NeuroIPS (
ç§ãã¡ DS ãšã³ãžãã¢ããæ°ãã XNUMX 幎ã§çç©åŠãèšèªåŠãå¿çåŠãç¿åŸã§ããã§ãããã? ã¬ãã¥ãŒã§ãç¥ããããŸãã
ä»å¹Žã®äŒè°ã«ã¯ã13500 ãåœãã 80 人以äžãã«ããã®ãã³ã¯ãŒããŒã«éãŸããŸããã ãºãã«ãã³ã¯ããã·ã¢ã代衚ããŠã«ã³ãã¡ã¬ã³ã¹ã«åå ããã®ã¯ä»å¹ŽãåããŠã§ã¯ãªããDS ããŒã ã¯éè¡æ¥åããã»ã¹ã«ããã ML ã®å°å
¥ãML ã³ã³ãã¹ãããºãã«ãã³ã¯ DS ãã©ãããã©ãŒã ã®æ©èœã«ã€ããŠè©±ããã ML ã³ãã¥ããã£ã«ããã 2019 幎ã®äž»ãªãã¬ã³ãã¯äœã§ããã? ã«ã³ãã¡ã¬ã³ã¹ã®åå è
ã¯ããèšããŸãã
ä»å¹ŽãNeurIPS ã¯ãã¢ã«ãŽãªãºã ãæ°ããã¢ãã«ãæ°ããããŒã¿ã®æ°ããã¢ããªã±ãŒã·ã§ã³ãªã©ã1400 件ãè¶
ããè«æãåãå
¥ããŸããã
å 容ïŒ
- ãã¬ã³ã
-
- ã¢ãã«ã®è§£éå¯èœæ§
- åŠéæ§
- æšè«
- RL
- GAN
- åºæ¬çãªæåŸ è¬æŒ
-
- ããœãŒã·ã£ã« ã€ã³ããªãžã§ã³ã¹ãããã¬ã€ãº ã¢ã²ã© 〠ã¢ã«ã«ã¹ (Google)
- ãVeridical Data ScienceããBin Yu (ããŒã¯ã¬ãŒ)
- ãæ©æ¢°åŠç¿ã«ãã人éã®è¡åã¢ããªã³ã°: æ©äŒãšèª²é¡ããããªã¢ã»Mã»ãªãªããŒãã¢ã«ããŒãã»ã¢ãªã»ãµã©ãŒ
- ãã·ã¹ãã 1 ããã·ã¹ãã 2 ãã£ãŒãã©ãŒãã³ã°ãžãããšã·ã¥ã¢ ãã³ãžãª
ãã¬ã³ã2019 of the year
1. ã¢ãã«ã®è§£éå¯èœæ§ãšæ°ãã ML ææ³
ã«ã³ãã¡ã¬ã³ã¹ã®äž»ãªããŒãã¯ããªãç¹å®ã®çµæãåŸãããã®ãã®è§£éãšèšŒæ ã§ãã ããã©ãã¯ããã¯ã¹ã解éã®å²åŠçéèŠæ§ã«ã€ããŠã¯é·ãéèªãããŠããŸãããããã®åéã§ã¯ãã£ãšçŸå®çãªæ¹æ³ãšæè¡éçºããããŸããã
ã¢ãã«ãè€è£œããããããç¥èãæœåºããæ¹æ³è«ã¯ãç§åŠã®ããã®æ°ããããŒã«ãããã§ãã ã¢ãã«ã¯ãæ°ããç¥èãååŸããããããã¹ãããããã®ããŒã«ãšããŠæ©èœããŸããã¢ãã«ã®ååŠçããã¬ãŒãã³ã°ãããã³é©çšã®å段éã¯åçŸå¯èœã§ããå¿
èŠããããŸãã
åºçç©ã®ããªãã®éšåã¯ãã¢ãã«ãããŒã«ã®æ§ç¯ã§ã¯ãªããçµæã®ã»ãã¥ãªãã£ãéææ§ãæ€èšŒå¯èœæ§ã®ç¢ºä¿ã®åé¡ã«åœãŠãããŠããŸãã ç¹ã«ãã¢ãã«ã«å¯Ÿããæ»æ (æµå¯Ÿçæ»æ) ã«ã€ããŠã¯å¥ã®ã¹ããªãŒã ãç»å Žãããã¬ãŒãã³ã°ã«å¯Ÿããæ»æãšã¢ããªã±ãŒã·ã§ã³ã«å¯Ÿããæ»æã®äž¡æ¹ã®ãªãã·ã§ã³ãæ€èšãããŠããŸãã
èšäºïŒ
æ€èšŒããŒã¿ç§åŠ â ã¢ãã«æ€èšŒã®æ¹æ³è«ã«é¢ããããã°ã©ã ã«é¢ããèšäºã ã¢ãã«ã解éããããã®ææ°ããŒã«ã®æŠèŠãç¹ã«æ³šæã®äœ¿çšãšãç·åœ¢ã¢ãã«ã䜿çšããŠãã¥ãŒã©ã« ãããã¯ãŒã¯ããèžçãããããšã«ãã£ãŠç¹åŸŽã®éèŠæ§ãååŸããæ¹æ³ã®æŠèŠãå«ãŸããŠããŸãããããªæã: 解éå¯èœãªç»åèªèã®ããã®æ·±å±€åŠç¿ ãã£ãªãã¡ã³ã»ãã§ã³ããªã¹ã«ãŒã»ãªãŒããããšã«ã»ã¿ãªãã¢ãªãŒãã»ããŒããããã·ã³ã·ã¢ã»ã«ãŒãã£ã³ããžã§ããµã³ã»Kã»ã¹ãŒãã£ãŒã ãã¥ãŒã©ã« ãããã¯ãŒã¯ã«ããã解éå¯èœæ§ã¡ãœããã®ãã³ãããŒã¯ ãµã©ã»ããã«ãŒããã¥ããã«ã»ãšã«ãã³ãããŒã¿ãŒïŒã€ã³ã»ãã³ããŒãã³ã¹ãããŒã³ã»ãã泚ææ¡åŒµãšãŒãžã§ã³ãã䜿çšãã解éå¯èœãªåŒ·ååŠç¿ã«åã㊠ã¢ã¬ã¯ãµã³ããŒã»ã¢ããããããšã«ã»ãŸã©ã³ããã€ã¯ã»ãã£ããã¹ããŒãããŒã³ã»ãŽã£ã¢ã¹ãã©ããããŒãã»ãã¡ãã¹ã»ã¬ãŒã³ãã©ã³ãã ãã©ã¬ã¹ãã«å¯Ÿããåãã®ãªã MDI æ©èœéèŠåºŠã®æž¬å® ã·ã£ãªã»ãªãŒããŠãŒã»ã¯ã³ãã¹ãã³ã¿ã»ãã¹ãã«ãŒã«ã»ã¯ã³ãã¢ããã³ã»ãŠãŒèŠ³å¯å¯èœãªããŒã¿ããªãå Žåã®ç¥èã®æœåº ãŠã»ãžã§ãã³ããã§ã»ããã§ã³ããã ã»ããã ããŠã»ã¬ã³ç¬ç«ããŠåçŸå¯èœãªæ©æ¢°åŠç¿ç 究ã®å®éåã«åããäžæ© ãšãã¯ãŒãã»ã©ã
ExBert.net ã¯ããã¹ãåŠçã¿ã¹ã¯ã®ã¢ãã«è§£éã瀺ããŠããŸã
2. åŠéæ§
ä¿¡é Œæ§ã®é«ãæ€èšŒã確ä¿ããç¥èãæ€èšŒããã³æ¡åŒµããããã®ã¡ã«ããºã ãéçºããã«ã¯ãML ãšå¯Ÿè±¡åé (å»åŠãèšèªåŠãç¥çµçç©åŠãæè²ãªã©) ã®èœåãåæã«åããé¢é£åéã®å°é家ãå¿ èŠã§ãã ç¹ã«ãç¥çµç§åŠãšèªç¥ç§åŠã«ãããèäœãè¬æŒã®ååšæãããé¡èã§ããããšã¯æ³šç®ã«å€ããŸããå°é家ã®æ¥è¿ãã¢ã€ãã¢ã®åçšãèŠãããŸãã
ãã®æ¥è¿ã«å ããŠãããã¹ããšåçãããã¹ããšã²ãŒã ãã°ã©ãããŒã¿ããŒã¹ + ããã¹ããšåçãªã©ãããŸããŸãªãœãŒã¹ããã®æ å ±ã®å ±ååŠçã«ãããŠåŠéæ§ãçãŸããŠããŸãã
èšäºïŒ
- ç¥çµç§åŠ + ML â
èªç¶èšèªåŠçïŒè³å ïŒã«ããèªç¶èšèªåŠçïŒæ©æ¢°å ïŒã®è§£éãšæ¹å - VisualQA -
æœè±¡åã«ããåŠç¿: ãã¥ãŒã©ã« ã¹ããŒã ãã·ã³ - RL + NLP -
èªç¶èšèªåœä»€ã®çæãšããã«åŸãããšã«ããéå±€çãªææ決å®
RL ããã³ NLP ãã¬ã€ ãªã³ã©ã€ã³æŠç¥ã«åºã¥ã XNUMX ã€ã®ã¢ãã« - ã¹ãã©ããžã¹ããšãšã°ãŒã¯ãã£ã -
3.æšè«
人工ç¥èœã®åŒ·åã¯ãèªå·±åŠç¿ã·ã¹ãã ããæèããæšè«ãæšè«ãžã®åãã§ãã ç¹ã«ãå ææšè«ãšåžžèæšè«ãçºéããŠããŸãã ã¬ããŒãã®äžã«ã¯ãã¡ã¿åŠç¿ (åŠç¿æ¹æ³ã«ã€ããŠ) ããDL ãã¯ãããžãŒãš 1 次ããã³ 2 次ããžãã¯ã®çµã¿åããã«ç¹åãããã®ããããŸããäžè¬äººå·¥ç¥èœ (AGI) ãšããçšèªã¯ãè¬æŒè ã®è¬æŒã§ãã䜿ãããçšèªã«ãªãã€ã€ãããŸãã
èšäºïŒ
èŠèŠçåžžèæšè«ã®ããã®ç°çš®ã°ã©ãåŠç¿ ãŠã§ã€ãžã£ã³ã»ãŠãŒããžã³ãŠã§ã³ã»ãã§ãŠããŠã§ã€ããªã»ãŠãŒãã·ã£ãªãã³ã»ãªã£ã³ããã³ã»ã·ã£ãªã¢ããã¯ãã£ãåŠç¿ã«ããæ©æ¢°åŠç¿ãšè«ççæšè«ã®æ©æž¡ã Wang-Zhou DaiãQiuling XuãYang YuãZhi-Hua Zhouäžéè«çã§æšè«ããããšãæé»çã«åŠç¿ãã ãŽã¡ã€ã·ã£ãã¯ã»ãã«ããã¬ã³ãã³ã»ãžã¥ãPHYRE: ç©ççæšè«ã®æ°ãããã³ãããŒã¯ ã¢ã³ãã³ã»ãã¯ãã³ãããŒã¬ã³ã¹ã»ãã¡ã³ã»ãã«ã»ããŒãã³ããžã£ã¹ãã£ã³ã»ãžã§ã³ãœã³ãããŒã©ã»ã°ã¹ã¿ããœã³ããã¹ã»ã¬ãŒã·ãã¯æšè«ã®ããã®ç¥èã®éååã蟌㿠ãã£ãã·ã¥ ã¬ãŒã°ãã·ã£ãžã¹ ã€ã¯ãã«ããµã³ãã·ã¥ K. ã¹ãªãŽã¡ã¹ã¿ãŽã¡ãããªãã ãŽã£ã·ã¥ã¯ã«ã«ãããã ã«ã©ãã ãL ãŽã§ã³ã«ã¿ ã¹ãã©ããã¢ã
4.匷ååŠç¿
äœæ¥ã®ã»ãšãã©ã¯ãã¢ãŒããã¯ãã£ãšã³ã³ãã¥ãŒã¿ ããžã§ã³ãNLPãã°ã©ã ããŒã¿ããŒã¹ãçµã¿åããã RL - DOTA2ãStarcraft ãªã©ã®åŸæ¥ã®é åã®éçºãç¶ç¶ããŠããŸãã
ã«ã³ãã¡ã¬ã³ã¹ã®å¥ã®æ¥ã¯ RL ã¯ãŒã¯ã·ã§ããã«åœãŠãããããã§ã¯ãããŸã§ã®ãã¹ãŠã®ã¢ãã«ãç¹ã«ãœãã ã¢ã¯ã¿ãŒ ã¯ãªãã£ã«ã«ãããåªãããªããã£ãã¹ãã£ã㯠ã¢ã¯ã¿ãŒ ã¯ãªãã£ã«ã« ã¢ãã« ã¢ãŒããã¯ãã£ãçºè¡šãããŸããã
èšäºïŒ
楜芳çãªä¿³åªè©è«å®¶ã«ããããè¯ãæ¢çŽ¢ ; ã«ãã«ã»ãã§ã»ã¯ãã¯ã¢ã³ã»ãŽãªã³ããããŒãã»ãããã£ã³ãã«ãã£ã»ãããã³ChainerRL: 深局匷ååŠç¿ã©ã€ãã©ãª ; è€ç° åº·è£ (Preferred Networks, Inc.)*; ç岡 ä¿æš¹ (Preferred Networks, Inc.); Prabhat Nagarajan (Preferred Networks); ç³å·è²Žè£ (æ±äº¬å€§åŠ) [å€éš PDF ãªã³ã¯]ãã³ã³ãããŒã«ãã倢: æœåšçãªæ³ååã«ããåŠç¿è¡å ; ãããã«ã»ããã㌠(Google)*; ãã£ã¢ã·ãŒã»ãªãªãã¯ã©ãã (ãã£ãŒããã€ã³ã); ãžããŒã»ã (ããã³ã倧åŠ); ã¢ãããã»ãã«ãŒãž (Google Brain)ã¯ãŒã¯ã·ã§ããã®è³æ
StarCraft ãã¬ã€ã€ãŒã¯ Alphastar ã¢ãã«ãšæŠããŸã (DeepMind)
5.ã¬ã³
çæãããã¯ãŒã¯ã¯äŸç¶ãšããŠæ³šç®ãéããŠããŸããå€ãã®äœåã§ã¯ãæ°åŠç蚌æã«ããã© GAN ã䜿çšãããŠããããŸããæ°ããçããæ¹æ³ (ã°ã©ãçæã¢ãã«ãç³»åã®æäœãããŒã¿å ã®å æé¢ä¿ãžã®é©çšãªã©) ã«ãå¿çšãããŠããŸãã
èšäºïŒ
æ¡ä»¶ä»ã GAN çšã® GOLD ãµã³ãã«ã®ãã€ãã³ã° ã¢ã»ãµã³ãŠããã ã»ããã³ããã ã»ãœã³ãŠã³ããã§ã»ãã³ã¹ãã·ã³ã»ãžãGAN ã®æŒžé²çãªæ¡åŒµ ãã³ã»ãã£ã³ãã¢ã³ãã»ã³ã¬ãŽã¡æ¡ä»¶ä»ã GAN ã䜿çšãã衚圢åŒããŒã¿ã®ã¢ããªã³ã° ã¬ã€ã»ã·ã¥ãŒãããªã¢ã»ã¹ã¯ãŒã©ãªãã¥ãã¢ã«ãã¬ãã»ã¯ãšã¹ã¿ïŒã€ã³ãã¡ã³ããã«ãªã€ã³ã»ãŽã£ãŒã©ããã£ããpapers.nips.cc/paper/9377-a-domain-agnostic-measure-for-monitoring-and-evaluating-gans
ããå€ãã®ãä»äºãåãå
¥ããŠããã ããã®ã§ã
æåŸ è¬æŒ
ããœãŒã·ã£ã« ã€ã³ããªãžã§ã³ã¹ãããã¬ã€ãº ã¢ã²ã© 〠ã¢ã«ã«ã¹ (Google)
è¬æŒã§ã¯ãæ©æ¢°åŠç¿ã®äžè¬çãªæ¹æ³è«ãšãçŸåšæ¥çãå€ããèŠéããã€ãŸãç§ãã¡ãçŽé¢ããŠããå²è·¯ã«çŠç¹ãåœãŠãŸãã è³ãšé²åã¯ã©ã®ããã«æ©èœããã®ã§ãããã?ãŸãããªãç§ãã¡ã¯èªç¶ã·ã¹ãã ã®çºéã«ã€ããŠãã§ã«ç¥ã£ãŠããããšãã»ãšãã©æŽ»çšããŠããªãã®ã§ãããã?
ML ã®ç£æ¥çºå±ã¯ãNeurIPS ã«é¢ããç 究ãæ¯å¹Žçºè¡šããŠãã Google ã®çºå±ã®ãã€ã«ã¹ããŒã³ãšã»ãŒäžèŽããŠããŸãã
- 1997 â æ€çŽ¢æ©èœãæåã®ãµãŒããŒãå°èŠæš¡ãªã³ã³ãã¥ãŒãã£ã³ã°èœåã®ç«ã¡äžã
- 2010 â ãžã§ã ãã£ãŒã³ã Google Brain ãããžã§ã¯ããç«ã¡äžãããã¥ãŒã©ã« ãããã¯ãŒã¯ã®ããŒã ãå§ãŸããŸãã
- 2015 â ãã¥ãŒã©ã« ãããã¯ãŒã¯ã®ç£æ¥å®è£ ãããŒã«ã« ããã€ã¹ã§ã®çŽæ¥ã®é«éé¡èªèããã³ãœã« ã³ã³ãã¥ãŒãã£ã³ã°çšã«èª¿æŽãããäœã¬ãã« ããã»ããµ - TPUã Googleãå®éšæœèšã«ãã¥ãŒã©ã«ãããã¯ãŒã¯ãå°å ¥ããããã®ããã³ã³ãã¥ãŒã¿ãŒãã©ãºããªãŒãã€ã®é¡äŒŒåã§ããCoral aiãçºå£²
- 2017 â Google 㯠Android äžã§åæ£ãã¬ãŒãã³ã°ã®éçºãéå§ããããŸããŸãªããã€ã¹ããã®ãã¥ãŒã©ã« ãããã¯ãŒã¯ ãã¬ãŒãã³ã°ã®çµæã XNUMX ã€ã®ã¢ãã«ã«çµåããŸãã
çŸåšãæ¥çå šäœãããŒã¿ã®ã»ãã¥ãªãã£ãéçŽãããŒã«ã« ããã€ã¹ã§ã®åŠç¿çµæã®è€è£œã«æ³šåããŠããŸãã
Google ã«ããã°ããã§ãã¬ãŒããã ã©ãŒãã³ã°ã«åºã¥ãçæã¢ãã«ã¯å°æ¥ææãªæ¹åæ§ã§ããããææ°é¢æ°çãªæé·ã®åæ段éã«ããããšã®ããšã§ãã è¬åž«ã«ãããšãGAN ã¯çç©éå£ã®éå£è¡åãšæèã¢ã«ãŽãªãºã ãåçŸããæ¹æ³ãåŠç¿ããããšãã§ããŸãã
XNUMX ã€ã®åçŽãª GAN ã¢ãŒããã¯ãã£ã®äŸã䜿çšãããšãæé©åãã¹ã®æ€çŽ¢ãåãæãããã«ããŸãã£ãŠããããšãããããŸããããã¯ãæé©åèªäœãçºçããªãããšãæå³ããŸãã åæã«ããããã®ã¢ãã«ã¯ãçç©åŠè ã现èéå£ã«å¯ŸããŠè¡ãå®éšãã·ãã¥ã¬ãŒãããã®ã«éåžžã«æåããé£ç©ãæ¢ãéã®æ°ããè¡åæŠç¥ã®åŠç¿ã现èã«åŒ·å¶ããŸãã 人çã¯æé©åé¢æ°ãšã¯ç°ãªãåãããããšçµè«ä»ããããšãã§ããŸãã
ãŠã©ãŒãã³ã° GAN ã®æé©å
çŸåšãæ©æ¢°åŠç¿ã®æ çµã¿ã§ç§ãã¡ãè¡ã£ãŠããããšã¯ãã¹ãŠãçããŠéåžžã«åœ¢åŒåãããã¿ã¹ã¯ã§ããããããã®åœ¢åŒäž»çŸ©ã¯äžè¬åãäžååã§ãããç¥çµççåŠãçç©åŠãªã©ã®åéã«ãããäž»é¡ã®ç¥èã«å¯Ÿå¿ããŠããŸããã
è¿ãå°æ¥ãç¥çµççåŠåéããæ¬åœã«åãã䟡å€ã®ãããã®ã¯ãæ°ãããã¥ãŒãã³ã®ã¢ãŒããã¯ãã£ãšã誀差ã®éäŒæã¡ã«ããºã ã®ããããªæ¹èšã§ãã
人éã®è³èªäœã¯ãã¥ãŒã©ã« ãããã¯ãŒã¯ã®ããã«åŠç¿ããããã§ã¯ãããŸããã
- 圌ã«ã¯ãæèŠã幌å°æã«äžãããããã®ãªã©ãã©ã³ãã ãªäžæ¬¡å ¥åããããŸããã
- 圌ã¯æ¬èœççºéã®åºæã®æ¹åæ§ãæã£ãŠããŸãïŒå¹Œå ããèšèªãåŠã³ãããçŽç«ããŠæ©ããããšããé¡æïŒ
åã ã®è³ãèšç·Žããããšã¯äœã¬ãã«ã®ã¿ã¹ã¯ã§ãããéå£é²åã®ã¡ã«ããºã ãåçŸããããã«ãæ¥éã«å€åããåäœãäºãã«ç¥èãäŒéãããã³ãããŒããèæ ®ããå¿ èŠããããããããŸããã
çŸåš ML ã¢ã«ãŽãªãºã ã«å°å ¥ã§ãããã®ã¯æ¬¡ã®ãšããã§ãã
- éå£ã®åŠç¿ã確å®ã«ããããå人 (ãå人ã®è³ã) ã®å¯¿åœã¯çã现è系統ã¢ãã«ãé©çšããŸãã
- å°æ°ã®äŸã䜿çšããå°æ°ã·ã§ããåŠç¿
- ããè€éãªãã¥ãŒãã³æ§é ããããã«ç°ãªã掻æ§åé¢æ°
- ãã²ãã ãã次äžä»£ã«äŒãã - ããã¯ãããã²ãŒã·ã§ã³ã¢ã«ãŽãªãºã
- ç¥çµççåŠãšãã¥ãŒã©ã«ãããã¯ãŒã¯ãçµã³ã€ãããšãå€ãã®ã³ã³ããŒãã³ãããå€æ©èœãªè³ãæ§ç¯ããæ¹æ³ããããããã«ãªããŸãã
ãã®èŠ³ç¹ããèŠããšãSOTA ãœãªã¥ãŒã·ã§ã³ã®å®è·µã¯æ害ã§ãããå ±éã®ã¿ã¹ã¯ (ãã³ãããŒã¯) ãéçºããããã«èŠçŽãããå¿ èŠããããŸãã
ãVeridical Data ScienceããBin Yu (ããŒã¯ã¬ãŒ)
ãã®ã¬ããŒãã¯ãæ©æ¢°åŠç¿ã¢ãã«ã®è§£éã®åé¡ãšããã®çŽæ¥çãªãã¹ããšæ€èšŒã®æ¹æ³è«ã«çŠç¹ãåœãŠãŠããŸãã ãã¬ãŒãã³ã°ããã ML ã¢ãã«ã¯ãããããæœåºããå¿
èŠãããç¥èã®ãœãŒã¹ãšããŠèªèã§ããŸãã
å€ãã®åéãç¹ã«å»åŠã§ã¯ããã®é ãããç¥èãæœåºããã¢ãã«ã®çµæã解éããããšãªãã«ã¢ãã«ã䜿çšããããšã¯äžå¯èœã§ããããã§ãªããšãçµæãå®å®ããŠãããã©ã³ãã ã§ã¯ãªããä¿¡é Œæ§ããããæ£è ã殺ããªããšãã確信ãæãŠãªããªããŸããå¿è匷ãã äœæ¥æ¹æ³è«ã®å šäœçãªæ¹åæ§ã¯ãã£ãŒãã©ãŒãã³ã°ãã©ãã€ã å ã§çºå±ããŠããããã®å¢çãè¶ ããŠããŸã - æ€èšŒçãªããŒã¿ãµã€ãšã³ã¹ã§ãã ããã¯äœã§ããïŒ
ç§ãã¡ã¯ã次ã®ãããªç§åŠåºçç©ã®å質ãšã¢ãã«ã®åçŸæ§ãéæããããšèããŠããŸãã
- äºæž¬å¯èœãª
- èšç®å¯èœãª
- å®å®ãã
ããã XNUMX ã€ã®ååãæ°ããæ¹æ³è«ã®åºç€ã圢æããŸãã ML ã¢ãã«ããããã®åºæºã«ç §ãããŠãã§ãã¯ããã«ã¯ã©ãããã°ããã§ãããã? æãç°¡åãªæ¹æ³ã¯ãããã«è§£éã§ããã¢ãã« (ååž°ã決å®æš) ãæ§ç¯ããããšã§ãã ãã ãããã£ãŒãã©ãŒãã³ã°ã®ã¡ãªãããããã«åŸãããšãèããŠããŸãã
ãã®åé¡ã«å¯ŸåŠããæ¢åã®æ¹æ³ã¯ããã€ããããŸãã
- ã¢ãã«ã解éããã
- 泚æã«åºã¥ããæ¹æ³ã䜿çšããã
- ãã¬ãŒãã³ã°æã«ã¢ã«ãŽãªãºã ã®ã¢ã³ãµã³ãã«ã䜿çšããç·åœ¢è§£éå¯èœãªã¢ãã«ãç·åœ¢ã¢ãã«ã®ç¹åŸŽã解éããŠãã¥ãŒã©ã« ãããã¯ãŒã¯ãšåãçããäºæž¬ããããšã確å®ã«åŠç¿ããŸãã
- ãã¬ãŒãã³ã° ããŒã¿ãå€æŽããã³åŒ·åããŸãã ããã«ã¯ããã€ãºãå¹²æžãããŒã¿æ¡åŒµã®è¿œå ãå«ãŸããŸãã
- ã¢ãã«ã®çµæãã©ã³ãã ã§ã¯ãªãã軜床ã®æãŸãããªãå¹²æž (æµå¯Ÿçæ»æ) ã«äŸåããªãããšãä¿èšŒããããã®ããããæ¹æ³ã
- ãã¬ãŒãã³ã°åŸã«ãäºåŸçã«ã¢ãã«ã解éããŸãã
- ããŸããŸãªæ¹æ³ã§ç¹åŸŽã®éã¿ãç 究ããŸãã
- ãã¹ãŠã®ä»®èª¬ã®ç¢ºçãã¯ã©ã¹ååžãç 究ããŸãã
æµå¯Ÿçæ»æ
ã¢ããªã³ã°ã®ãšã©ãŒã¯èª°ã«ãšã£ãŠã代åã䌎ããŸãããã®ä»£è¡šçãªäŸãã©ã€ã³ãã«ããšããŽãã®ç 究ã§ããã
ã©ã® ML ãã¯ãããžãŒã«ããå®è£ ããå®è£ ãŸã§ã®ç¬èªã®ã©ã€ããµã€ã¯ã«ããããŸãã æ°ããæ¹æ³è«ã®ç®æšã¯ãã¢ãã«ã®å¯¿åœã®å段é㧠XNUMX ã€ã®åºæ¬ååããã§ãã¯ããããšã§ãã
çµæïŒ
- ML ã¢ãã«ã®ä¿¡é Œæ§ãé«ããããã«ãããã€ãã®ãããžã§ã¯ããéçºãããŠããŸãã ããã¯ãããšãã°ãdeeptune (ãªã³ã¯å
:
github.com/ChrisCummins/paper-end2end-dl ); - ãã®æ¹æ³è«ãããã«çºå±ãããã«ã¯ãML åéã®åºçç©ã®å質ãå€§å¹ ã«åäžãããå¿ èŠããããŸãã
- æ©æ¢°åŠç¿ã«ã¯ãæè¡åéãšäººæç§åŠåéã®äž¡æ¹ã§åŠéçãªãã¬ãŒãã³ã°ãšå°éç¥èãæã€ãªãŒããŒãå¿ èŠã§ãã
ãæ©æ¢°åŠç¿ã«ãã人éã®è¡åã¢ããªã³ã°: æ©äŒãšèª²é¡ãããªã¢ã»Mã»ãªãªããŒãã¢ã«ããŒãã»ã¢ãªã»ãµã©ãŒ
人éã®è¡åã®ã¢ãã«åããã®æè¡çåºç€ãå¿çšã®å±æã«ç¹åããè¬çŸ©ã
人éã®è¡åã¢ããªã³ã°ã¯æ¬¡ã®ããã«åé¡ã§ããŸãã
- å人ã®è¡å
- å°äººæ°ã®ã°ã«ãŒãã®è¡å
- éå£è¡å
ãããã®ã¿ã€ãã¯ãããããML ã䜿çšããŠã¢ãã«åã§ããŸãããå ¥åæ å ±ãšç¹åŸŽã¯ãŸã£ããç°ãªããŸãã åã¿ã€ãã«ã¯ãåãããžã§ã¯ããééããç¬èªã®å«ççåé¡ããããŸãã
- å人ã®è¡å - å人æ å ±ã®çé£ããã£ãŒããã§ã€ã¯ã
- 人ã ã®ã°ã«ãŒãã®è¡å - å¿ååã®è§£é€ã移åãé»è©±ãªã©ã«é¢ããæ å ±ã®ååŸã
å人ã®è¡å
äž»ã«ã人éã®ææ ãšåå¿ã®èªèã§ããã³ã³ãã¥ãŒã¿ãŒ ããžã§ã³ã®ãããã¯ã«é¢é£ããŠããŸãã ãããããæèãæéããããã¯åœŒèªèº«ã®ææ ã®å€åã®çžå¯Ÿçãªå°ºåºŠã«ãã£ãŠã®ã¿ã ããã ãã®ã¹ã©ã€ãã¯ãå°äžæµ·å¥³æ§ã®ææ ã¹ãã¯ãã«ã®æèã䜿çšããŠã¢ã ãªã¶ã®ææ ãèªèããããšã瀺ããŠããŸãã çµæã¯ãåã³ã®ç¬ã¿ã§ããã軜èãšå«æªæã䌎ããŸããã ãã®çç±ã¯ããããããäžç«çãªãææ ãå®çŸ©ããæè¡çãªæ¹æ³ã«ãããŸãã
å°äººæ°ã®ã°ã«ãŒãã®è¡å
ãããŸã§ã®ãšããææªã®ã¢ãã«ã¯æ å ±äžè¶³ã«ãããã®ã§ãã äŸãšããŠ2018幎ãã2019幎ã®äœåãå±ç€ºããŸããã æ°å人 X æ°åã®ãã㪠(100k++ ç»åããŒã¿ã»ãããåç §)ã ãã®ã¿ã¹ã¯ãæé©ã«ã¢ãã«åããã«ã¯ãã§ããã°é«åºŠèšã枩床èšããã€ã¯é²é³ãªã©ã®ã»ã³ãµãŒããã®ãã«ãã¢ãŒãã«æ å ±ãå¿ èŠã§ãã
éå£è¡å
顧客ãåœé£ãšå€ãã®å·ã§ãããããæãçºå±ããåéã§ãã å±å€ç£èŠã«ã¡ã©ãé»è©±å¡ããã®ããŒã¿ïŒè«æ±æžãSMSãé話ãå·å¢éã®ç§»åã«é¢ããããŒã¿ïŒããããã¹ãŠãã人ã ã®ç§»åãšç€ŸäŒçäžå®å®æ§ã«ã€ããŠã®éåžžã«ä¿¡é Œæ§ã®é«ãç»åãæäŸããŸãã ãã®æè¡ã®æœåšçãªçšé: ç·æ¥æã®æå©æŽ»åãæŽå©ãäœæ°ã®ã¿ã€ã ãªãŒãªé¿é£ã®æé©åã 䜿çšãããã¢ãã«ã¯äž»ã«ãŸã ååã«è§£éãããŠããŸããããããã¯ããŸããŸãª LSTM ãšç³ã¿èŸŒã¿ãããã¯ãŒã¯ã§ãã åœé£ã欧å·äŒæ¥ã«ç 究ã«å¿ èŠãªå¿ååããŒã¿ã®å ±æã矩åä»ããæ°æ³ã®å¶å®ãæ±ããŠãããŒæŽ»åãè¡ã£ãŠãããšã®çãçºèšããã£ãã
ãã·ã¹ãã 1 ããã·ã¹ãã 2 ãã£ãŒãã©ãŒãã³ã°ãžãããšã·ã¥ã¢ ãã³ãžãª
ãžã§ã·ã¥ã¢ã»ãã³ãžãªæ°ã®è¬çŸ©ã§ã¯ããã£ãŒãã©ãŒãã³ã°ãšç¥çµç§åŠãç®æšèšå®ã®ã¬ãã«ã§åºäŒããŸãã
ãã³ãžãªæ°ã¯ãããŒãã«è³åè³è
ãããšã«ã»ã«ãŒããã³æ°ã®æ¹æ³è«ã«åŸã£ãŠãXNUMXã€ã®äž»èŠãªã¿ã€ãã®åé¡ãç¹å®ããŠããŸãïŒèæžã
ã¿ã€ã 1 - ã·ã¹ãã 1ãç§ãã¡ããèªåçã«ãè¡ãç¡æèã®è¡å (å€ä»£ã®è³): 銎æã¿ã®ããå Žæã§è»ãé転ãããæ©ããé¡ãèªèããã
ã¿ã€ã 2 - ã·ã¹ãã 2ãæèçãªè¡å (倧è³ç®è³ª)ãç®æšèšå®ãåæãæèãè€åã¿ã¹ã¯ã
AI ã¯ãããŸã§ã®ãšãããæåã®ã¿ã€ãã®ã¿ã¹ã¯ã§ã®ã¿ååãªé«ã¿ã«å°éããŠããŸãããç§ãã¡ã®èª²é¡ã¯ãAI ã XNUMX çªç®ã®ã¿ã€ãã®ã¿ã¹ã¯ã«åŒãäžããåŠéçãªæäœãå®è¡ããè«çãšé«åºŠãªèªç¥ã¹ãã«ã§åäœããããã«æããããšã§ãã
ãã®ç®æšãéæããããã«ã次ã®ããšãææ¡ãããŠããŸãã
- NLP ã¿ã¹ã¯ã§ã¯ãæèãã¢ãã«åããããã®éèŠãªã¡ã«ããºã ãšããŠæ³šæã䜿çšããŸãã
- ã¡ã¿åŠç¿ãšè¡šçŸåŠç¿ã䜿çšããŠãæèãšãã®äœçœ®ã«åœ±é¿ãäžããç¹åŸŽãããé©åã«ã¢ãã«åãããããã«åºã¥ããŠããé«ãã¬ãã«ã®æŠå¿µã§ã®æäœã«é²ã¿ãŸãã
çµè«ã®ä»£ããã«ãããã§æåŸ
è¬æŒã玹ä»ããŸããBengio ã¯ãæé©ååé¡ãSOTAãæ°ããã¢ãŒããã¯ãã£ãè¶
ã㊠ML ã®åéãæ¡å€§ããããšããŠããå€ãã®ç§åŠè
ã® XNUMX 人ã§ãã
æèã®åé¡ãæèãžã®èšèªã®åœ±é¿ãç¥çµçç©åŠãã¢ã«ãŽãªãºã ãã©ã®çšåºŠçµã¿åããã£ãŠãå°æ¥ç§ãã¡ãåŸ
ã£ãŠããã人éãšåãããã«ãèãããæ©æ¢°ãžã®ç§»è¡ãå¯èœã«ããã®ããšããçåã¯ãŸã æ®ã£ãŠããªãã
ããããšãããããŸãïŒ
åºæïŒ habr.com