|
¸ðµÎ¸¦ À§ÇÑ ¸ÞŸ·¯´×PyTorch¸¦ È°¿ëÇÑ Few-shot ÇнÀ ¸ðµ¨°ú ºü¸¥ °ÈÇнÀ ¿¡ÀÌÀüÆ® ¸¸µé±âÁ¤°¡ : 26,000¿øÆǸŰ¡ 23,400¿ø (10% ÇÒÀÎ)
Ç°¸ñÁ¤º¸
Ã¥¼Ò°³ÁöµµÇнÀ°ú °ÈÇнÀÀ» À§ÇÑ ÃֽŠ±â¼ú ¸ÞŸ·¯´×À» ÀÌÇØÇÏÀÚ! ÀÌ Ã¥Àº ÃÖ±Ù ÀΰøÁö´É°ú ¸Ó½Å·¯´× ºÐ¾ß¿¡¼ °¢±¤¹Þ°í ÀÖ´Â ¸ÞŸ·¯´×¿¡ ´ëÇÑ ÀÔ¹® ¼ÀûÀÔ´Ï´Ù. µ¶ÀÚ ¿©·¯ºÐ²²¼ ´Ù¼Ò »ý¼ÒÇÒ ¼ö ÀÖ´Â ¸ÞŸ·¯´×¿¡ ´ëÇÑ °³³äÀ» ÀÌÇØÇÏ°í ½ÇÁ¦·Î ¼¼ºÎ ¾Ë°í¸®Áòµé±îÁö ±¸ÇöÇغ¸´Â ±âȸ¸¦ Á¦°øÇÏ´Â °ÍÀ» ¸ñÇ¥·Î ÇÕ´Ï´Ù. ƯÈ÷ ¸ÞŸ·¯´×¿¡¼ ȸ±Í ¹®Á¦, ºÐ·ù ¹®Á¦¸¦ ´Ù·ç´Â ¸ÞŸ ÁöµµÇнÀ»Ó ¾Æ´Ï¶ó, °ÈÇнÀÀ» ¼Ò°³ÇÏ°í ÀÌ¿¡ ´ëÇØ ¸ÞŸ·¯´×À» Àû¿ëÇÑ ¸ÞŸ °ÈÇнÀ±îÁö ½Éµµ ÀÖ°Ô ´Ù·ç´Â °ÍÀÌ ÀÌ Ã¥ÀÇ Å« ÀåÁ¡ÀÔ´Ï´Ù. óÀ½¿¡´Â ´Ù¼Ò »ý¼ÒÇÒ ¼ö ÀÖÁö¸¸, ²ÙÁØÈ÷ ÀÌ Ã¥À» ¹Ýº¹Çؼ ÀÐ°í ½Ç½ÀÇÏ¸é ¸ÚÁø ÃֽŠ¸Ó½Å·¯´× ±â¼úÀÎ ¸ÞŸ·¯´×À» ÇÑÃþ ±íÀÌ ÀÌÇØÇÒ ¼ö ÀÖÀ» °ÍÀÔ´Ï´Ù.
¸ñÂ÷¢Ã 1Àå: ¸ÞŸ·¯´× °³¿ä 1.1 ¸Ó½Å·¯´×°ú µö·¯´× 1.2 ¸ÞŸ·¯´×À̶õ? 1.3 ¸ÞŸ·¯´× ÇнÀ ȯ°æ ±¸Ãà ___1.3.1 ¾Æ³ªÄÜ´Ù ¼³Ä¡¿Í »ç¿ë ___1.3.2 ¾Æ³ªÄÜ´Ù ¼³Ä¡ ___1.3.3 ±êÇãºê ÀúÀå¼Ò Ŭ·Ð ¹× ȯ°æ ±¸Ãà ¢Ã 2Àå: ¸ÞŸ ÁöµµÇнÀ 2.1 ¸ÞŸ·¯´× ¹®Á¦ Á¤ÀÇ ___2.1.1 ŽºÅ© Á¤ÀÇ ___2.1.2 ¸ÞŸ·¯´× µ¥ÀÌÅͼ ___2.1.3 ¸ÞŸ·¯´× ___2.1.4 ½Ç½À: Torchmeta ¶óÀ̺귯¸® ¼Ò°³ 2.2 ¸ðµ¨ ±â¹Ý ¸ÞŸ·¯´× ___2.2.1 ¸ðµ¨ ±â¹Ý ¸ÞŸ·¯´×ÀÇ ÇÙ½É °³³ä ___2.2.2 NTM(Neural Turing Machines) ___2.2.3 MANN(Memory-Augmented Neural Networks) ___2.2.4 ½Ç½À: MANN ±¸Çö ___2.2.5 SNAIL(Simple Neural Attentive Meta-Learner) ___2.2.6 ½Ç½À: SNAIL ±¸Çö 2.3 ÃÖÀûÈ ±â¹Ý ¸ÞŸ·¯´× ___2.3.1 ÀüÀÌÇнÀ°ú ÃÖÀûÈ ±â¹Ý ¸ÞŸ·¯´× ___2.3.2 MAML°ú FOMAML ___2.3.3 ½Ç½À: MAML-Regression ___2.3.4 ½Ç½À: MAML-Classification 2.4 ¸ÞÆ®¸¯ ±â¹Ý ¸ÞŸ·¯´× ___2.4.1 KNN°ú ¸ÞÆ®¸¯ ±â¹Ý ¸ÞŸ·¯´× ___2.4.2 Matching ³×Æ®¿öÅ© ___2.4.3 ½Ç½À: Matching ³×Æ®¿öÅ© ±¸Çö ___2.4.4 Prototypical ³×Æ®¿öÅ© ___2.4.5 ½Ç½À: Prototypical ³×Æ®¿öÅ© ±¸Çö 2.5 ¸ÞŸ·¯´× ¾Ë°í¸®Áò ¼Ó¼º°ú Àå´ÜÁ¡ ___2.5.1 ¸ÞŸ·¯´× ¾Ë°í¸®ÁòÀÇ ¼¼ °¡Áö ¼Ó¼º ___2.5.2 ¸ÞŸ·¯´× ¾Ë°í¸®Áò ºñ±³ ¢Ã 3Àå: °ÈÇнÀ °³¿ä 3.1 ¸¶¸£ÄÚÇÁ °áÁ¤ °úÁ¤, Á¤Ã¥, °¡Ä¡ÇÔ¼ö ___3.1.1 ¸¶¸£ÄÚÇÁ °áÁ¤ °úÁ¤ ___3.1.2 Á¤Ã¥°ú °ÈÇнÀÀÇ ¸ñÇ¥ ___3.1.3 °¡Ä¡ ÇÔ¼ö 3.2 ŽÇè°ú È°¿ë 3.3 °ÈÇнÀ ¾Ë°í¸®ÁòÀÇ Á¾·ù ___3.3.1 On-policy¿Í Off-policy ___3.3.2 Á¤Ã¥ ±â¹Ý ¾Ë°í¸®Áò ___3.3.3 °¡Ä¡ ±â¹Ý ¾Ë°í¸®Áò ___3.3.4 ¾×ÅÍ Å©¸®Æ½ ¾Ë°í¸®Áò 3.4 TRPO(Trust Region Policy Optimization) ___3.4.1 TRPO ¾ÆÀ̵ð¾î ___3.4.2 Surrogate ¸ñÀû ÇÔ¼ö¿Í Á¦¾à Á¶°Ç ___3.4.3 ÄÓ·¹ ±×¶óµð¾ðÆ®¹ý ±â¹Ý ÃÖÀûÈ 3.5 PPO(Proximal Policy Optimzation) ___3.5.1 PPO ¾ÆÀ̵ð¾î ___3.5.2 Clipped Surrogate ¸ñÀûÇÔ¼ö ___3.5.3 PPO ¾Ë°í¸®Áò 3.6 SAC(Soft Actor Critic) ___3.6.1 ¿£Æ®·ÎÇÇ ___3.6.2 ÃÖ´ë ¿£Æ®·ÎÇÇ °ÈÇнÀ ___3.6.3 °¡Ä¡ÇÔ¼ö ¹× Á¤Ã¥ ÇнÀ ___3.6.4 SAC ¾Ë°í¸®Áò ¢Ã 4Àå: ¸ÞŸ °ÈÇнÀ 4.1 ¸ÞŸ °ÈÇнÀ ___4.1.1 ŽºÅ© °³³ä ¼Ò°³ ___4.1.2 ¸ÞŸ °ÈÇнÀ ¹®Á¦ Á¤ÀÇ ___4.1.3 MuJoCo ¹× Half-Cheetah ȯ°æ °³³ä ¼Ò°³ 4.2 ¼øȯ Á¤Ã¥ ¸ÞŸ °ÈÇнÀ ___4.2.1 GRU ___4.2.2 ¼øȯ Á¤Ã¥ ¸ÞŸ °ÈÇнÀ ___4.2.3 RL2 ___4.2.4 ½Ç½À: RL2 ±¸Çö 4.3 ÃÖÀûÈ ±â¹Ý ¸ÞŸ °ÈÇнÀ ___4.3.1 MAML-RL ___4.3.2 ½Ç½À: MAML-RL ±¸Çö 4.4 ÄÁÅؽºÆ® ±â¹Ý ¸ÞŸ °ÈÇнÀ ___4.4.1 ŽºÅ© Ãß·Ð °üÁ¡¿¡¼ÀÇ ¸ÞŸ °ÈÇнÀ ___4.4.2 ÄÁÅؽºÆ® ±â¹Ý Á¤Ã¥ ___4.4.3 º¯ºÐÀû Ãß·Ð ___4.4.4 PEARL(Probabilistic Embeddings for Actor critic RL) ___4.4.5 ½Ç½À: PEARL ±¸Çö ¢Ã 5Àå: ¿ÀǠ縰Áö¿Í ¸ÞŸ·¯´× ¾ÖÇø®ÄÉÀÌ¼Ç 5.1 ¿ÀǠ縰Áö(Open Chanllenges) ___5.1.1 ¸ÞŸ °úÀûÇÕ ___5.1.2 Ä¡¸íÀû ¸Á°¢°ú Áö¼Ó ÇнÀ ___5.1.3 ºÎÁ·ÇÑ º¥Ä¡¸¶Å© ___5.1.4 ºÎÁ·ÇÑ ·¹À̺íµÈ µ¥ÀÌÅÍ¿Í ¸ÞŸ ºñÁöµµ ÇнÀ 5.2 ¸ÞŸ·¯´× ¾ÖÇø®ÄÉÀÌ¼Ç ___5.2.1 ÄÄÇ»ÅÍ ºñÀü ___5.2.2 °ÈÇнÀ ___5.2.3 ÀÚ¿¬¾î ó¸® ___5.2.4 ÀÇ·á ___5.2.5 ¸¶Ä¡¸ç »ó¼¼ À̹ÌÁö
(°³)
1. ¹è¼Û
¹è¼ÛÀº ´çÀÏ ¿ÀÈÄ 4½ÃÀÌÀü(Åä¿äÀÏÀº ¿ÀÈÄ12½Ã)¿¡ ÁÖ¹®¹× °áÀ縦 ÇÏ½Ã¸é ´çÀϹè¼ÛµË´Ï´Ù. ¹è¼ÛÀº ÁÖ¹® ÈÄ °áÀç¿Ï·áÀÏÀ» ±âÁØÀ¸·Î 1~2ÀÏ ¾È¿¡ ¹è¼Û¿Ï·á¸¦ ¿øÄ¢À¸·Î ÇÏ°í ÀÖ½À´Ï´Ù. ´Ù¸¸ ºÎµæÀÌÇÑ °æ¿ì(ÇØ´ç»óÇ°ÀÇ Ç°Àý ¹× ÀýÆÇ) °í°´´Ô²² Àüȸ¦ µå·Á¼ 󸮸¦ ÇÕ´Ï´Ù. 2. ¹è¼Û·á ¹è¼Û·á´Â 25,000¿ø ÀÌ»óÀÌ °æ¿ì À̱×ÀëÇ÷¯½º¿¡¼ ºÎ´ãÇÏ°í, ¹Ì¸¸ÀÏ °æ¿ì °í°´ÀÌ 2,500¿øÀ» ºÎ´ãÇÕ´Ï´Ù. ´Ù¸¸ ±¸¸Å±Ý¾×ÀÌ 25,000¿ø ÀÌ»óÀε¥ Àû¸³±Ý µîÀ» »ç¿ëÇÏ¿© ½ÇÁ¦ °áÀç±Ý¾×ÀÌ 25,000¿ø ¹Ì¸¸ÀÌ µÇ´õ¶óµµ À̱×ÀëÇ÷¯½º¿¡¼ ¹è¼Û·á¸¦ ºÎ´ãÇÏ°í, Á¦ÁÖ¸¦ Á¦¿ÜÇÑ µµ¼Áö¿ªÀÇ °æ¿ì´Â 5,000¿øÀÌ Ãß°¡µË´Ï´Ù. Çؿܹè¼ÛÀº Àü¾× °í°´ºÎ´ãÀÌ µË´Ï´Ù. 5. ¹ÝÇ°/±³È¯
¹ÞÀ¸½Å ³¯ ºÎÅÍ ÀÏÁÖÀÏ À̳» ¹ÝÇ°/±³È¯ÀÌ °¡´ÉÇϽʴϴÙ. ½Ã°£ÀÌ Áö³ª¸é 󸮰¡ ºÒ°¡ÇÔÀ» ¾Ë·Á µå¸³´Ï´Ù. ´Ù¸¸ ÁÖ¹®ÇϽŠ»óÇ°°ú ´Ù¸¥ »óÇ°ÀÌ ¹è¼ÛµÇ°Å³ª Æĺ», ³«ÀåÀÌ ÀÖ´Â µµ¼ÀÇ °æ¿ì ÀÌ¿ë¾à°ü¿¡ ÁØÇÏ¿© À̱×ÀëÇ÷¯½ºÀÇ ºñ¿ëÀ¸·Î ±³È¯, ¹ÝÇ° ¹× ȯºÒ µîÀ» Çص帳´Ï´Ù. 4. ȯºÒ ¹ÞÀ¸½Å ³¯ºÎÅÍ 2ÀÏ À̳»·Î Àüȳª 1:1 °Ô½Ã±Û·Î Àû¾îÁÖ¼Å¾ß °¡´ÉÇϽʴϴÙ. 5. ÁÖÀÇ °í°´´Ô º¯½ÉÀ¸·Î ÀÎÇÑ ±³È¯/¹ÝÇ°/ȯºÒ °í°´´Ô²²¼ Åùèºñ(¿Õº¹Åùèºñ)¸¦ ºÎ´ãÀÌ µÇ´Ï ÀÌÁ¡ À¯ÀÇÇØ ÁֽʽÿÀ. ¶ÇÇÑ °ÀÇÅ×ÀÙ/¾ãÀº¹®Á¦Áý/ºñ´ÒÆ÷ÀåµÈÁ¦Ç°µîÀº Çѹø ±¸¸ÅÇÏ½Ã¸é ±³È¯/¹ÝÇ°/ȯºÒÀÌ ÀüÇô ¾ÈµË´Ï´Ù. ½ÅÁßÈ÷ »ý°¢Çϼż ±¸¸Å ºÎŹµå¸³´Ï´Ù. ´Ù¸¸ ºÒ·®Å×ÀÙÀÇ °æ¿ì ±× ºÒ·®ÀÌ µÈ °³º°Å×ÀÙÀ» À̱×ÀëÇ÷¯½ºÀÇ ºñ¿ëÀ¸·Î A/S¸¦ Çص帳´Ï´Ù. 6. ÁÖ¹®Ãë¼Ò, ±³È¯, ¹ÝÇ° ¹× ȯºÒÀº ´çÀÏ 3½Ã ÀÌÀü¿¡ ÇØÁÖ¼Å¾ß °¡´ÉÇÕ´Ï´Ù. ±× ÀÌÈÄ´Â ¹è¼Û µî¿¡ µû¸¥ Á¦ºñ¿ëÀ» ºÎ´ãÇÏ¼Å¾ß ÇÕ´Ï´Ù.
|
|
||||||||
|
||||||||
|
||||||||
* »óÇ°»çÀÌÁî Ä¡¼ö´Â Àç´Â ¹æ¹ý°ú À§Ä¡¿¡ µû¶ó 1~3cm ¿ÀÂ÷°¡ ÀÖÀ» ¼ö ÀÖ½À´Ï´Ù.