±¨ ¸æ ÈË£ºÀîÓî·å ¸±½ÌÊÚ
£¬ÄϾ©´óѧ

»ã±¨¹¦·ò£º12ÔÂ8ÈÕ£¨ÖÜÈÕ£©10:20¡«11:10

»ã±¨µØÖ·£ºÐ£±¾²¿ÀÖºõÐÂÂ¥2ºÅÂ¥¶þ²ã
£¬Ñ§º£Ìü

Ñû Çë ÈË£ºÔÀÏþ¶¬ ¸±½ÌÊÚ

»ã±¨ÌáÒª£º

Weakly supervised data is widely existed in real-world scenarios. Unlike supervised learning, which has achieved relatively mature solutions, weakly supervised learning is far from mature compared to supervised learning. Although various weakly supervised learning techniques have been applied in the industry, they lack clear guidelines to model the data due to the quality of raw data, so as to require lots of human intelligence involved into the modeling processes, which turn out to be not so reliable and smart. This talk first tries to figure out a couple of conditions that would be useful for deriving reliable weakly supervised learning. We then point out some possible conditions or resources that may be useful to derive reliable weakly supervised learning in future; finally we highlight some challenges remained in several complicated yet realistic weakly supervised learning scenarios.

»ã±¨È˼ò½é:

ÀîÓî·å
£¬±ðÀëÓÚ2006ÄêºÍ2013ÄêÔÚÄϾ©´óѧÍÆËã»ú¿Æѧϵ»ñѧʿºÍ²©Ê¿Ñ§Î»¡£2013Äê½øÈëÄϾ©´óѧÍÆËã»ú¿ÆѧÓë¼¼ÊõϵÈÎÖúÀí×êÑÐÔ±
£¬ÏÖÈÎÈí¼þм¼Êõ¹ú¶È³Áµã³¢ÊÔÊÒ¸±½ÌÊÚ¡£ËûÊÇLAMDA×éµÄ³ÉÔ±¡£ËûµÄ×êÑÐÐËÖÂÖØÒªÊÇ»úе½ø½¨¡£³ö¸ñÊÇ
£¬Ëû¶ÔÈõ¼à¶½½ø½¨¡¢Í³¼Æ½ø½¨ºÍÓÅ»¯¸ÐÐËÖÂ¡£ËûÔÚ¶¥¼¶ÆÚ¿¯ºÍ»áÒéÉÏ°ä·¢ÁË40¶àƪÂÛÎÄ
£¬ÈçJMLR¡¢TPAMI¡¢AIJ¡¢ICML¡¢NIPS¡¢AAAIµÈ¡£ËûÔøÊǶ¥¼¶ÈËΪÖÇÄÜ»áÒéµÄ¸ß¼¶·¨Ê½Î¯Ô±
£¬ÈçIJCAI'15/17/19¡¢AAAI'19/20µÈ
£¬ÒÔ¼°¡¶Neural Network¡·±àί¡£Ôøµ£ÈÎIEEE Bigcomp2020¹²Í¬·¨Ê½Ö÷ϯ¡¢ACML2019 Tutorial¹²Í¬Ö÷ϯµÈ¡£Ôø»ñÖйúÍÆËã»úѧ»áÓÅÁ¼²©Ê¿ÂÛÎĽ±¡¢½­ËÕÊ¡ÓÅÁ¼²©Ê¿ÂÛÎĽ±µÈ¡£