÀδÙ. |
¿µ¹®³»¿ë (English Abstract) |
Achieving algorithmic fairness is becoming increasingly essential for various vision applications. Although a state-of-the-art fairness method, dubbed as MMD-based Fair feature Distillation (MFD), significantly improved accuracy and fairness via feature distillation based on Maximum Mean Discrepancy (MMD) compared to previous works, MFD could be limitedly applied into when a teacher model has the same architecture as a student model. In this paper, based on MFD, we propose a systematic approach that mitigates unfair biases via feature distillation of a teacher model of larger architecture, dubbed as MMD-based Fair feature Distillation with a regressor (MFD-R). Throughout the extensive experiments, we showed that our MFD-R benefits from the use of the larger teacher compared to MFD as well as other baseline methods. |
Å°¿öµå(Keyword) |
°øÁ¤¼º
ÆíÇ⼺
Áö½Ä Áõ·ù
½Ã°¢Àû ÀνÄ
fairness
bias
knowledge distillation
visual recognition
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|