人工èžè¡å®¶
v,1.3 åŠå€å±ä»æ§
ç®æ¬¡
- æŠèŠãã¯ããã«
- ãŸããããå¶äœåæ©
- åµé æ§ãåµé ããããšã«ã€ããŠ
- ã¢ãŒãã£ã¹ãã®å®çŸ©
- æ©æ¢°ãã¡ã®èžè¡
- ãŸã èªãããããšããã-æ©æ¢°éã«ãŸã€ããæ§ã
ãªè©±
- ã¬ã³ãã©ã³ãã®âæ°äœâ
- 17æ³ã®äººå·¥ç¥èœ
- 人工ç¥èœãšã¯
- å®éšæé
- AIã¯äžèœãããªã- åºç€çšèªã®èª¬æ
- å®éã®å®éš
- æ©æ¢°ã«åµé æ§ãç§»æ€ããæ¹æ³ãæ³åããŠã¿ã
- çµããã«
- è¬èŸ
- åèæç®ãbibliography
- ãªã¹ã¹ã¡äœå
æŠèŠãã¯ããã«
ãã®æç« ã¯ããã®äœåã«ã€ããŠããå°ã詳ããç¥ããã人ã®ããã®ãã®ã§ããååã«äœåãšç ç©¶ã®ã³ã³ã»ãããåŸåã«å®éã®å®éšæé ã瀺ããŠããŸããã³ã³ã»ããããæèã®è«žåŠã«ã€ããŠã¯ååãèªãäºã§8岿¹æžãããã«ãªã£ãŠããŸãã
ã€ãã§ã«èšããŸããšãæ©æ¢°åŠç¿ãAIã®ããšãããç¥ããªã人ãåãã£ãæ°ã«ãªã£ãŠããããšå¬ããã§ãã
ïŒ. ãŸãããã補äœåæ©
åãã®åæ©ã¯æ¥µããŠäžçŽãªãã®ã§ããèªåã§äœåãäœãã®ã¯ãšãŠããã©ã€ã®ã§ãããã«æ©æ¢°ã«ãã£ãŠããããããªã©ãšèšãèããããã®èšç»ã¯å§ãŸããŸããã
âŠåè«ã§ã¯ãããŸããã倧çé¢ç®ã§ãããšã«ããäœåãäœããšããã®ã¯ããçš®ã®äººéã«ãšã£ãŠéåžžã«ã€ããããšã§ãç£ã¿ã®èŠãã¿ããšããèšãåãããããããã§ããæéã粟ç¥åãäœåã ã£ãŠæ¶èããŸãã å··ã«æº¢ããã€ã©ã¹ããäžãäºæéã§ãã£ãã£ã£ãšæãããã®ã ãšæã£ãŠããæ¹ã
ãæã
ããããããã§ãããããã人ã«ã¯äžåºŠãçœçŽããäžæã®çµµãåºæ¥äžãããŸã§ãããŒã«ããçåéã§äžã€ç¬ããèš±ãã¬ç¶æ³ã§èŠãŠããããããã®ã§ãããããåè«ã§ããã
ãé¢åäºãåé¿ããããã®é¢åãæããŸãªãããç®æšãšããç§ããèªåã§äœåãäœããªããªã代ããã«äœåãäœãã·ã¹ãã ãäœãã°è¯ããšèããã®ã¯èªç¶ãªæµãã§ããã2016幎æ«ã¯ãŸãã«ãè¶ã®éã«ãAIããŒã ãæ¥ãŠãããæµè¡ãã«ä¹ã£ããæ°æã¡ã§âAIâã«æãåºããŸããããããå§ãŸãã®ç¬¬äžæ©ã§ãã
æ©æ¢°ã䜿ã£ãŠèªåã®ä»£ããã«äœåãäœã£ãŠããããšã¯èŠããã«ãèªåã®äœåãäœãã¯ããŒã³ãã®ãããªãã®ãäœãããšã«ãªããŸãããããããã®ãäœã£ããšããŠããŒã¹ã¯èªåã§ããããã©ããªè²ã奜ãããã©ããªã¢ããŒãããã©ããªæ§å³ã奜ãããªã©ããããäºæ³ãã€ããŸããããã¯å°ãããããããããŸãããæ¹åãå€ããŠã¯ããŒã³ã§ã¯ãªãèªåã®èšé²ãããŒã¹ã«ããŠä»äººã®èšé²ãå°ãæ··ããããæ©æ¢°ç¬ç¹ã®çã®ãããªãã®ãçãããŠæ©æ¢°ç¬èªã®ã¢ãŒããäœã£ãŠãããããã«èšç»ãç·Žãããšã«ããŸããã
ããããŠçŽäœæ²æããããæ©æ¢°ã«åµé å(ã¯ãªãšã€ãã£ããã£)ãæãããããšããç®æšãèšå®ãã人éã®åµé æ§ã«å¯Ÿãã宣æŠåžåãšããŠåæ¥å¶äœã®äœåãäœãããšãšããŸããã
åµé æ§ãåµé ããããšã«ã€ããŠ
åµé æ§ïŒã¯ãªãšã€ãã£ããã£ïŒãšã¯äœã§ããããã
ããã¯çŽ çŽã«èŸæžãåŒãããšã«ããŸãã
"æ°å¥ã§ç¬èªãã€çç£çãªçºæ³ãèãåºãããšããŸãã¯ãã®èœåãåµé æ§ã«ã€ããŠã¯ããŸããŸãªç ç©¶ãè¡ãããŠããããããŸã ã«ãã®æ¬æ
ã«ã€ããŠæå¿«ãªçµè«ã¯åŸãããŠããªããããããããã€ãã®æ³šç®ãã¹ãç¹åŸŽã¯ã¿ãããã髿¬¡ã®åµé æ§ââããšãã°æ°åŠäžã®çºæã»çºèŠã®éçšã¯å
žåã§ãããæ¯èŒçåã«ã¯ãŸã£ãäœçžã远ã£ãŠé²è¡ããã..."ïŒä»¥äžç¥ïŒ
ïŒæ¥æ¬å€§çŸç§èŸæžããããã«ããåŒçšïŒ
âæ°å¥ã§ç¬èªãã€çç£çãªåµé æ§âããã·ã³ã«æèŒãããªãã°ãããã¯ã¢ãŒãã£ã¹ããšããŠã®åœãæããããããããŸããã
ãã·ã³ã«åµé æ§ãæèŒã§ãããªããåµé æ§ã¯0ã1ãšãã£ãæååãæ°å€ã§è¡šããããšã«ãªããŸãã人é¡ã®å
ã«ããåµé ã®ã¿ãããæ°å€åã§ãããšãããïŒæã人é以å€ã®çåœãååšãæã
ãšåããããã®åµé åãæã£ããšãããïŒ
人ã«ãã£ãŠã¯ãããªã®èããããªããšèšãæ¹ãããããšã§ãããã
ãã人ã¯äººéã®åµé åããèšæ¶ã®å¿åŽãšééããããªããã®ããšãããŸãã
å¥ã®äººã¯ãäžå¯ç¥ã®äžäœååšããæãããããã®ããšãããŸãã
ãšããã§ç§ã¯ãã¿ããªãç¥ã£ãŠãAãšã¿ããªãç¥ã£ãŠãBãçµã¿åãããŠèª°ãç¥ããªãã£ãXãã€ããããšããšèããŠããŸãã
人éã®åµé æ§ã®ãªãããããã¯ã£ããããªãä»ãåµé åãæååã«ããããªã©ãç¥ããæãã¬è¡çºã ãšããããç¥ããŸãããããããã¯ããã§æ¥œããããªã®ã§ç§ã¯ãã®ãŸãŸåµé æ§ã®æ¢æ±ã«ããããããšã«ããŸãã
ã¢ãŒãã£ã¹ãã®å®çŸ©
ã¢ãŒãã£ã¹ãã¯å€§æµã0ãã10ãŸã§ãã¢ãŒãã£ã¹ãæ¬äººãäœã£ãŠããã®ã ãšèããããŠããŸããå®éãå€ãã®æ¹ãèªãã®æã§äœåãç·Žãäžããã®ã§ããäŸå€ããŸãååšããŸããããããã¢ãŒãã£ã¹ãã¯ã³ã³ã»ãããèšèšãèªåã§äœã£ãã®ã¡å®éã®è£œäœã¯ä»ã®ã¹ã¿ããã«ä»»ããŠãèªåã¯èŠæèŠæã§ææ®ãå·ããšãã£ãå
·åã«äœããŸãããšãŠã倧ããªäœåãã建ç¯ç©ãæ³åãããšåãããããã§ããããããããŠäœããããã®ã ã£ãŠãç«æŽŸãªã¢ãŒãã£ã¹ãã«ããäœåã§ãã
æ©æ¢°åŠç¿ã®åéã¯éåžžã«åºãæ·±ã(ãã£ãŒãã©ãŒãã³ã°ã ãã«)ãå
šãŠãèªåãã³ã³ãããŒã«ããã®ã¯äžå¯èœã«çããããšã§ãããã§ã«å
é§è
ãå±
ãåéã§ããã°åœŒãã®ç¥æµãåããŠããŸãã®ãæžåœã§ããããè»èŒªãåéçºããã«ã¯ããŸãã«ãæéãè¶³ããŸãããããã¯ç§ãæå°ããŠãã ãã£ãŠããææã®åã売ãã§ããâŠäœããŠã§ããç§ã¯å
šãŠãèªäœããè·äººïŒcraftsmanïŒã§ã¯ãªãåæã«æç€ºãåºãç£ç£ïŒdirectorïŒãšããŠãããã°ãªããªãã®ã§ã
æ©æ¢°éã®ããã®èžè¡
æå²ä»¥åããã¢ãŒããèžè¡ãšãããã®ã¯å€§æ¹â人éâã®æã§äœã£ãŠãä¿¡ä»°ãããã®ã ã£ãããç¥ã ã£ãããâ人éâã®ããäœããããã®ã§ããä»ã®æä»£ã ãšâ人éâãäœã£ãŠãâ人éâãè©äŸ¡ãããã®ãã»ãšãã©ãå ããŸãããã£ããèšããšäººã®äŸ¡å€èгãäœåã«é¢ããäžã§ã®å€§ããªæ±ãšãªã£ãŠãããã§ããä»ãŸã§æ°ããããªãæ°ã®ã人ã®ããã®èžè¡ããäœãããã§ãããããããŠè©±ãé£èºããŸãã1940幎代ã«ã³ã³ãã¥ãŒã¿ã®å§ç¥éãçãŸããŠãã80幎ã»ã©çµã¡ãŸãããããšã¡ãã£ãšã§100幎ã«ãªãããã§ãããããããæ©æ¢°ã«ããæ©æ¢°ã®ããã®èžè¡ãçé¢ç®ã«äœã£ãŠã¿ãŠãããããªãšæãã®ã§ãã
äœãå§ãã«äººéãã©ãããŠã絡ãã§ããã®ãã ãã¶ããã¯ã§ã¯ãããŸããããã¯ãŸã人ãªãäžäœååšãæ©æ¢°ã«èžè¡ãæããçãªè§£éã§ç®ãçãããšã«ããŸãããã£ãã¿ãŸããåååè«ãªãã§ãç³ãæããªãã§ãã ããïŒäººéãæ©æ¢°ã®äžäœååšãªããŠæã£ãŠãŸããããïŒç³ãæããªãã§ãã ããïŒïŒããïŒç³ãçªã«ïŒçªã«ïŒ
ãã®è©±ã¯èããã°èããã»ã©ãããã«ããã£ãŠãããŸãã
æ©æ¢°éã®çŸæèãšã¯ãåºæºãšã¯ãäœããã£ãŠããŠèžè¡ãšãã®ä»ãåããããèª°ãæ©æ¢°ã®äŸ¡å€èгãå®çŸ©ããŠä¿èšŒããŠãããã®ããããããæ©æ¢°ã®ããã®èžè¡ã£ãŠäœã®ããã«äœããã ãæ©æ¢°ã¯æåãšãã§ããã®ãïŒ
ããŒããå
šãåŠæ¡ãåºãŠããŸãããäžçèãç¶ããªããšãã¡ã£ãœãã§ãã
ãŸã èªãããããšããã-æ©æ¢°éã«ãŸã€ããæ§ã ãªè©±
æ¬åœã®ãšããã¯ã§ãããæŠéåŠç²Ÿéªé¢šã®éªé¢šã«ã€ããŠã§ããšããæè¿è©±é¡ã®éåã³ã³ãã¥ãŒã¿ãšãããã£ãšè²ã ãšããšãã®ãªã話ããããã£ãã®ã§ããæãæžãã®ãäžçã§7çªç®ãããã«èŠæã ã£ãã®ã§ããŸãæžãé²ããããŸããã§ãããåç¡ã
ã¬ã³ãã©ã³ãã®âæ°äœâ
2016幎ã®å§ãã«ã¬ã³ãã©ã³ãã®æ°äœçµµç»ãäœãããããšã¯ãåç¥ã§ãããããæ°ããâçºèŠãããçµµç»âã§ã¯ãããŸãããæ°äœã§ãã
人工ç¥èœãéå»ã®ã¬ã³ãã©ã³ãäœåãåŠç¿ããŠãäœåã«ãã£ãšãããç»å Žããã¢ããŒãã§çµµç»ãæããã®ã§ããèŠãããšã®ãªãæ¹ã¯äžåºŠèŠãŠãã ãããäžèšã®ãªã³ã¯ã®ãããããŒãžã«ããã®ãããã§ãã
THE NEXT RENBRANDT : https://www.nextrembrandt.com/
èŠãŸãããïŒã©ãã§ããããã±ã£ãšèŠã§äººãæããããã«ããèŠããªããšæããŸããããããã ã®ç»åãšããããã§ã¯ãªãâçµµç»âãªã®ã§çµµã®å
·ãšçè·¡ã«ãã衚é¢ã®å¹åžãåçŸãããŠãããå®ç©ã¯çµµç»ã®ãã¯ã¹ãã£ãåçŸããŠãããšãã€ãŸãïŒå¹Žåã®æç¹ã§ãã·ã³ãã¯ãŒããããã°æ©æ¢°ã¯ãããŸã§ç²Ÿå·§ã«çµµãæããŠã¿ããã®ã§ãã
â»ã¬ã³ããªã³ã°ã«500æé(ããã21æ¥ã«å¹æµ)ãåºåãããçµµç»ã¯150GBã»ã©ãããå¹åžæ
å ±ã¯140Mãã¯ã»ã«ããªã©ãªããªãã«ãšã°ãæ°åã䞊ãã§ã¯ããŸãã
ããããªãã§ããïŒç§ã¯ããããšæããŸããæ©æ¢°ãã¡ã¯çµµãæãããã®ãã¯ãŒãååã«æã£ãŠããããã§ããããšã¯æ³ååããã£ãéïŒãšãŠãææ çãªã¯ãŒãã§ããïŒãäžããã°ãã£ã±ãªã¢ãŒãã£ã¹ãã«ãªããã ãããšèããŠããŸãã3Dããªã³ã¿ãCADã«è»¢çšã§ããã°åœ«å»ãäœããã§ãããã倢ãåºãããŸããïŒïŒïŒ
17æ³ã®äººå·¥ç¥èœ
ã»ãã®æ°æ¥åã11/29ã®èšäºã§é ãæ±ããããªããã®ãèŠãŸãããä»å¹Žã®ïŒïŒæã«æµ·ã®åããã§17æ³ã®é«æ ¡çã人工ç¥èœã®è«æãçºè¡šãããšã®ããšã
https://wired.jp/2017/11/29/high-schooler-shaking-up-ai/
https://s3-us-west-2.amazonaws.com/openai-assets/MLSH/mlsh_paper.pdf
15æ³ã§èªåã®ãã¥ãŒã©ã«ããããèšç·ŽããŠãã ãšããOpenAI 1 ãšããçµç¹ã§ã€ã³ã¿ãŒã³ããŠãã ãšããäºå®ã¯å°èª¬ãããå¥ãªããå°ã§è¡ãã¯ãŒããããããåºãŠããŸããçŽ æŽãããããšã ããã©äžæ¹ã§äœæ³ã幎äžã®ç§ãããã°ã©ã èšèªã®ææ³ã§ããŒããŒæ©ãã§ããã«äœãã§ããªã奎ã ãšæããšé ãçããªããŸããã
15æ³ã§ãã¥ãŒã©ã«ããããèªåšã«èšç·Žããã®ã¯ç¡çã§ãããã幎ããããã°ã©ãè·ã®äººãè¶£å³ã§ããã°ã©ã ãæžããŠããã²ãšãã¡ãèªåã®ãã¥ãŒã©ã«ããããæžããŠåãããŠéãã§ããäºäŸã¯ãããšãããããããŸãã
AIãæ·±å±€åŠç¿ãæ©æ¢°åŠç¿ã¯ãªã«ãå€§äŒæ¥ã倩æããã«ãŒã ãã䜿ãããã®ã§ã¯ãªããå®ã¯ãã®èŸºã®ïŒç§ã®ãããªïŒã¯ãœã¬ãã§ãäœãåŸããã®ãªã®ã§ãã
人工ç¥èœã¯äžèœãããªã
人工ç¥èœã¯äžèœã§ã¯ãããŸãããããæ¹ã®èšèãåããã°ãæ©æ¢°åŠç¿ã¯çŽæãéããŠããã«éããªãããšèšããŸããä¹±æŽãªèšãæ¹ãèš±ããããªãã°ãäººã®æã«è² ããªã倧éã®ããŒã¿ãã匷倧ãªãã·ã³ãã¯ãŒãšã¢ã«ãŽãªãºã ã§ãŽãªæŒããŠåæããããšããšèšããŸãã
人工ç¥èœãšããèšèã®é¿ãã®ãããã§AIãšããã®ã¯åãã€ããããã®ã£ã°ãæãããããšç¬ã£ãããã¿ãã人ãšåãããã«æèãããã®ãšèããŠãã人ãå€ãã§ãããå®éã®ãšãããããã£ãäŒè©±ç³»AIããããããã®æ
å ±ãåæããç³»AIã®ã»ããå§åçã«å€ãã¯ãã§ããããã倧æµã¯ãããç¹å®ã®æ
å ±ã«ç¹åããŠåæããããã«èšç·Žããããäžç¹ç¹ååã®ããã°ã©ã ã倧åã§ãæ ç»ãå°èª¬ã«åºãŠãããããªãã¿ãŒãããŒã¿ãŒçã«äººé¡ãè
ããããããäœã§ãã§ããã¹ãŒããŒåŒ·ãAIãçŸå®ã«ãªãã«ã¯ããã¡ãã£ãšã ããããããã§ãã
å®éã®å¶äœã«ã€ããŠã®èª¬æ
ããããã¯ãã®äœåã«ãããŠã©ããã£ãæè¡ã䜿ã£ãŠããã®ããæé ã远ã£ãŠèª¬æããŸãã
æåã«åºç€çšèªã®èª¬æãèŒããã®ã§æ©æ¢°ã«ã€ããŠã¡ãã·ãããã·ããªæ¹ã§ãé°å²æ°çšåºŠã¯æŽãã§é ããããšæããŸãã
2. åºè«ãåºç€çšèªã®èª¬æ
ç°¡åãªç¥èãçšèªã®è§£èª¬ãããŸãã以éã®æç« ã«çŽæ¥åºãŠããªãçšèªããããŸãããã®æã®è©±ã«èå³ããããªãç¥ã£ãŠãããŠæã¯ãªããšæããŸã
-
ããŒã»ãããã³
ãã¥ãŒã©ã«ãããã¯ãŒã¯ã®åºç€ããªããïŒã€ã®å ¥åãå ã«ïŒãããã¯ïŒã®åºåãè¡ãç°¡æãªæ§é ããããŸãã- çºç«
ããã§ã¯åºåä¿¡å·ãïŒã®æããçºç«ããŠããªãããïŒããçºç«ããŠããããšããŸããããŒã»ãããã³ã®åºåãïŒã ãšãçºç«ããããªã©ãšäœ¿ããŸã
- çºç«
-
æ©æ¢°åŠç¿
è±èªã§machine LearningããšæžããŸããæ·±å±€åŠç¿ãšã¯äŒŒãŠããããã§ç°ãªããã®
â ãã£ãŒãã©ãŒãã³ã°ïŒæ·±å±€åŠç¿ïŒ
AIã®èšèãšå
±ã«ãããæã
䜿ãããæ°ãããŸã
æ©æ¢°åŠç¿ãšã¯äŒŒãŠããããã§éããã®
-
ããŒã¿ã»ãã
ããã°ã©ã ã®äžã§ãšãŠãéèŠãªèŠçŽ ã§ããæåž«ãèŠæ¬ã«åœãããã®ã§ã圢åŒã¯ç»åã§ãã£ããæç« ã§ãã£ããé³å£°ãªã©æ§ã ã§ãã
ç»ååé¡ãããå Žåã¯äžã€ã®ãã¿ãŒã³ã«ã€ããŠ5äžãã10äžæã®ç»åãå¿ èŠååãšãããŠããŸãã
äŸãã°ãã«ããã®ç»åãèå¥ããããã«ã¯ããããªãã«ããã®ç»åãäžåäœã§çšæããå¿ èŠãããã£ãŠããšã§ãã -
è¡å
æ°åŠçšèªã§çžŠãšæšªã«åºãããæã€ãæ°å€ã®ãããŸãã®ãããªãã®ãæããŸãã
æ©æ¢°åŠç¿ã®åºæ¬ã¯è¡åå士ã®èšç®ã§æãç«ã£ãŠããŸãã
é¢é£ãã倧äºãªåéãšããŠç·åœ¢ä»£æ°ããã¯ãã«ããããŸãã -
é å
ããã°ã©ã çšèªãè¡åãšåããäžã€ä»¥äžã®æ°å€ã®ãããŸãããããè¡åãšé åã¯äŒŒãŠéãªããã®ãè¡åãšç°ãªãã®ã¯2次å 以äžã®åºãããæãŠãããšã§ã3次å ã5次å ã12次å ãšééãªãå¢ãããŠãããŸãã -
python
ããã°ã©ãã³ã°èšèªã®äžã€ã§ããèšç®ãåŸæãšããæ©æ¢°åŠç¿ã«é©ããèšèªãšããããŠããŸããç¡æã§å©çšã§ããŸãã
ã»ã»ã»
pythonã®å²åŠãè©©ã®åœ¢ã«ãŸãšãããThe Zen of Pythonããšããç©ããããŸããpythonã®äžã«ä»èŸŒãŸããã€ãŒã¹ã¿ãŒãšãã°ã§ãpythonã«éãããã®ã¥ãããããäžã§å¿åŸãŠããããããšã19ãæžããŠããã®ã§èª°ã§ãäžåºŠã¯èªãã§ã»ããè©©ã§ãã
åèURL:
pythonå ¬åŒãµã€ããªãã¡ã¬ã³ã¹(è±èª) - https://www.python.org/dev/peps/pep-0020/
ããã°ã©ããæã€ã¹ãå¿æ§ã (The Zen of Python) - https://qiita.com/IshitaTakeshi/items/e4145921c8dbf7ba57ef#%E6%9C%80%E5%BE%8C%E3%81%AB -
ã©ã€ãã©ãª
è¡åèšç®ãã©ãã«ä»äžãªã©ãç¹å®ã®äœæ¥ã簡䟿ã«è¡ããããã«äœãããããã°ã©ã ã®éãŸãã§ããããããã³ãŒãå ã§åŒçšããããšã§çŽ çŽã«æžãããšãããšæ°çŸè¡ããããããªæ©èœã§ã2,3è¡ã§æžããªã©ããã®æ©æµã¯åãŸãããã®ã§ããããã§ã¯èšç®çã ã«åœ¹ç«ã€ã©ã€ãã©ãªãšpythonããããããããã±ãŒãžãããŠããAnacondaãšãã䟿å©ãªãã®ã䜿ããŸãã -
CNN
Convolutional Neural NetWork ã®ç¥ã
ç³ã¿èŸŒã¿ãã¥ãŒã©ã«ãããã¯ãŒã¯ãšããããã®åã®éãç¥çµã®ä»çµã¿ãæš¡ããæ§é ãæã£ãŠããŸãã
éšåããå šäœããã¶ãåºããããªããããã¯å šäœããéšåã玡ããããæ§é ãããŠããŸãã -
DCGAN
Deep Convolutional Generative Adversarial Networks ã®ç¥
ç»åçæã«åšåãçºæ®ããã¢ã«ãŽãªãºã ã§ãã
adversarialã¯å察ã®ãæµå¯Ÿãããšãã£ãæå³ãæã¡ãDCGANã®ããŒã¹ã«ãªã£ãŠããGAN(Generative Adversarial Networks)ã¯æš¡å£è ãšèå¥è ã®2ã€ã®åœ¹å²ãäžã€ã®ã»ãããšããŠæãç«ã£ãŠããŸãã
æš¡å£è ã¯åŠç¿ããŒã¿ãããšã«ç»åãçæããŸããèå¥è ã¯ç»åãæ¬ç©ãåœç©ããå€å®ããŸãããã³ã¢ãã®ç»åãšæš¡å£è ã®äœã£ãåœç»åãèŠãŠæ¬ç©ã¯æ¬ç©ãåœç©ã¯åœç©ãšå€å®ã§ããããã«ããŸããæš¡å£è ã¯èå¥è ãæ¬ç©ãšèª€èªãããããªç»åãçæããŸãããããç¡éã«ç¹°ãè¿ãããšã§ããæ¬ç©ãããç»åãäœãã®ãGANã®ææ³ã§ãã
ããããã£ãŒãã©ãŒãã³ã°ãžè»¢çšããã®ãDCGANã§ãã
(åè)DCGAN-ãhttp://pip-install-deeplearning.hatenadiary.jp/entry/2016/01/21/232837
-
processing
ã°ã©ãã£ãã¯æç»ã«é©ããããã°ã©ã èšèªã§ããåæã®ããŒã¿ã»ããã®ã»ãŒå šãŠãprocessingã§äœããŸãããjavaããŒã¹ã -
vvvv
Processingãããããã«ã°ã©ãã£ãã¯ã«ç¹åããèšèªã§ããæŠå¿µçã§ã°ã©ãã£ã«ã«ãªå³ãäœæããããã«äœ¿çšããŸãããwindowså°çšãœãããŠã§ã¢ -
qiita
Qiitaã¯ãããã°ã©ãã³ã°ã«é¢ããç¥èãèšé²ã»å ±æããããã®ãµãŒãã¹ã§ããïŒå ¬åŒãµã€ãããåŒçšïŒ
ç°¡æãªè£ é£Ÿèšæ³ã䜿ãããµãŒãã¹ã§å匷ã«è¡ãè©°ãŸã£ãæã¯ãã°ãã°ãäžè©±ã«ãªããŸããã -
stackOverFlow, teratail
ãšãŠãä¹±æŽã«èšããŸããšãããã°ã©ãçšã®ç¥æµè¢ãã§ããã€ã³ã¿ãŒãããã®æµ·åºããšããã©ã調ã¹ãŠãæ¬ãèªãã§ãããããªãããšã¯ãããŸããããããæã«ã¯ãã¯ã人ã«èãã®ãäžçªæã£åãæ©ãããããã§ããå®éããã§ãããteratailã«è³ªåãæžã蟌ãã§å©ããä¹ãããšããïŒïŒåã§åçãæ¥ãŠãšãŠãã³ã£ããããŸããã -
transfer Leaningã«ã€ããŠ
çŽèš³ããã°ç§»è»¢åŠç¿ãšãªããŸããæ·±å±€åŠç¿ã§åŸãããåŠç¿çµæãä»ã®ããã°ã©ã ãžå¿çšïŒããªãã¡ç§»è»¢ïŒããããšããã®ãtransferLearningãšãããŠããŸã
3. äœåã®å¶äœæé ãçµæ
ããã°ã©ãã³ã°èšèªã¯python3x, 2x,ã»ãã©ã€ãã©ãªãšããŠtensorflow, numpy, matplotlibãªã©ã䜿ããŸãã
3.2 äžã€ç®ã®å¶äœããããã¿ã€ã
ãŸãã¯ããç°¡æãªåé¡ã®ããã°ã©ã ãäœããŸãã
numpyãškerasãäž»ã«å©çšããŠã³ãŒããæžããæ©æ¢°åŠç¿ã®åºæ¬æ§é ãå®è·µããããšãç®çãšã
ãŸãã
å
äººã®æ®ããã³ãŒããæš¡å£ããŠãèªåã®åŠç¿çšã®ããŒã¿ã«åã£ãåœ¢ãžæžãæããŠãããŸãã
3.2.1 ããŒã¿äœæ
䜿çšããããŒã¿ç»åã¯èªåã§æžãåºããäžžãšå€è§åœ¢ã®ã¢ãã¯ãç»åã§ãã

28*28ãã¯ã»ã«ã®ã¢ãã¯ãç»åã1ã€ã®å³åœ¢ã«ä»ã3ïœ6äžæã»ã©ãprocessingãçšããŠäœæããŸãããã©ã³ãã ãªå³åœ¢ã衚瀺ãããã¬ãŒã ããšã«ç»é¢ãpngç»åãšããŠä¿åããããã°ã©ã ã§ãã

28*28=784ã®ããŒãããå§ãŸãäžå±€ã®ã¢ãã«ãäœããŸãã
ã€ã¡ãŒãžãšããŠã¯ãã®ãããªæãã§ãã

å
¥å784-784-4åºå
ç·èšç®åæ°ã¯784784+7844=61,7792åããããšããããšã«ãªããŸãã
è€éãªããã«èŠããŸããããã£ãŠãããšã¯ç·åœããæŠã®ãããªãã®ã§ãããããã«æ°ã倧ããã®ã§ææ°ã¥ããŠããŸããŸããèŠã¯åããããªèšç®ãç¹°ãè¿ããŠãã ãã§ãããã®äžèšç®ãã®ãã®ã¯ã³ã³ãã¥ãŒã¿ã«ä»»ããã®ã§å²ãšæ¥œã§ãã
4çš®é¡ã®å³åœ¢ãããã®ã§åºåã¯4ã€ãšããŸãã
ä»åã¯ããã°ã©ã äžã®æ§é ãææ¡ããã®ãç®çãªã®ã§å°ãªãã«ã1å³åœ¢ã«ã€ã3åæã»ã©ãããã¯ã¢ããããŠãèš1.6äžæã»ã©ã§å®éšããŸãã
ç»åããŒã¿æŽåœ¢
ç»åããŒã¿ã¯ãã¯ã»ã«ãäºæ¬¡å
圢ç¶ã«äžŠãã ãã®ãªã®ã§1次å
ã®åœ¢ç¶ã«ããŸãã
éåžžã®ç»åã ãšïŒãã¯ã»ã«ã«ã€ãRãGãBã®3ã€å€ãæã£ãŠããã®ã§ãããæäœããå¿
èŠããããããªã®ã§ãããä»å䜿ãç»åã¯ã¢ãã¯ãã§1ãã¯ã»ã«ã«ã€ã1ã€ã®å€ããæããªãã®ã§ç¹ã«æäœã¯å¿
èŠãªãããã§ãã
ãŸãããã¯ã»ã«ã®å€ã¯æ¬æ¥0ïœ255ãåãã®ã§ããã0ïœ1ã®å€ã«å€æãããã®ã§ãããæŽå€ãæäœããŸãã
ç»åã®å³åœ¢ããšã«ãã©ã«ãåãããŠåãã©ã«ãããšã«å³åœ¢ãåºå¥ããããã®ã©ãã«ãæ¯ããŸãã
ãå®è¡
åŠç¿ãèšç·Žã«ã¯ReLu颿°ãšãããã®ã䜿ããŸãããããå€ãšãããã®ãèšå®ããŠå ¥åããããå€ããäœããã°0ãé«ããã°å ¥åã®å€ããã®ãŸãŸåºåãããšãã颿°ã§ãã
åç»åã®ããŒã¿ãšã©ãã«ãçŽã¥ããŠåŠç¿ããŠãçµããã«çãåããçšã®ããŒã¿ã§æ©æ¢°ãäºæ³ããã©ãã«ãšå®éã®ã©ãã«ããã£ãŠãããã©ããã¯ã€ãºãšçãåããããã®ãããé床ã§è¡ããŸãã
ãããåé¡ã®æµãã§ã
å®è¡ã§ããããã¯ãã®ã®æ°åã§çµãããŸãã

âŠã
ãªããšããããšã§ããããæ£è§£çã¯è
åšã®38ïŒ
ã§ããååãåã£ãŠãŸãããããã人ããã£ãã»ãããŸã ãŸããšãããã®ã§ãã
42 2 ã§ãã£ãããã£ãšé¢çœããã¿ã«ã§ããã®ã§ããäžéå端ãããŠãã¿ã«ããªããŸããã
ã¡ãã£ãšè©Šãã«å¥ã®ç»åã䜿ã£ãŠåãåŠç¿ããããŠã¿ãŸãããã
ã»ã»ã»

詊ãã«10åã®æ°ã®ããŒã¿ã§åŠç¿ãããŠã¿ãŸãããå
šãå€åããããŸãããã¡ãªã¿ã«åŠç¿ã«ããã£ãæéã¯10åã©ããã§ã¯ãªãã2,3æéã¯ããããŸããã
200ã®ç¹°ãè¿ãåŠç¿ïŒãšããã¯ïŒã§ã³ã¬ã ãã®æ°åãªã®ã§1000ã§ã200000åã§ã倧ããäžæã¯èŠèŸŒããªãã§ãããã
ä»ã«ãæ¹åãã¹ããšããããããŸãããã¢ãã«ãããåŠç¿ããŒã¿ã®ã»ãã«åé¡ããã£ãã§ãããããšã¯å®¹æã«æ³åã§ããŸãããã®ããšã¯æ¬¡ã®ãããã¿ã€ãã§åå ãæ¢ãããšã«ããŸãã
âŠ
æ²ãã¿ã®ããŒã¿ã»ãã
ãããªããšãèšããªããããããèŠããšãããŒã¿ã»ããã«äžåããããŸãããããã§ã¯ã§ãããã®ãã§ããŸããã
ãŸãã¯æ··ãã£ãããŒã¿ãåãé€ããããŒã¿ãšããŠãµãããããªãç»åãåãé€ããææ°ãããå°ãæžãããäžã§å床詊ãããšã«ããŸããåé¡ç»åãäžžãšäžè§ã®ïŒã€ã«æžãããŸãã

ã©ãã§ããããèŠäºã«100ïŒ
ã§ããæ°åãã£ãŠã¿ãŠ99%, 100%ãé£ç¶ã§å©ãåºããŸããã
ããã§ãããããã¹ãã§ããçµæãèŠãããªãã®ã§ããŸãéææã¯ãããŸããã確ãã«ããã§æ©æ¢°åŠç¿ãè¡ããããšæããšãšãŠãããããã§ããã
3.3 äºã€ç®ã®å®éš
å®éšãšéæã£ãŠãããªããããŒã¿ã»ãããã³ãŒããããšãããããã®ã䜿çšããŸããç³ãèš³ãªã
DCGANã䜿ã£ãtensorflowã«ããç»åçæã§ããDCGANã«ã€ããŠã¯çšèªèª¬æã§ãã軜ãè§ŠããŠããã®ã§èª¬æãçç¥ããŸãã
MNISTãšããæåãªææžãæ°åã®ããŒã¿ã»ãããã䜿ã£ãŠã³ã³ãã¥ãŒã¿ã«æ°åãæžããŠããããŸã
A tensorflow implementation of "Deep Convolutional Generative Adversarial Networks" ãhttps://github.com/carpedm20/DCGAN-tensorflow
ããŠããã§ã¯å®è¡ããŠã¿ãŸããããã©ãã«ã¯ã€ãºãããã ãã®åé¡ããã°ã©ã ãšéãç»åãæžãåºãå¿ èŠãããã®ã§ã¹ããŒãã¯åœç¶ãã¡ãã®ã»ããé ããšæãããŸãã
14epochãå®äºãããŸã§ã«ããã19æéããããŸãããæãããïŒ
19æéã®åŠç¿ã®ææãããã§ãïŒgifã¢ãã¡ãŒã·ã§ã³ïŒ
ã
è¬ã®ã¢ã€ã¢ã€ããé²ããåŸã
ã«æ°åã£ãœããã®ã圢äœãããŠããã®ããããã§ãããããå®éã«ç®ã®åã§ã ãã ãç»åãæ°åã£ãœããªã£ãŠããã®ãèŠããšæåãã®ã§ããããŸãã§åºç£ã«ç«ã¡äŒã£ãŠãããã®ãããªâŠãããªçµéšã¯ä»ã®ãšããããããšããããŸãããã
ãã®çšåºŠã®ããšã§åãã§ãããšæ¬è·ã®æ¹ãã錻ã§ç¬ããããã§ãããšã«ããå人ææã¬ãã«ã®PCã§ããããããã®ç»åçæãªãå¯èœã ãšããããšãããããŸãããããã€ã«ãªãããããããªã次ã®ã¹ããããžã®ã¢ãããŒã·ã§ã³ã«ããªããŸããã
æ©æ¢°ã«åµé æ§ãç§»æ€ããæ¹æ³ãæ³åããŠã¿ã
âŠããåã®ç« ã¯çµãã£ãŠããŸããŸãããã¯ãã人ã«ãåºãã§ããã¬ãã«ã®å®éšã¯ããã ãã§ããèšèªç¿åŸããé 匵ã£ãã®ã§èš±ããŠãããšã¯ãšãŠããšãŠãèšããŸãããã眵åã®ä»£ããã«åœ¹ç«ã€æ å ±ã圹ç«ã€æ¹è©ãæ¬ã®ãå§ããããŠããããšèº«äœããå¬ãæ¶ãéæµããŸãã(èš³:芪åã«ããŠãããŠããããšãããããŸã)
ããŠãããŸã§ã¯æ¢åã®æ¹æ³ããªãã£ãŠã³ã³ãã¥ãŒã¿ã«AIãœãããšãããŠããã£ãŠããŸããã
ã§ã¯æ©æ¢°ã«åµé æ§ãããããããã«ã¯ã©ããªæ¹æ³ããšãã°ããã®ããé倿Œ¢ãªãã«èããŠããã€ãæ¡ãåºãããšã§ãã®æã®ç· ããšãããŠããã ããŸãã
GANã®æ¹æ³ãå¿çšããŠãã£ã¹ã¯ãªãããŒã¿åœ¹ã«æã ããžã£ãŠããã
- æ©æ¢°ã«ããæ©æ¢°ã®ããã®èžè¡ã«ã€ããŠå®çŸ©ãã
- æ©æ¢°ã«æ©æ¢°èªèº«ã®æŽå²ïŒäººã®äººçã«åœãããã®ïŒãèšé²ãããå©çšãããã
- è§£ååºŠã®æŠå¿µãæããªãæ¶²æ¶ç»é¢
- ããŒã»ãããã³ã«å€ããæèåäœãäœãã人éã®ç¥çµãã¢ãã«ã«ããããŒã»ãããã³ã§ãªãããæ©æ¢°ã®äŸ¡å€èŠ³ã«æ²¿ã£ãã¢ãã«ãäœã
- æ©æ¢°ã®äŸ¡å€èŠ³ãæ©æ¢°ã®çŸæèã誰ãä¿èšŒããã®ããèãã
- ã³ã³ãã¥ãŒã¿éã®ééããèšæ¶ã®æ¬ èœãå£åãå¿åŽãæå³çã«çºçãããŠã¿ã
ã»ã»ã»ã
è¬èŸ
å
šãé¢ããã®ãªãã£ãåéã®å匷ãšèå¯ã®ãããå€ãã®æ¹ã®å©ããããã ããŸãã
(å
šçç¥)
åèæç®-bibliography
æè€åº·æ¯
: ãŒãããäœãDeep LearningââPythonã§åŠã¶ãã£ãŒãã©ãŒãã³ã°ã®çè«ãšå®è£
, ãªã©ã€ãªãŒã»ãžã£ãã³
岡åŽç« 倫: ã¯ãããŠã®ç»ååŠçæè¡ 第2ç, 森ååºç
Bill Lubanovic, æè€åº·æ¯
, é·å°Ÿé«åŒ: å
¥éPython3, ãªã©ã€ãªãŒã»ãžã£ãã³
ããŒã«ã»ã°ã¬ã¢ã : ããã«ãŒãšç»å®¶ ã³ã³ãã¥ãŒã¿æä»£ã®åµé è
ãã¡, ãªãŒã 瀟
Eã»Hã»ãŽã³ããªããž :çŸè¡ã®ç©èª ãã±ããç ,ãã¡ã€ãã³
ã±ãŽã£ã³ã»ã±ãªãŒ:ã€ã³ã¿ãŒãããã®æ¬¡ã«ãããã®, NHKåºç
ã¯ãžã©é£è¡æº: Pythonã«ããã¹ã¯ã¬ã€ãã³ã°ïŒæ©æ¢°åŠç¿ïŒ»éçºãã¯ããã¯ïŒœ,ãœã·ã
矀ãã¯æèããã€:é¡åžãã®ãª-幞倫, PHPãµã€ãšã³ã¹ã¯ãŒã«ãåºç
éæ²³ããããã€ã¯ã¬ã€ãïŒãã°ã©ã¹ã»ã¢ãã ã¹, å®ååèŠ, æ²³äºåºæåº«
TEXTãç·åœ¢ä»£æ°
髿 ¡æ°åŠã§ãããç·åœ¢ä»£æ°
èžè¡è³ãšç§åŠ
Simone Bianco ä»: Image Aesthetics, http://www.ivl.disco.unimib.it/activities/image-aesthetics/
ïŒé äžåïŒ
åæ¥è«æã»ä¿®å£«è«æãæžã人ãžãhttp://isabelle.cc.kyushu-u.ac.jp/~amano/how_to_write/
generative adversarial netsãhttp://datascienceassn.org/sites/default/files/Generative%20Adversarial%20Nets.pdf
OpenAI:Requests for Research https://openai.com/requests-for-research/#inverse-draw
æ©æ¢°éã®ããšããã£ãšç¥ã£ãŠã¿ãã -è²ããªäœå
人工ç¥èœãAIã®ç»å Žããäœåãæ©æ¢°ãšäººéã®éäºãæãããã®ããã®ä»èå¯ã®äžå©ã«ãªã£ãäœåã玹ä»ããŸãã
è°·åŽæœ€äžé: é°åœ±ç€Œè³, äžå ¬æåº«
- æå€äž»çŸ©ã®ãããããååŒãæããããšãã»ã€ãšãèšããããããŠèªãå£ã¯äžå¯§ã§10幎åŸã«ãŸãèªã¿ãããªãæç« ã§ãã
ç¥æé·å¹³: ããŸéåçç¡æèãã,æ©å·æžæ¿
ç¥æé·å¹³: æŠéåŠç²Ÿã»éªé¢šïŒæ¹ïŒ, æ©å·æžæ¿
ç¥æé·å¹³: ã°ããã©ãã¯æŠéåŠç²Ÿã»éªé¢š, æ©å·æžæ¿
- éªé¢šã·ãªãŒãºã¯åäœãªã®ã§æ¯éãªã¹ã¹ã¡ããŸãã空æŠãšãã³ãã¬AIã奜ããªè²Žæ¹ã«ã©ããã
ç¥æé·å¹³: ããªãã®éã«å®ãããã, æ©å·æžæ¿
Aã»Aã»ã¹ã¯ãšã¢: ãã©ããã©ã³ãã®åéº
- äžäººã®2次å ã®äœæ°ã3次å ã®ã®äžçãèªèããåèãããŸã§ã®èªåäŒé¢šçç·šã§ããèªåã®äžçã®äžäœã«äœçœ®ããååšãèªèããã®ãã©ãããããšãããããæŽããã£ããã«ãªããŸããäºæ¬¡å ãšèããŠå¯æã女ã®åãæ³åããæ¹ã¯å°ãèœã¡çããŠããäœåã¿ã€ãã«ãã°ã°ã£ãŠæŠèŠãèªãã§ãã ããã
æ ç»:
ã¢ããã«ã·ãŒã
楜å远æŸ
psçšã²ãŒã :
ãªã¡ã¬ããŒã¹ã, ããªãã©ããŒããžã¿ã«
ps2çšã²ãŒã :
ã¢ãŒããŒãã³ã¢3, ããã ãœãããŠã§ã¢
Z.O.E zone of the enders
ANUBIS Z.O.E zone of the enders
ps3çšã²ãŒã :
ã¢ãŒããŒãã³ã¢ ãŽã¡ãŒãã£ã¯ããã€, ããã ãœãããŠã§ã¢
psvita,3dsçšã²ãŒã :
é»è³äžçããã®è±åº



