·µ»Ø- ½¹µãÊÖÒÕ
- ÒÔÔ´´ÊÖÒÕϵͳΪ»ù±¾£¬£¬£¬£¬£¬SenseCoreÉÌÌÀAI´ó×°ÖÃΪ½¹µã»ù×ù£¬£¬£¬£¬£¬½á¹¹¶àÁìÓò¡¢¶àÆ«ÏòÇ°ÑØÑо¿£¬£¬£¬£¬£¬
¿ìËÙÂòͨAIÔÚ¸÷¸ö±ÊÖ±³¡¾°ÖеÄÓ¦Ó㬣¬£¬£¬£¬ÏòÐÐÒµ¸³ÄÜ¡£¡£¡£¡£¡£¡£¡£¡£
ÉÌÑÐØICLR 2019ÂÛÎĽâ¶Á£ºËæ»úÉñ¾ÍøÂç½á¹¹ËÑË÷ (SNAS)
¿ËÈÕ£¬£¬£¬£¬£¬ÉÌÌÀ¿Æ¼¼Ñо¿Ôº½ÒÏþÂÛÎÄ¡¶Ëæ»úÉñ¾ÍøÂç½á¹¹ËÑË÷¡·(SNAS£¬£¬£¬£¬£¬Stochastic Neural Architecture Search)£¬£¬£¬£¬£¬Òѱ»Éî¶Èѧϰ¾Û»áICLR(International Conference Of Learning Representation)ÈÎÃü¡£¡£¡£¡£¡£¡£¡£¡£
¸ÃÂÛÎÄÌá³öÁËÒ»ÖÖȫеľ¼Ã¡¢¸ßЧÇÒ×Ô¶¯»¯Ë®Æ½¸ßµÄÉñ¾ÍøÂç½á¹¹ËÑË÷(NAS)ÒªÁ죬£¬£¬£¬£¬Êܵ½Éî¶ÈѧϰѧÊõÁìÓòµÄÆÕ±é¹Ø×¢¡£¡£¡£¡£¡£¡£¡£¡£
NASÊÇ×ÔÄîͷеѧϰ(AutoML)ÖеÄÈÈÃÅÎÊÌâÖ®Ò»£¬£¬£¬£¬£¬Í¨¹ýÉè¼Æ¾¼Ã¸ßЧµÄËÑË÷ÒªÁ죬£¬£¬£¬£¬×Ô¶¯»ñµÃ·º»¯ÄÜÁ¦Ç¿¡¢Ó²¼þÓѺõÄÉñ¾ÍøÂç½á¹¹£¬£¬£¬£¬£¬¿ÉÒÔ´ó×Ú½µµÍÑз¢±¾Ç®£¬£¬£¬£¬£¬½â·ÅÑо¿Ô±µÄ´´Á¢Á¦¡£¡£¡£¡£¡£¡£¡£¡£
ÔÚ¾µäµÄNASÒªÁìÖУ¬£¬£¬£¬£¬Öð²ã´î½¨Éñ¾ÍøÂçµÄÀú³Ì±»½¨Ä£³ÉÒ»¸öÇ¿»¯Ñ§Ï°(Reinforcement Learning)ÖÇÄÜÌåÂí¶û¿É·ò¾öÒéÀú³Ì(MDP ,Markov Decision Process)¡£¡£¡£¡£¡£¡£¡£¡£
ÉÌÌÀ¿Æ¼¼Ñо¿ÍŶÓͨ¹ýÉîÈëÆÊÎöNASʹÃüµÄMDP£¬£¬£¬£¬£¬»Ø¸²ÁËÉÏÊöÎÊÌ⣬£¬£¬£¬£¬²¢Ìá³öÁËÒ»¸ö¸ü¸ßЧµÄÒªÁ졪¡ªËæ»úÉñ¾ÍøÂç½á¹¹ËÑË÷£¬£¬£¬£¬£¬ÖØн¨Ä£ÁËNASÎÊÌâ¡£¡£¡£¡£¡£¡£¡£¡£Óë»ùÓÚÇ¿»¯Ñ§Ï°µÄÒªÁì(ENAS)Ïà±È£¬£¬£¬£¬£¬SNASµÄËÑË÷ÓÅ»¯¿É΢·Ö£¬£¬£¬£¬£¬ËÑË÷ЧÂʸü¸ß¡£¡£¡£¡£¡£¡£¡£¡£
ÓëÆäËû¿É΢·ÖµÄÒªÁì(DARTS)Ïà±È£¬£¬£¬£¬£¬SNASÖ±½ÓÓÅ»¯NASʹÃüµÄÄ¿µÄº¯Êý£¬£¬£¬£¬£¬ËÑË÷Ч¹ûÎó²î¸üС¡£¡£¡£¡£¡£¡£¡£¡£±ðµÄ£¬£¬£¬£¬£¬»ùÓÚSNAS¼á³ÖÁËËæ»úÐÔ(Stochasticity)µÄÓÅÊÆ£¬£¬£¬£¬£¬¸ÃÂÛÎĽøÒ»²½Ìá³öͬʱÓÅ»¯ÍøÂçËðʧº¯ÊýµÄÆÚÍûºÍÍøÂçÕýÏòʱÑÓµÄÆÚÍû£¬£¬£¬£¬£¬×Ô¶¯ÌìÉúÓ²¼þÓѺõÄÏ£º±ÍøÂç¡£¡£¡£¡£¡£¡£¡£¡£
±ðµÄ£¬£¬£¬£¬£¬ÉÌÌÀ¿Æ¼¼È¥ÄêÓÚCVPR½ÒÏþOralÂÛÎÄ¡¶Éî¶ÈÔöǿѧϰ×Ô¶¯ÍøÂç½á¹¹Éè¼Æ¡·£¬£¬£¬£¬£¬Ìá³öÁËÒ»ÖÖ¸ßЧËã·¨£¬£¬£¬£¬£¬Í¨¹ýÇ¿»¯Ñ§Ï°À´×Ô¶¯Éè¼Æ×îÓŵÄÍøÂç½á¹¹¡£¡£¡£¡£¡£¡£¡£¡£Ëæ×ÅÉÌÌÀ¿Æ¼¼Í¨Óûù´¡Ëã·¨µÄÉú³¤£¬£¬£¬£¬£¬ÔÚδÀ´£¬£¬£¬£¬£¬SNAS¿ÉÒÔ½â¾ö¸ü¶àÌض¨Ãü¾ÝºÍÌØÊâµÄÓ²¼þÓ¦Óó¡¾°µÄÉñ¾ÍøÂç½á¹¹×Ô¶¯¶¨ÖÆ»¯ÐèÇ󡣡£¡£¡£¡£¡£¡£¡£
