(´º½ºÅë½Å=±è»ó¼· ±âÀÚ) ÀÎõ´ë Á¤º¸±â¼ú´ëÇпø ¹éÀç¿ë, ÄÄÇ»ÅÍ°øÇкΠ2Çгâ À¯¿ë»ó ÇлýÀÌ Á¤º¸±â¼ú´ëÇпø°³¿ø ÀÌ·¡ ÃÖÃÊ·Î SCI±Þ ³í¹®À» °ÔÀçÇß´Ù.
21ÀÏ ÀÎõ´ë¿¡ µû¸£¸é ¹éÀç¿ë, À¯¿ë»ó µÎÇлý(ÄÄÇ»ÅͺñÀü¿¬±¸½Ç ¼Ò¼Ó, Áöµµ±³¼ö ¹è½Âȯ)Àº Áö³´Þ ¡®Adversarial Learning With Knowledge of
Image Classification for Improving GANs¡¯Á¦¸ñÀÇ SCIE ³í¹®À» IEEE Access(IF:3.557, Q1) 1ÀúÀÚ, 2ÀúÀÚ·Î °ÔÀçÇß´Ù.
ƯÈ÷, ¹éÀç¿ë ÇлýÀº 1996³â 11¿ù Á¤º¸±â¼ú´ëÇпø°³¿ø ÀÌ·¡ Á¤º¸±â¼ú´ëÇпø»ý ÃÖÃÊ·Î SCI±Þ ³í¹®À» °ÔÀçÇÏ´Â ¼º°ú¸¦ ÀÌ·ð´Ù.
¹éÀç¿ë, À¯¿ë»ó ÇлýÀº transfer learningÀ» ÅëÇØ GANÀÇ ¼º´ÉÀ» Çâ»ó½ÃÅ°±â À§ÇØ image classification networkÀÇ knowledge¸¦ ÀÌ¿ëÇÏ´Â ¸ðµ¨À» °³¹ßÇß´Ù.
ÀÌÁ¾ ºÐº° ³×Æ®¿öÅ©(Heterogeneous discriminative networks)°áÇսà ¹ß»ýÇÏ´Â excessive negative feedbackÀ» ±Ô¸íÇß´Ù.
¶Ç, GANÇнÀ°úÁ¤ Áß À̸¦ È¿°úÀûÀ¸·Î ¿ÏÈÇϱâ À§ÇØincremental learning°ú »õ·Î¿î GAN energy functionÀ» Á¦¾ÈÇß´Ù.
GAN¿¡¼ Á¦¾ÈÇÑ ¹æ¹ýÀ» Àû¿ëÇÒ ¶§, ´Ù¸¥ multiple discriminator¸¦ ÀÌ¿ëÇÏ´Â GANÀ̳ª standard CNNÀ» ÀÌ¿ëÇÏ´Â GAN¿¡ ºñÇØ ¿ì¼öÇÑ À̹ÌÁö »ý¼º ¼º´ÉÀ» È®ÀÎÇÒ ¼ö ÀÖ¾ú´Ù.