Slide 1

Slide 1 text

Naoki Jogan 1 Classmethod, Inc. AWS Business Unit Consulting Dev. 2018/12/5 %FFQ3BDFS%FFQ %JWF re:Growth 2018

Slide 2

Slide 2 text

ࣗݾ঺հ  • ιϦϡʔγϣϯΞʔΩςΫτ • ೥݄ೖࣾ —೥ؒ4*FSͰ"1ΤϯδχΞ • ࠷ۙ͞Θ͍ͬͯΔαʔϏε —&,4 4VNFSJBO ৓؛௚رʢδϣ΢ΨϯφΦΩʣ "84ࣄۀຊ෦ίϯαϧςΟϯά෦

Slide 3

Slide 3 text

re:Invent 2018 KeyNote 

Slide 4

Slide 4 text

ࠓ೔࿩͢͜ͱ  w%FFQ3BDFSͰԿ͕Ͱ͖Δͷ͔ɹʙNJO w%FFQ3BDFSΛͲ͏΍ͬͯಈ͔͢ͷ͔ɹʙNJO w%FFQ3BDFS͸Ͳ͏΍ֶͬͯΜͰ͍Δͷ͔ɹʙNJO

Slide 5

Slide 5 text

%FFQ3BDFSͰԿ͕Ͱ͖Δͷʁ 5

Slide 6

Slide 6 text

%FFQ3BDFSͰԿ͕Ͱ͖Δͷʁ  ڧԽֶशΛֶ΂Δ

Slide 7

Slide 7 text

%FFQ3BDFS஀ੜͷഎܠ  How can we put Reinforcement Learning in the hands of all developers?

Slide 8

Slide 8 text

%FFQ3BDFS͸ڧԽֶशΛֶͿͨΊͷखஈ  wڧԽֶशΛ։ൃऀͷखʹ౉͢ ΤΩαΠςΟϯάͳखஈ wڧԽֶशͷϋϯζΦϯΛఏڙ

Slide 9

Slide 9 text

 ڧԽֶशͱ͸ʁ

Slide 10

Slide 10 text

ػցֶशͷछྨ  UCL Course on RL Lecture 1: ntroduction to Reinforcement Learning wڭࢣ͋Γֶश wڭࢣͳֶ͠श wڧԽֶशʢ%FFQ3BDFSʣ

Slide 11

Slide 11 text

ࣗಈӡస 

Slide 12

Slide 12 text

ғޟকع 

Slide 13

Slide 13 text

 ڧԽֶशͬͯͲ͏΍Δͷʁ

Slide 14

Slide 14 text

ڧԽֶशʢྫʣ  ใु RLΞϧΰϦζϜ wΤαΛͨ͘͞Μ৯΂ͯ wఢ͕͍ͳ͍ͱ͜ΖʹਐΜͰ wγϛϡϨʔγϣϯ wใुΛ࠷େԽ

Slide 15

Slide 15 text

%FFQ3BDFSΛಈ͔ͯ͠ΈΔ 15

Slide 16

Slide 16 text

%FFQ3BDFSΛಈ͔͢·Ͱ  wεςοϓ̍ɿ3FXBSEؔ਺࡞੒ wεςοϓ̎ɿγϛϡϨʔγϣϯ wεςοϓ̏ɿ෺ཧͰ૸ΒͤΔ

Slide 17

Slide 17 text

%FFQ3BDFSΛಈ͔͢·Ͱ  wεςοϓ̍ɿ3FXBSEؔ਺࡞੒ wεςοϓ̎ɿγϛϡϨʔγϣϯ wεςοϓ̏ɿ෺ཧͰ૸ΒͤΔ

Slide 18

Slide 18 text

εςοϓ̍ɿ3FXBSEؔ਺  ใुؔ਺ʢPythonʣ ɾEJTUBODF@GSPN@DFOUFS ɹηϯλʔϥΠϯ͔Βͷڑ཭ ɾPO@USBDL ɹं྆ͷલ෦͕നઢͷ֎ଆʹ͋Δ͔Ͳ͏͔ ɾUISPUUMF ɹंͷ଎౓ɹ͸ఀࢭΛࣔ͠ɺ͸࠷ߴ଎౓ ɾUSBDL@XJEUI ɹτϥοΫ෯ ɹ ɹͳͲͳͲ Πϯϓοτ Ξ΢τϓοτ

Slide 19

Slide 19 text

εςοϓ̍ɿ3FXBSEؔ਺  NBSLFS@ USBDL@XJEUI NBSLFS@ USBDL@XJEUI NBSLFS@ USBDL@XJEUI SFXBSEF JGEJTUBODF@GSPN@DFOUFSBOEEJTUBODF@GSPN@DFOUFSNBSLFS@ SFXBSE FMJGEJTUBODF@GSPN@DFOUFSNBSLFS@ SFXBSE FMJGEJTUBODF@GSPN@DFOUFSNBSLFS@ SFXBSE FMTF SFXBSEFMJLFMZDSBTIFEDMPTFUPPGGUSBDL

Slide 20

Slide 20 text

%FFQ3BDFSΛಈ͔͢·Ͱ  wεςοϓ̍ɿ3FXBSEؔ਺࡞੒ wεςοϓ̎ɿγϛϡϨʔγϣϯ wεςοϓ̏ɿ෺ཧͰ૸ΒͤΔ

Slide 21

Slide 21 text

εςοϓ̎ɿϞσϧ࡞੒ 

Slide 22

Slide 22 text

εςοϓ̎ɿϞσϧ࡞੒ 

Slide 23

Slide 23 text

%FFQ3BDFSΛಈ͔͢·Ͱ  wεςοϓ̍ɿ3FXBSEؔ਺࡞੒ wεςοϓ̎ɿγϛϡϨʔγϣϯ wεςοϓ̏ɿ෺ཧͰ૸ΒͤΔ

Slide 24

Slide 24 text

εςοϓ̏ɿ෺ཧͰ૸ΒͤΔ  ֶशϞσϧ

Slide 25

Slide 25 text

εςοϓ̏ɿ෺ཧͰ૸ΒͤΔ 

Slide 26

Slide 26 text

"84্ͷΞʔΩςΫνϟ  SageMaker RoboMaker S3 kinesis video streams CloudWatch Logs Client

Slide 27

Slide 27 text

%FFQ3BDFS͸Ͳ͏΍ֶͬͯΜͰ͍Δͷ͔ 27

Slide 28

Slide 28 text

τϨʔχϯά͸ͲͷΑ͏ʹͯ͠ߦΘΕΔ͔ʁ  https://www.slideshare.net/AmazonWebServices/robocar-rally-2018-aim206r20-aws-reinvent-2018 ɾ45"5&ɿঢ়ଶ ɹ%FFQ3BDFSͷϑϩϯτΧϝϥ ɾ3&8"3%ɿใु ɹηϯλʔϥΠϯʹ͚ۙΕ͹(PPE ɹεϐʔυ͕ग़͍ͯΕ͹(PPE ɾ"$5*0/ɿߦಈ ɹεϐʔυΛ্͛Δ ɹӈʹۂ͕Δɺࠨʹۂ͕Δ

Slide 29

Slide 29 text

τϨʔχϯά͸ͲͷΑ͏ʹͯ͠ߦΘΕΔ͔ʁ  "$5*0/ ɾεϐʔυΛ্͛Δ ɾ௚ਐ͢Δ ɾӈʹۂ͕Δ ɾࠨʹۂ͕Δ t=0 S0

Slide 30

Slide 30 text

τϨʔχϯά͸ͲͷΑ͏ʹͯ͠ߦΘΕΔ͔ʁ  "$5*0/ ɾεϐʔυΛ্͛Δ ɾ௚ਐ͢Δ ɾӈʹۂ͕Δ ɾࠨʹۂ͕Δ t=0 S0

Slide 31

Slide 31 text

τϨʔχϯά͸ͲͷΑ͏ʹͯ͠ߦΘΕΔ͔ʁ  t=1 S "$5*0/ ɾεϐʔυΛ্͛Δ ɾ௚ਐ͢Δ ɾӈʹۂ͕Δ ɾࠨʹۂ͕Δ 1

Slide 32

Slide 32 text

τϨʔχϯά͸ͲͷΑ͏ʹͯ͠ߦΘΕΔ͔ʁ  t=1 S "$5*0/ ɾεϐʔυΛ্͛Δ ɾ௚ਐ͢Δ ɾӈʹۂ͕Δ ɾࠨʹۂ͕Δ R :ಓ֎ΕͨͷͰใुͳ͠ 1

Slide 33

Slide 33 text

τϨʔχϯά͸ͲͷΑ͏ʹͯ͠ߦΘΕΔ͔ʁ  "$5*0/ ɾεϐʔυΛ্͛Δ ɾ௚ਐ͢Δ ɾӈʹۂ͕Δ ɾࠨʹۂ͕Δ t=0 S0

Slide 34

Slide 34 text

τϨʔχϯά͸ͲͷΑ͏ʹͯ͠ߦΘΕΔ͔ʁ  "$5*0/ ɾεϐʔυΛ্͛Δ ɾ௚ਐ͢Δ ɾӈʹۂ͕Δ ɾࠨʹۂ͕Δ t=0 S0

Slide 35

Slide 35 text

τϨʔχϯά͸ͲͷΑ͏ʹͯ͠ߦΘΕΔ͔ʁ  t=1 S "$5*0/ ɾεϐʔυΛ্͛Δ ɾ௚ਐ͢Δ ɾӈʹۂ͕Δ ɾࠨʹۂ͕Δ 1

Slide 36

Slide 36 text

τϨʔχϯά͸ͲͷΑ͏ʹͯ͠ߦΘΕΔ͔ʁ  t=1 S "$5*0/ ɾεϐʔυΛ্͛Δ ɾ௚ਐ͢Δ ɾӈʹۂ͕Δ ɾࠨʹۂ͕Δ R :ঢ়گʹԠͨ͡ใु 1

Slide 37

Slide 37 text

τϨʔχϯά͸ͲͷΑ͏ʹͯ͠ߦΘΕΔ͔ʁ  https://www.slideshare.net/AmazonWebServices/robocar-rally-2018-aim206r20-aws-reinvent-2018

Slide 38

Slide 38 text

·ͱΊ 38

Slide 39

Slide 39 text

·ͱΊ  w%FFQ3BDFS͸ڧԽֶशΛֶͿͨΊͷखஈ w%FFQ3BDFS͸ڧԽֶशͱ͔ؔ܎ͳ͘୯७ʹָ͍͠ wϨʔεେձ΋͋ΔͷͰΈΜͳࢀՃ͠Α͏ʂʂΞϝϦΧͰʁ

Slide 40

Slide 40 text

40 ԶͰڧԽֶशΛ ֶΜͰ͘Εʂʂ ࠓͳΒˈ249

Slide 41

Slide 41 text

 Let's start reinforcement learning with DeepRacer

Slide 42

Slide 42 text

42