定規とコンパスと ChainerRL

8efe990be3aad3b8c6bac487b8ef7b2b?s=47 horiem
June 09, 2018

定規とコンパスと ChainerRL

強化学習を使って作図問題を解く
Chainer Meetup #07, 9th Jun 2018

8efe990be3aad3b8c6bac487b8ef7b2b?s=128

horiem

June 09, 2018
Tweet

Transcript

  1. ఆنͱίϯύεͱ ChainerRL Chainer Meetup #07, 9th Jun 2018 horiem@yellowshippo

  2. ChainerRL Ͱ

  3. ࡞ਤ໰୊Λղ͖͍ͨ

  4. ࡞ਤ໰୊ • ఆنͱίϯύε͚ͩΛ࢖ͬͯ໨తͷਤܗΛඳ͘໰୊ http://mathworld.wolfram.com/GeometricConstruction.html

  5. σϞ

  6. ͷલʹ

  7. ਤͷݟํ ֶशϞσϧʹ౉͢৘ใ ʢObservationʣ ਓؒ༻ ໨తͷਤܗ ར༻Մೳͳ ఺

  8. σϞ

  9. શମ૾ ؀ڥ ΤʔδΣϯτ ߦಈ ؍ଌ

  10. શମ૾ ؀ڥ ΤʔδΣϯτ [p0_x, p0_y] [p1_x, p1_y] …… ը૾ ఺ͷ৘ใ

    [shape_flag, pi, pj] ৽͍͠ਤܗ
  11. ωοτϫʔΫΞʔΩςΫνϟ Conv MLP MLP Conv MLP [p0_x, p0_y] [p1_x, p1_y]

    …… ը૾ ఺ͷ৘ใ [shape_flag, pi, pj] ৽͍͠ਤܗ
  12. ωοτϫʔΫΞʔΩςΫνϟ (100, 100) (12, 3) Conv MLP MLP Conv MLP

    (2, 12, 12) [p0_x, p0_y] [p1_x, p1_y] …… ը૾ ఺ͷ৘ใ [shape_flag, pi, pj] ৽͍͠ਤܗ
  13. ωοτϫʔΫΞʔΩςΫνϟ (100, 100) (12, 3) Conv MLP MLP Conv MLP

    (2, 12, 12) [p0_x, p0_y] [p1_x, p1_y] …… ը૾ ఺ͷ৘ใ [shape_flag, pi, pj] ৽͍͠ਤܗ = 288
  14. ࢥͬͨ͜ͱͳͲ • ڧԽֶश΍ͬͨ͜ͱͳ͔͚ͬͨͲָ͍͠ • ChainerRL ϥΫͰΑ͍ • ߦಈۭ͕ؒେ͖͍ͷͰݮΒ͍ͨ͠ • AlphaGO

    ͕ࢀߟʹͳΔ͔΋ʁ • ίʔυ͖Ε͍ʹͨ͠Βެ։ && ղઆ͠·͢ • n ࣍ํఔࣜΛ ChainerRL Ͱղ͚Δ͔ʁ • ՝֎׆ಈ޷͖ͳਓɺҰॹʹ΍Γ·͠ΐ͏ʂ