Statistical Theory for Deep Learning
POSTECH »ê¾÷°æ¿µ°øÇаú ä¹Î¿ì ±³¼ö
º» °¿¬¿¡¼´Â ½ÉÃþ½Å°æ¸Á ¸ðÇüÀÇ ¶Ù¾î³ ¼º´ÉÀ» ¼³¸íÇϱâ À§ÇÑ ÃֽŠÅë°èÇÐ À̷еéÀ» ¼Ò°³ÇÑ´Ù. ºñ¸ð¼ö ÇÔ¼ö ÃßÁ¤ °üÁ¡¿¡¼ ½ÉÃþ½Å°æ¸Á ¸ðÇüÀÇ ÇÔ¼ö ±Ù»ç ÀÌ·Ð, ȸ±Í ÇÔ¼ö ¹× ºÐÆ÷ ÃßÁ¤ ÀÌ·Ð µîÀ» ´Ù·é´Ù.
Domain Adaptation and Generalization under Distribution Shifts
KAIST ¼ö¸®°úÇаú ÇÏ¿ì¼® ±³¼ö
µµ¸ÞÀÎ °£ µ¥ÀÌÅÍÀÇ ºÐÆ÷°¡ ´Ù¸¥ °æ¿ì ¸ðµ¨ÀÇ ¼º´ÉÀ» À¯ÁöÇϱâ À§ÇÑ domain adaptation°ú domain generalizationÀÇ ±âº»ÀûÀÎ ¾Ë°í¸®Áò°ú ÃֽŠ¿¬±¸¸¦ ¼Ò°³ÇÑ´Ù. ÇöÀç domain adaptationÀÇ ÀÌ·ÐÀû ¹ßÀü°ú ÇÑ°èÁ¡, ±×¸®°í ¾Ë°í¸®ÁòÀÇ ÀϹÝȸ¦ À§ÇÑ µ¥ÀÌÅÍÀÇ Åë°èÀû Àΰú°ü°èÀÇ Á߿伺¿¡ ´ëÇØ ³íÀÇÇÑ´Ù.
Theoretical Understanding of Foundation Models
¿¬¼¼´ëÇб³ ÀÀ¿ëÅë°èÇаú ¼ÕÁö¿ë ±³¼ö
ÃÖ±Ù ChatGPT, Stable DiffusionÀ» Æ÷ÇÔÇÑ ´Ù¾çÇÑ Transformer ±â¹ÝÀÇ Foundation ModelµéÀÌ ³î¶ó¿î ¼º´ÉÀ» º¸ÀÌ°í ÀÖ´Ù. º» °¿¬¿¡¼´Â, Foundation ModelÀ» ÀÌ·ÐÀûÀ¸·Î ºÐ¼®ÇÏ´Â ´Ù¾çÇÑ ³í¹®µé¿¡ ³ª¿Â ¼ö½ÄÀ» ¸®ºäÇÏ°í, ÇâÈÄ ¿¬±¸ ¹ßÀü ¹æÇâ¿¡ ´ëÇØ ³íÀÇÇÑ´Ù.
Brief Overview of Theoretical Deep Learning: Some Classic and Modern Topics
°í·Á´ëÇб³ ¼öÇаú À̵¿Çå ±³¼ö
µö·¯´×ÀÇ ÀÌ·ÐÀû ±Ù°£À» ÀÌ·ç´Â universal approximation theoremºÎÅÍ, ÀÌ·ÐÀû ±â°èÇнÀÀÇ Å¬·¡½Ä ÅäÇÈÀÎ complexity theory¿Í error analysis, ±×¸®°í ÃÖ±Ù µö·¯´× ÀÌ·Ð ¿¬±¸¿¡¼ ºÎ°¢µÈ neural tangent kernel¿¡ ´ëÇØ °£·«È÷ »ìÆ캻´Ù.