Slide 1

Slide 1 text

%FFQ3BDFSಉ޷ձ ୈೋճΦϑϥΠϯษڧձ!େࡕ େᖒ༐ే "84%FFQ3BDFSͰ·ͳͿڧԽֶश

Slide 2

Slide 2 text

ࣗݾ঺հ େᖒ༐ే w σʔλΠϯςάϨʔγϣϯ෦ػցֶशνʔϜ w ΤϯδχΞ w ೥݄ೖࣾ w લ৬ɿ6OJUZͱ͔1)1౳ͰήʔϜ։ൃ

Slide 3

Slide 3 text

ࠓ೔࿩͢಺༰ %FFQ3BDFSͰ࢖ΘΕ͍ͯΔڧԽֶश΁ͷ௒ೖ໳ wػցֶश wڧԽֶश wํࡦϕʔεͱՁ஋ϕʔεɺ"DUPS$SJUJDɺ110 w%FFQ3BDFS w֤ϋΠύʔύϥϝʔλͷҙຯ

Slide 4

Slide 4 text

ࠓ೔࿩͞ͳ͍಺༰ w਺ࣜ wใुઃܭΛ۩ମతʹͲ͏͢Ε͹͍͍ͷ͔ wϋΠύʔύϥϝʔλʔ͸۩ମతʹͲ͏ઃఆͨ͠Β͍͍ͷ͔

Slide 5

Slide 5 text

ػցֶश

Slide 6

Slide 6 text

ػցֶश ֓ཁ ͬ͘͟Γݴ͏ͱنଇੑΛֶशͤ͞Δ͜ͱɻ wڭࢣ͋Γֶश౴͑Λ༩͑ͯೖྗͱ౴͑ͷنଇੑΛֶशͤ͞Δ wڭࢣͳֶ͠शɿ౴͑Λ༩͑ͣʹೖྗͷنଇੑΛֶशͤ͞Δ w൒ڭࢣ͋Γֶशɿجຊతʹ͸౴͑Λ༩͑ͣʹೖྗͷنଇੑΛֶशͤ͞ Δ͕ɺ෦෼తʹ౴͑Λ༩͑ͯิॿ͢Δ wڧԽֶशɿ͋Δ؀ڥͰใु͕࠷େԽ͢Δߦಈબ୒ͷنଇੑΛֶशͤ͞Δ ͳͲ

Slide 7

Slide 7 text

ػցֶश χϡʔϥϧωοτϫʔΫ֓ཁ ϑΟʔυϑΥϫʔυ χϡʔϥϧωοτϫʔΫ wೖྗ wόονσʔληοτ͔ΒϥϯμϜʹநग़ wΤϙοΫ਺ಉ͡σʔληοτΛ࢖͏ճ਺ wग़ྗ wೖྗ͕//Λ௨ͬͯग़͖ͯͨ΋ͷ wྫ dͷ࣮਺ͳͲ

Slide 8

Slide 8 text

ڧԽֶश

Slide 9

Slide 9 text

ڧԽֶश ֓ཁ ؀ڥ͔ΒΤʔδΣϯτʹঢ়ଶ ΤʔδΣϯτ͕ঢ়ଶʹج͍ͮͯ ߦಈ ؀ڥ͕ߦಈ݁Ռʹج͍ͮͨใु ΛΤʔδΣϯτ΁ dΛ܁Γฦ͢ ࠷େใु͕΋Β͑ΔߦಈΛબͿΑ͏ʹֶश͢Δ

Slide 10

Slide 10 text

ڧԽֶश "84%FFQ3BDFS֓ཁ RoboMakerͰͷγϛϡϨʔγϣϯͱ SageMakerͰͷϞσϧͷֶशΛ܁Γฦ͢

Slide 11

Slide 11 text

ڧԽֶश %FFQ3BDFS༻ޠ ΤʔδΣϯτɿ%FFQ3BDFS ं ঢ়ଶɿ%FFQ3BDFSͷΧϝϥ͔Βͷը૾ ߦಈɿεςΞϦϯάͱ଎౓ͷ૊Έ߹Θͤ

Slide 12

Slide 12 text

ڧԽֶश ख๏঺հ ڧԽֶशͷख๏͸͍Ζ͍Ζ͋ΔͷͰߜͬͯ঺հ w֤ख๏ͷجຊͱͳΔߟ͑ํɺՁ஋ϕʔεͱํࡦϕʔε wݸਓతʹཧղʹۤ࿑ͨ͠"DUPS$SJUJD w%FFQ3BDFSͰ࢖ΘΕ͍ͯΔ1SPYJNBM1PMJDZ0QUJNJ[BUJPO

Slide 13

Slide 13 text

ڧԽֶश ख๏঺հ Ձ஋ϕʔε wֶश୳ࡧΛਐΊΔதͰɺ֤ঢ়ଶͱߦಈͷՁ஋ΛٻΊ ͍ͯ͘ wݱࡏͷঢ়ଶʹ͓͍ͯՁ஋͕ߴ͍ߦಈ΄Ͳબ͹Ε΍͍͢ wঢ়ଶͱߦಈͷՁ஋2 T B ࣍εςοϓҎ߱ಘΒΕΔใ ुͷ૯࿨ͷظ଴஋

Slide 14

Slide 14 text

ڧԽֶश ख๏঺հ ํࡦϕʔε wؔ਺ ํࡦؔ਺ ΛٻΊͯɺͦͷؔ਺Λ༻͍ͯ ߦಈΛબͿ wํࡦؔ਺ঢ়ଶʹదͨ͠ߦಈ ͷ֬཰ Λग़ྗ ͢Δ

Slide 15

Slide 15 text

ڧԽֶश ख๏঺հ "DUPS$SJUJD w֤ΞϧΰϦζϜͰঢ়ଶՁ஋ͱͯ͠ใुͷฏۉΛ࢖͍ͬͯ ͨͱ͜ΖΛՁ஋ؔ਺Ͱஔ͖׵͑Δ wՁ஋ؔ਺ͱํࡦؔ਺ڞʹֶशͤ͞Δ

Slide 16

Slide 16 text

ڧԽֶश ख๏঺հ $MJQQFE 1SPYJNBM1PMJDZ0QUJNJ[BUJPO w%FFQ3BDFSͷֶशʹ࢖ΘΕͯΔ wํࡦϕʔε wߋ৽લޙͷํࡦ ߦಈͷ֬཰ ͷൺΛݩʹֶश͢Δ wߋ৽ྔ͕େ͖͘ͳΓ͗͢ͳ͍Α͏ʹ্ݶͱԼݶΛܾΊͯ$MJQ͢Δ

Slide 17

Slide 17 text

ڧԽֶश ख๏঺հ 1SPYJNBM1PMJDZ0QUJNJ[BUJPO https://nervanasystems.github.io/coach/components/agents/policy_optimization/ppo.html

Slide 18

Slide 18 text

AWS DeepRacer

Slide 19

Slide 19 text

"84%FFQ3BDFS RoboMakerͰͷγϛϡϨʔγϣϯͱ SageMakerͰͷϞσϧͷֶशΛ܁Γฦ͢

Slide 20

Slide 20 text

"84%FFQ3BDFS ߦಈۭؒ w଎౓ͱਐߦํ޲ͷ૊Έ߹Θͤ ૊Έ߹Θͤ਺͕ଟ͍ͱ ಈ͖͕ॊೈʹͳΔ ҙਤͨ͠௨Γಈ͘Α͏ʹ ֶशͤ͞Δͷ͸೉͍͠

Slide 21

Slide 21 text

"84%FFQ3BDFS ใुؔ਺ w଎౓΍ҐஔͳͲͷσʔλʹج͍ͮͯใुΛ༩ ͑Δ w1ZUIPOͰࣗ༝ʹॻ͚Δ wෛͷ஋΋Մ wֶश࣌ʹ͸ઌͷεςοϓͷใु΋ߟྀ͞ΕΔ ͜ͱʹཹҙ DeepRacerʹͲ͏ಈ͍ͯ΄͍͔͠ ใुͱ͍͏ܗͰ ఆٛ͢Δ

Slide 22

Slide 22 text

"84%FFQ3BDFS ϋΠύʔύϥϝʔλ wόοναΠζҰճͷֶशʹ࢖༻͢Δσʔλ ͷ਺ wΤϙοΫ਺ֶशϑΣΠζͰͷֶशճ਺ wֶश཰Ϟσϧߋ৽࣌ͷޯ഑ͷ܎਺ ֶश͠ ΍͢͞ wΤϯτϩϐʔߦಈબ୒࣌ͷϥϯμϜ͞

Slide 23

Slide 23 text

"84%FFQ3BDFS ϋΠύʔύϥϝʔλ wׂҾ཰ઌͷεςοϓΛͲΕ͚ͩॏࢹ͢Δ͔ ઌͷ εςοϓͰͷใुʹର͢Δ܎਺ wଛࣦλΠϓՁ஋ؔ਺ͷϞσϧͷଛࣦؔ਺ΛͲ͏͢ Δ͔ wγϛϡϨʔγϣϯϑΣΠζͷΤϐιʔυ਺

Slide 24

Slide 24 text

·ͱΊ wڧԽֶशͰ΋χϡʔϥϧωοτϫʔΫ // Λ࢖ͬͯΔ΋ͷ͕ଟ͍ͷͰɺ·ͣ͸//ͰͷճؼͳͲڭࢣ͋ Γֶश͔Β࢝ΊΔͷ΋Ұͭͷख wڧԽֶशͷख๏͸͍Ζ͍Ζ͋Δ͚ͲɺํࡦϕʔεͱՁ஋ϕʔε͕جຊ w%FFQ3BDFS͸ใुؔ਺͕ॏཁ wใुؔ਺͸ओʹͲ͏͍͏ಈ͖Λֶश͍͔ͤͨ͞Ͱฦ͢ใुͷ஋ΛܾΊΔ wجຊతʹର৅εςοϓͰͷใु͚ͩͰ͸ͳ͘ɺকདྷͷεςοϓͰͷใु΋ؚΊͯධՁ͞ΕΔ wϋΠύʔύϥϝʔλ͸ֶशͷ࢓ํͷௐ੔ʹ࢖͏ͷ͕ྑ͛͞ wใु΍ϋΠύʔύϥϝʔλͷ֤஋͕Ͳ͏ֶशʹ׆͔͞ΕΔ͔ͱ͍͏ͱ͜Ζ͔ΒڧԽֶशʹೖΔͷ͕Φε εϝ ڵຯΛ࣋ͬͨํ͸ͥͻ͍Ζ͍ΖάάͬͯΈ͍ͯͩ͘͞ʔ ͦͯ͠ɺҰॹʹۤ͠Έͳ͕Βֶͼ·͠ΐʔ