game dev light

Transcript

1. ͜Ε͔ΒͷϏσΦήʔϜ࢈ۀʹ͍ͭͯ
   ը૾ͳ͠൛
   
   

   View Slide

2. ೥ޙΛݟਾ͑Δ
   ͜Ε͔ΒͷϏσΦήʔϜ࢈ۀʹ͍ͭͯ
   

   View Slide

3. IUUQTXXXVOSFBMFOHJOFDPNKBCMPHXFMDPNFUPVOSFBMFOHJOF
   GGXΛඪ४γΣʔμͱ͢ΔUnreal Engine 4͕ϦϦʔε͞ΕΔ
   2014೥
   Unity͕ඪ४γΣʔμΛGGXʹมߋ
   IUUQTVOJUZEDPNKQVOJUZXIBUTOFXVOJUZ
   2015೥
   

   View Slide

4. IUUQTXXXVOSFBMFOHJOFDPNKBCMPHXFMDPNFUPVOSFBMFOHJOF
   GGXΛඪ४γΣʔμͱ͢ΔUnreal Engine 4͕ϦϦʔε͞ΕΔ
   ࣍ʹϏσΦήʔϜ࢈ۀʹ޿·Δٕज़͸ಡΊΔ
   GGXʹؔ͢Δ࠷ॳͷ࿦จ
   Bruce Walter, Stephen R. Marschner, Hongsong Li, and Kenneth E. Torrance. 2007. Microfacet
   models for refraction through rough surfaces. In Proceedings of the 18th Eurographics
   conference on Rendering Techniques (EGSR'07), Jan Kautz and Sumanta Pattanaik (Eds.).
   Eurographics Association, Aire-la-Ville, Switzerland, Switzerland, 195-206. DOI=http://
   dx.doi.org/10.2312/EGWR/EGSR07/195-206
   2007೥
   Stephen McAuley, Stephen Hill, Naty Hoffman, Yoshiharu Gotanda, Brian Smits, Brent
   Burley, and Adam Martinez. 2012. Practical physically-based shading in film and game
   production. In ACM SIGGRAPH 2012 Courses (SIGGRAPH '12). ACM, New York, NY,
   USA, Article 10, 7 pages. DOI: https://doi.org/10.1145/2343483.2343493
   σΟεχʔ͕GGXΛ࢖͍ͬͯΔࣄΛެද
   2012೥
   2014೥
   Unity͕ඪ४γΣʔμΛGGXʹมߋ
   IUUQTVOJUZEDPNKQVOJUZXIBUTOFXVOJUZ
   2015೥
   

   View Slide

5. ࠓݚڀ͕੝ΜʹߦΘΕ͍ͯͯ
   ਺೥Ҏ಺ʹήʔϜ࢈ۀͰԠ༻͕޿͕Γͦ͏ͳٕज़͸
   ػցֶश
   Machine Learning
   ಛʹήʔϜ԰ͷ࢓ࣄΛେ͖͘ม͑ͦ͏ͳͷ͕
   ఢରతੜ੒ωοτϫʔΫͱਂ૚2ωοτϫʔΫ
   

   View Slide

6. https://www.reddit.com/r/zelda/comments/5olf6h/
   oc_i_fixed_my_zelda_map_size_comparison_graphic/
   θϧμͷ఻આ࣌ͷΦΧϦφ
   
   θϧμͷ఻આ෩ͷλΫτ
   
   θϧμͷ఻આ
   τϫΠϥΠτϓϦϯηε
   
   5IF&MEFS4DSPMMT7
   4LZSJN
   
   θϧμͷ఻આ
   ϒϨεΦϒβϫΠϧυ
   
   

   View Slide

7. https://www.reddit.com/r/gaming/comments/5p6db2/
   zelda_breath_of_the_wild_fair_map_comparison/
   5IF8JUDIFS8JME)VOU
   
   (SBOE5IFGU"VUP7
   
   5IF&MEFS4DSPMMT7
   4LZSJN
   
   θϧμͷ఻આ
   ϒϨεΦϒβϫΠϧυ
   
   

   View Slide

8. ͭͷήʔϜΛ࡞ΔͨΊʹ
   ࡞Βͳ͚Ε͹ͳΒͳ͍෺͸૿͍͑ͯΔ
   ͭͷήʔϜΛ࡞ΔͨΊʹ
   ࢖͏͜ͱ͕Ͱ͖ΔϦιʔε͸༗ݶͰ͋Δ
   ͍͔ʹޮ཰Α͘ήʔϜΛ࡞Δ͔͕
   ॏཁʹͳ͍ͬͯΔ
   

   View Slide

9. Coins
   $ - 7 gold pieces
   ---------- ------------ Armor
   #-........| ##...........+####### ---------- a - an uncursed +2 pair of leather gloves (being worn)
   |.........#####|.....)....| #####|........| b - an uncursed +1 robe (being worn)
   .........| ####...........-# #|........| Comestibles
   |........| # ------------# #||........| # # #-........- h - 6 uncursed apples
   ----------## # #|........| i - 8 uncursed oranges
   #### # #---------- j - 4 uncursed fortune cookies
   # # ########### l - a food ration
   # # # # m - 2 jackal corpses
   ####### # # p - a kobold corpse
   ## ## ## Scrolls
   -------# # # d - an uncursed scroll of identify
   |.f.@.-# # ## n - a scroll labeled ZLORFIK
   |...>.|# #------------# Spellbooks
   |.....| #|..........|# c - a blessed spellbook of protection
   |...... #...........-# Potions
   |.....| |..........| e - 2 uncursed potions of healing
   ------- --------|--- f - a blessed potion of healing
   # Tools
   Foo the Candidate St:15 Dx:14 Co:12 In:9 Wi:1 k - a magic marker (0:43)
   Dlvl:1 $:7 HP:14(14) Pw:5(5) AC:4 Xp:1 Stressed Gems/Stones
   o - 6 rocks
   (end)
   େ͖ͳ෺Λখ͘͞࡞Δʹ͸
   ࣗಈੜ੒
   ೥୅ॳ಄ʹ͸طʹ
   μϯδϣϯΛࣗಈੜ੒͢ΔήʔϜ͕࡞ΒΕ͍ͯͨ
   

   View Slide

10. ۙ୅తͳ%ήʔϜͰ
    ࣗಈੜ੒Λߦ͓͏ͱ͢ΔͱͲ͏ͳΔ͔
    

    View Slide

11. https://www.gdcvault.com/play/1020197/Landscape-Creation-and-Rendering-in
    ۙ୅తͳ%ήʔϜͰ
    ࣗಈੜ੒Λߦ͓͏ͱ͢ΔͱͲ͏ͳΔ͔
    ஍ܗΛਓ͕ؒ༻ҙͯ͠
    ຊ෺Β͍͠ςΫενϟͱ
    ຊ෺Β͍͠৭ͱ഑ஔͷ૲໦Λੜ੒͢Δ
    ͜Ε͚ͩͰ΋
    ෳࡶͳϧʔϧ
    Λཁ͢Δ
    

    View Slide

12. https://minecraft.net/ja-jp/
    ࢥ͍੾ͬͯࣗಈੜ੒͠΍͘͢ͳΔ੍໿ͷ΋ͱͰ஍ܗΛ࡞Δ
    ݡ͍બ୒Ͱ͸͋Δ͕
    ൚༻తʹ࢖͑Δํ๏Ͱ͸ͳ͍
    ޿େͳੈքΛ
    খ͘͞࡞Δ͜ͱʹ
    ੒ޭ͍ͯ͠Δ
    

    View Slide

13. େ͖ͳ෺Λখ͘͞࡞Δʹ͸
    ࣗಈੜ੒
    ͕༗ޮ͕ͩ
    ΠϚυΩͷήʔϜͷίϯςϯπΛ
    ࣗಈੜ੒͢ΔϧʔϧΛߟ͑Δͷ͸༰қͰ͸ͳ͍
    

    View Slide

14. ?
    ԿΒ͔ͷϧʔϧ
    ͜͏͍͏࣌ ͜͏ͳͬͯཉ͍͠
    ࣗಈੜ੒͸ԿͰ΋ϥϯμϜʹੜ੒͢Ε͹ྑ͍Θ͚Ͱ͸ͳ͘
    ଟ͘ͷ৔߹͜Μͳ෩ʹੜ੒͍ͨ͠ͱ͍͏໨ඪ͕͋Δ
    

    View Slide

15. ?
    ԿΒ͔ͷϧʔϧ
    ͜͏͍͏࣌ ͜͏ͳͬͯཉ͍͠
    ط஌ ط஌
    ະ஌
    ͍͔ͭ͘ͷࣄྫ͔Β
    ͦͷഎޙʹ͋Δϧʔϧ͕ԿͰ͋Δ͔Λߟ͑Δ໰୊ʹͳΔ
    ࣗಈੜ੒͸ԿͰ΋ϥϯμϜʹੜ੒͢Ε͹ྑ͍Θ͚Ͱ͸ͳ͘
    ଟ͘ͷ৔߹͜Μͳ෩ʹੜ੒͍ͨ͠ͱ͍͏໨ඪ͕͋Δ
    

    View Slide

16. ػցֶश
    ?
    ͜ΕΛਓ͕ؒߟ͑ΔͷͰ͸ͳ͘
    ܭࢉػΛ࢖ͬͯٻΊΑ͏ͱ͍͏ͷ͕
    ೖྗ ग़ྗ
    ͍͔ͭ͘ͷࣄྫ͔Β
    ͦͷഎޙʹ͋Δϧʔϧ͕ԿͰ͋Δ͔Λߟ͑Δ໰୊ʹͳΔ
    Machine Learning
    

    View Slide

17. ڱ͍
    ޿͍
    ఢ͕ଟ͍
    ఢ͕গͳ͍
    ෦԰ͷಛ௃
    ͷ͓खຊ͸͜͜ͰๅശΛग़͢΂͖ͩͱݴ͍ͬͯΔ
    ͷ͓खຊ͸͜͜ͰๅശΛग़͢ͷ͸ૣ͍ͱݴ͍ͬͯΔ
    ͷ৚݅Ͱ࡞ΒΕͨ෦԰ʹๅശ͸ग़Δ΂͖͔
    ๅശग़ͤ೿
    ๅശ·ͩ೿
    

    View Slide

18. ڱ͍
    ޿͍
    ఢ͕ଟ͍
    ఢ͕গͳ͍
    ෦԰ͷಛ௃
    ͳΜͱͳ͘͜ͷลͰ෼͔Εͯͦ͏͔ͩΒๅശΛग़ͦ͏
    ͷ৚݅Ͱ࡞ΒΕͨ෦԰ʹๅശ͸ग़Δ΂͖͔
    ๅശग़ͤ೿
    ๅശ·ͩ೿
    

    View Slide

19. ڱ͍
    ޿͍
    ఢ͕ଟ͍
    ఢ͕গͳ͍
    ෦԰ͷಛ௃
    αϙʔτϕΫλϚγϯ
    ૬൓͢Δछྨͷ͓खຊ͔Β࠷େͷڑ཭ΛऔΕΔͱ͜ΖͰઢҾ͖
    Support Vector Machine
    

    View Slide

20. ڱ͍
    ޿͍
    ఢ͕ଟ͍
    ఢ͕গͳ͍
    ͬͪͩͬͨ͜Β
    ๅശ͕ग़Δ
    ͬͪͩͬͨ͜Β
    ๅശ͕ग़ͳ͍
    ෦԰ͷಛ௃
    σʔλ͔Βϧʔϧ͕Ͱ͖ͨ
    

    View Slide

21. ڱ͍
    ޿͍
    ఢ͕ଟ͍
    ఢ͕গͳ͍
    ෦԰ͷಛ௃
    ৚͕݅ෳࡶʹͳΔͱ
    ͜ͷํ๏ͰϧʔϧΛੜΈग़͢ͷ͸ແཧͦ͏
    

    View Slide

22. ڱ͍
    ޿͍
    ఢ͕ଟ͍
    ఢ͕গͳ͍
    ෦԰ͷಛ௃
    ͲΜͳઢͰ෼ྨͰ͖Δ͔΋෼͔Βͳ͍ॴ͔Β
    ϧʔϧΛಘ͍ͨ
    

    View Slide

23. ৚݅෇͖ఢରతੜ੒ωοτϫʔΫΛ༻͍ͨ
    ຊ෺Β͍͠஍ܗͷੜ੒
    Éric Guérin, Julie Digne, Éric Galin, Adrien Peytavie, Christian Wolf, Bedrich Benes, and
    Benoît Martinez. 2017. Interactive example-based terrain authoring with conditional
    generative adversarial networks. ACM Trans. Graph. 36, 6, Article 228 (November 2017),
    13 pages. DOI: https://doi.org/10.1145/3130800.3130804
    ෼Ͱඳ͔Εͨεέον͔Βੜ੒͞Εͨ஍ܗ
    

    View Slide

24. ೔ຊޠͰཔΉ
    ৚݅෇͖ఢରతੜ੒ωοτϫʔΫΛ༻͍ͨ
    ຊ෺Β͍͠஍ܗͷੜ੒
    

    View Slide

25. χϡʔϥϧωοτϫʔΫ
    ೖྗ ग़ྗ
    Neural Network
    ৚݅෇͖ఢରతੜ੒ωοτϫʔΫΛ༻͍ͨ
    ຊ෺Β͍͠஍ܗͷੜ੒
    ૚
    ܗࣜχϡʔϩϯ
    

    View Slide

26. ܗࣜχϡʔϩϯ
    ೖྗ1
    ೖྗ2
    ೖྗ3
    ೖྗ4
    ೖྗ5
    w1
    w2
    w3
    w4
    w5
    શͯͷೖྗʹ͋ΔॏΈwΛ
    ͔͚ͨ΋ͷͷ࿨ΛٻΊͯ
    ͜ΕΛແ਺ʹܨ͛ͨ΋ͷ͸
    ॏΈ࣍ୈͰ
    ೚ҙͷؔ਺ΛۙࣅͰ͖Δ
    ReLU
    ग़ྗ
    ׆ੑԽؔ਺(ReLU)
    ʹ௨ͨ͠஋Λग़ྗ
    Formal Neuron
    

    View Slide

27. ͱΓ͋͑ͣద౰ͳॏΈΛઃఆͨ͠ωοτϫʔΫʹ
    ೖྗΛྲྀ͢ͱ
    ೖྗ
    ཉ͔ͬͨ͠ग़ྗͱ͸͍ͩͿҧ͏෺͕ग़ͯ͘Δ
    ≠
    ཉ͔ͬͨ͠ग़ྗ
    ग़ྗ
    

    View Slide

28. ग़ྗ͕ͲΕ͘Β͍ཉ͔ͬͨ͠෺ʹ͍͔ۙΛද͢஋Λఆٛ͢Δ
    ଛࣦؔ਺
    ໨ඪ ݁ߏࣅͯΔ ͪΐͬͱࣅͯΔ ࣅͯͳ͍
    Loss Function
    

    View Slide

29. w0k
    w1k
    w2k
    w3k
    w4k
    yk
    x0
    x1
    x2
    x3
    x4
    yk
    = R
    (∑
    n
    wnk
    xn)
    YͱX͕෼͔͍ͬͯͯZΛٻΊΔํ๏
    ∂E
    ∂wjk
    =
    ∂E
    ∂xk
    R′
    k (∑
    n
    wnk
    xn)
    xj
    XͷมԽ͕ଛࣦؔ਺ʹ༩͑ΔӨڹXʹΑΔภඍ෼
    R′
    k͸3F-6ΛLʹ͍ͭͯඍ෼ͨ͠΋ͷ
    R͸׆ੑԽؔ਺3F-6
    

    View Slide

30. ∂E
    ∂wij
    = ∑
    k
    ∂E
    ∂xk
    R′
    k (∑
    n
    wnk
    xn)
    wjk
    R′
    j (∑
    n
    wnj
    xn)
    xi
    i j k
    ࠷ޙ͔Β૚໨ͷภඍ෼͸
    ߹੒ؔ਺ͷඍ෼Λ࢖ͬͯٻΊΔ
    ઌ΄ͲٻΊͨ࠷ޙͷ૚ͷภඍ෼ͷ஋Λ࢖͑Δ
    ࿈࠯཯
    Chain Rule
    

    View Slide

31. ଛࣦؔ਺ͷ஋
    ग़ޱଆ͔Βॱ൪ʹ
    XΛม͑Δͱग़ྗ͕Ͳ͏มΘΔ͔
    ΛٻΊ͍ͯ͘
    ޡࠩٯ఻೻๏
    Back Propagation
    

    View Slide

32. ଛࣦؔ਺ͷ஋Λখ͍ͨ͘͞͠
    ଛࣦؔ਺͕ࢁͩͱ͢Δͱ
    Ұ൪௿͍ॴʹߦ͖͍ͨ
    ͨͩ͠ࢁશମͷܗ͸Θ͔Βͳ͍
    Xʹ͍ͭͯͷภඍ෼͸
    ࠓཱ͍ͬͯΔ৔ॴͷ
    ܏ࣼ
    ܏ࣼʹԊͬͯ
    ΑΓ௿͍ํ΁Ҡಈ͍͚ͯ͠͹
    Ұ൪௿͍ͱ͜ΖʹͨͲΓண͚Δؾ͕͢Δ
    ֶश
    Training
    ภඍ෼ͷ݁ՌʹԊͬͯXΛগ͠ಈ͔ͤ͹
    ໨ඪͱ͢Δؔ਺ʹগۙͮ͘͠ؾ͕͢Δ
    

    View Slide

33. ޡࠩٯ఻೻ͰXΛमਖ਼
    ͜ΕΛେྔͷσʔλʹ͍ͭͯ܁Γฦ͢ͱ
    χϡʔϥϧωοτϫʔΫ͸
    ೖྗʹରͯ͠ظ଴͢Δग़ྗ͕ಘΒΕΔؔ਺ʹ͍͍ۙͮͯ͘
    ೖྗ
    ଛࣦؔ਺
    ग़ྗ
    σʔλ͔Βؔ਺͕Ͱ͖ͨ
    

    View Slide

34. ∂E
    ∂wij
    = ∑
    k
    ∂E
    ∂xk
    R′
    k (∑
    n
    wnk
    xn)
    wjk
    R′
    j (∑
    n
    wnj
    xn)
    xi
    ஈḪΔຖʹ׆ੑԽؔ਺ͷඍ෼ֻ͕͔Δ
    ͜ͷ஋͕খ͍͞ͱ্ͷ૚·ͰXͷमਖ਼͕ಧ͔ͳ͍
    ReLU ReLUͷඍ෼
    ඍ෼ͯ͠΋஋͕খ͘͞ͳΒͳ͍
    ׆ੑԽؔ਺3F-6ͷൃݟʹΑΓ
    ௕͍ωοτϫʔΫ͕࣮༻తʹͳͬͨ
    σΟʔϓϥʔχϯά
    Deep Learning
    

    View Slide

35. ը૾ͷΑ͏ͳҐஔؔ܎Λ࣋ͬͨσʔλͰ͸
    ࿈ଓͨ͠ϐΫηϧ͕Ͳ͏ฒΜͰ͍Δ͔͸
    ॏཁͳ৘ใͰ͋Δ
    ? ΋͔ͯ͠͠ͳ
    

    View Slide

36. w13
    w23
    w33
    w43
    w53
    y3
    x1
    x2
    x3
    x4
    x5
    w93
    ʜ
    ͷΑ͏ͳԕ͘ͷϐΫηϧͱͷ઀ଓ͸
    ແବʹͳΔՄೳੑ͕ߴ͍
    w93
    

    View Slide

37. ը૾ͷہॴతͳಛ௃͸
    ը૾ʹϑΟϧλΛద༻͢ΔࣄͰऔΓग़ͤΔ
    
    
    
    ×
    ྫΤοδݕग़
    

    View Slide

38. ϑΟϧλ
    ೖྗ
    ग़ྗ
    ֶश͢Δը૾ॲཧϑΟϧλ ৞ΈࠐΈ
    Convolution
    

    View Slide

39. ͜Μͳ෩ʹܨ͍Ͱ
    χϡʔϩϯΛݮΒ͍ͯ͘͠ࣄͰ
    ը૾͔Βॏཁͳಛ௃͚ͩΛ
    औΓग़͢ࣄΛࢼΈΔ
    େྔͷೖྗ
    ݶΒΕͨग़ྗ
    ը૾ʹԿ͕ඳ͔Ε͍ͯΔ͔Λ
    ೝࣝ͢ΔΑ͏ͳ৔߹ʹΑ͘࢖͏
    ৞ΈࠐΈ
    Convolution
    

    View Slide

40. ٯ৞ΈࠐΈ
    େྔͷೖྗ
    େྔͷग़ྗ
    Ұ౓ॏཁͳಛ௃͚ͩΛऔΓग़͠
    ৄࡉͳը૾ͷ෮ݩΛࢼΈΔ
    ը૾͔Βผͷը૾Λ
    ੜ੒͢Δ৔߹ʹΑ͘࢖͏
    Deconvolution
    

    View Slide

41. ৞ΈࠐΈ
    χϡʔϥϧωοτϫʔΫ
    Convolutional Neural Network
    

    View Slide

42. ਤʹ͢Δͷ͕ਏ͍ͷͰ
    Ҏ߱χϡʔϥϧωοτϫʔΫͷ૚Λ
    ͜ͷΑ͏ʹॻ͘
    

    View Slide

43. ஍ܗͷੜ੒
    ৞ΈࠐΈ
    ٯ৞ΈࠐΈ
    ϥϯμϜͳը૾Λྲྀ͢ͱ
    ຊ෺ͬΆ͍஍ܗͷߴ͞Ϛοϓʹͳͬͯ
    ग़͖ͯͯ΄͍͠
    

    View Slide

44. ଛࣦؔ਺
    ஍ܗͷຊ෺ͬΆ͞ΛධՁ͢Δؔ਺͕ඞཁ
    Θ͔ΒΜ
    ຊ෺ͬΆ͍஍ܗ࡞Δ
    

    View Slide

45. ෱Ԭ Ѫ஌ ळా
    ਓ޻Ӵ੕ʹΑͬͯଌྔ͞Ε࣮ͨࡍͷ஍ܗσʔλ͸
    ൺֱత؆୯ʹखʹೖΕΔ͜ͱ͕Ͱ͖Δ
    

    View Slide

46. ຊ෺ͬΆ͍஍ܗ࡞Δ
    ຊ෺ͷ஍ܗ
    ੜ੒ͨ͠஍ܗ
    ஍ܗ͕ຊ෺͔൑ผ͢Δ΋͠൑ผʹࣦഊͨ͠ͳΒ
    ੜ੒ͨ͠஍ܗ͸
    ຊ෺Β͍͠஍ܗͰ͋Δ
    ఢରతੜ੒ωοτϫʔΫ
    Generative Adversarial Network
    I. J. Goodfellow, J. Pouget-Abadie, M.
    Mirza, B. Xu, D. Warde-Farley, S. Ozair,
    A. C. Courville, and Y. Bengio.
    Generative adversarial nets. In
    Proceedings of NIPS, pages 2672–
    2680, 2014. papers.nips.cc/paper/5423-
    generativeadversarial-nets.pdf.
    

    View Slide

47. ఢରతੜ੒ωοτϫʔΫ
    Generative Adversarial Network
    ຊ෺ͬΆ͍஍ܗ࡞Δ
    ஍ܗ͕ຊ෺͔൑ผ͢Δ
    ຊ෺ͷ஍ܗ
    ੜ੒ͨ͠஍ܗ
    ΋͠൑ผʹࣦഊͨ͠ͳΒ
    ੜ੒ͨ͠஍ܗ͸
    ຊ෺Β͍͠஍ܗͰ͋Δ
    ্ԼͷωοτϫʔΫͷXΛަޓʹमਖ਼͢Δ
    ্ͷωοτϫʔΫͷXΛमਖ਼͢Δࡍʹ͸ԼͷωοτϫʔΫ͸ݻఆ
    ԼͷωοτϫʔΫ͕
    ΑΓ൑ผͰ͖ͳ͍஍ܗ͕ੜ੒͞ΕΔΑ͏ʹमਖ਼Λߦ͏
    

    View Slide

48. ຊ෺ͬΆ͍஍ܗ࡞Δ
    ஍ܗ͕ຊ෺͔൑ผ͢Δ
    ຊ෺ͷ஍ܗ
    ੜ੒ͨ͠஍ܗ
    ΋͠൑ผʹࣦഊͨ͠ͳΒ
    ੜ੒ͨ͠஍ܗ͸
    ຊ෺Β͍͠஍ܗͰ͋Δ
    ԼͷωοτϫʔΫͷXΛमਖ਼͢Δࡍʹ͸্ͷωοτϫʔΫ͸ݻఆ
    ຊ෺ͷ஍ܗͱ্ͷωοτϫʔΫ͕ੜ੒ͨ͠΋ͷΛ
    ΑΓਖ਼ࣝ͘͠ผͰ͖ΔΑ͏ʹमਖ਼Λߦ͏
    ఢରతੜ੒ωοτϫʔΫ
    Generative Adversarial Network
    

    View Slide

49. ຊ෺ͬΆ͍஍ܗ࡞Δ
    ஍ܗ͕ຊ෺͔൑ผ͢Δ
    ຊ෺ͷ஍ܗ
    ΋͠൑ผʹࣦഊͨ͠ͳΒ
    ੜ੒ͨ͠஍ܗ͸
    ຊ෺Β͍͠஍ܗͰ͋Δ
    ৚݅෇͖ఢରతੜ੒ωοτϫʔΫ
    Conditional Generative Adversarial Network
    εέονੜ੒
    

    View Slide

50. ৚݅෇͖ఢରతੜ੒ωοτϫʔΫΛ༻͍ͨ
    ຊ෺Β͍͠஍ܗͷੜ੒
    Éric Guérin, Julie Digne, Éric Galin, Adrien Peytavie, Christian Wolf, Bedrich Benes, and
    Benoît Martinez. 2017. Interactive example-based terrain authoring with conditional
    generative adversarial networks. ACM Trans. Graph. 36, 6, Article 228 (November 2017),
    13 pages. DOI: https://doi.org/10.1145/3130800.3130804
    ෼Ͱඳ͔Εͨεέον͔Βੜ੒͞Εͨ஍ܗ
    

    View Slide

51. ("/Λ࢖ͬͨ%00.ͷεςʔδͷੜ੒
    (JBDPNFMMP &EPBSEPFUBMl%00.-FWFM(FOFSBUJPO6TJOH(FOFSBUJWF
    "EWFSTBSJBM/FUXPSLTz*&&&(BNFT &OUFSUBJONFOU .FEJB$POGFSFODF
    (&.
    
    
    ("/Ͱ
    ࣮ࡍͷ%ήʔϜͷεςʔδΛ
    ຊ෺Β͘͠
    ੜ੒ͨ͠ࣄྫ
    ιʔείʔυ
    https://github.com/DanieleLoiacono/DoomGAN
    ࿦จ
    https://arxiv.org/abs/1804.09154
    

    View Slide

52. %00.
    ᴈ໌ظͷ'14
    ιʔείʔυ͕ެ։͞Ε͍ͯͯվ଄͠΍͍͢ҝ
    ౓ʑݚڀऀͷ͓΋ͪΌʹͳΔ
    https://github.com/id-Software/DOOM
    

    View Slide

53. εςʔδΛ࡞Δ
    Ϛοϓ͕ຊ෺͔ࣝผ͢Δ
    ຊ෺ͷ
    %00.ͷεςʔδΛ
    ը૾ʹͨ͠΋ͷ
    ੜ੒ͨ͠εςʔδ
    ΋͠൑ผʹࣦഊͨ͠ͳΒ
    ຊ෺Β͍͠εςʔδͰ͋Δ
    ϝλσʔλ΍
    ෦԰਺౳ͷ
    ಛ௃ྔ
    

    View Slide

54. ࿦จͷஶऀʹΑΔσϞಈըIUUQTZPVUVCF,';UK21
    ΍ͬͨࣄ͕͋ΔਓͰͳ͍ͱΘ͔Γʹ͍͔͘΋͠Εͳ͍͚Ͳ
    ࣮ࡍʹ͋Γͦ͏ͳܗͷεςʔδ͕ग़དྷ্͕Δ
    

    View Slide

55. $ZDMF("/ΛԠ༻ͨ͠υοτֆͷੜ੒
    Chu Han, Qiang Wen, Shengfeng He, Qianshu Zhu, Yinjie Tan, Guoqiang Han, and Tien-
    Tsin Wong. 2018. Deep unsupervised pixelization. In SIGGRAPH Asia 2018 Technical
    Papers (SIGGRAPH Asia '18). ACM, New York, NY, USA, Article 243, 11 pages. DOI:
    https://doi.org/10.1145/3272127.3275082
    http://www.shengfenghe.com/deep-
    unsupervised-pixelization.html
    ࣸਅ΍ΠϥετΛ
    ಥͬࠐΜͩΒ
    υοτֆʹͳͬͯग़ͯ͘Δ
    ͨͩॖখ͢ΔͷͰ͸ͳ͘
    ಛ௃Λ͖ͬ͘Γ࢒ͯ͠
    ϑϥοτͳ৭࢖͍ʹͳΔ
    ͜Ε͸
    ਓ͕ඳ͍ͨ৔߹
    

    View Slide

56. A
    B
    B'
    ࣅ͍ͯΔ΄Ͳྑ͍
    ը૾܈"ͱը૾܈#͕
    ରͰରԠ͍ͯ͠Δ৔߹
    "Λೖྗͱͯ͠ಘΒΕͨੜ੒෺ͱ#ͷҧ͍Λ
    ଛࣦؔ਺ʹֶͯ͠श͢Ε͹ྑ͍
    "ͷΑ͏ͳը૾Λ#ͷΑ͏ͳը૾ʹม׵͍ͨ͠
    

    View Slide

57. $ZDMF("/
    "Λ#ʹม׵ ຊ෺ͷ#ͱࠨͷੜ੒෺Λࣝผ
    #Λ"ʹม׵
    ຊ෺ͷ"ͱӈͷੜ੒෺Λࣝผ
    A
    B
    ຊ෺ͷ#ͱੜ੒෺Λ
    ࣝผͰ͖ͳ͍ఔྑ͍
    ຊ෺ͷ"ͱੜ੒෺Λ
    ࣝผͰ͖ͳ͍ఔྑ͍
    #ˠ"ˠ#Ͱ
    ݩͱಉ͡ʹͳΔఔྑ͍
    "ˠ#ˠ"Ͱ
    ݩͱಉ͡ʹͳΔఔྑ͍
    Jun-Yan Zhu*, Taesung Park*, Phillip Isola, and Alexei A. Efros. "Unpaired
    Image-to-Image Translation using Cycle-Consistent Adversarial Networks",
    in IEEE International Conference on Computer Vision (ICCV), 2017.
    (* indicates equal contributions)
    

    View Slide

58. $ZDMF("/
    "Λ#ʹม׵ ຊ෺ͷ#ͱࠨͷੜ੒෺Λࣝผ
    #Λ"ʹม׵
    ຊ෺ͷ"ͱӈͷੜ੒෺Λࣝผ
    A
    B
    #ˠ"ˠ#Ͱ
    ݩͱಉ͡ʹͳΔఔྑ͍
    ຊ෺ͷ#ͱੜ੒෺Λ
    ࣝผͰ͖ͳ͍ఔྑ͍
    ຊ෺ͷ"ͱੜ੒෺Λ
    ࣝผͰ͖ͳ͍ఔྑ͍
    "ˠ#ˠ"Ͱ
    ݩͱಉ͡ʹͳΔఔྑ͍
    ຊ෺ͷ"ͱຊ෺ͷ#͸ରͰରԠ͍ͯ͠Δඞཁ͸ͳ͍
    "ͷྫͱ#ͷྫ͕୔ࢁ͋Ε͹૒ํ޲ͷม׵͕ಘΒΕΔ
    

    View Slide

59. υοτֆͷੜ੒
    ֦େ
    ຊ෺ͷ"ͱӈͷੜ੒෺Λࣝผ
    A
    B
    ֦େˠॖখˠλονͰ
    ݩͱಉ͡ʹͳΔఔྑ͍
    ॖখˠλονˠ֦େͰ
    ݩͱಉ͡ʹͳΔఔྑ͍
    ࣸਅ΍Πϥετ
    υοτֆ
    ݩͷ৭ͱڥքΛ
    อ͍ͯͯΔఔྑ͍
    ॖখ
    υοτֆͬΆ͍
    λονʹ͢Δ ຊ෺ͷυοτֆͱੜ੒෺Λࣝผ͢Δ
    ຊ෺ͱੜ੒෺Λ
    ࣝผͰ͖ͳ͍ఔྑ͍
    ਅٯͷૢ࡞ʹ
    ͳ͍ͬͯΔఔྑ͍
    http://www.shengfenghe.com/pixelization.html
    ͜͜ʹੜ੒ྫ͕୔ࢁ͋Δ
    

    View Slide

60. υοτֆͷੜ੒
    http://www.shengfenghe.com/deep-unsupervised-pixelization.html
    

    View Slide

61. ("/Λ୔ࢁ૊Έ߹Θͤͨ
    ౎ࢢͷσΟςΟʔϧͷੜ੒
    Tom Kelly, Paul Guerrero, Anthony Steed, Peter Wonka, and Niloy J. Mitra. 2018.
    FrankenGAN: guided detail synthesis for building mass models using style-synchonized
    GANs. In SIGGRAPH Asia 2018 Technical Papers (SIGGRAPH Asia '18). ACM, New York,
    NY, USA, Article 216, 14 pages. DOI: https://doi.org/10.1145/3272127.3275065
    ೖྗ ग़ྗ
    http://geometry.cs.ucl.ac.uk/projects/2018/frankengan/
    

    View Slide

62. ೉ॴ ૭౳ͷ഑ஔͷੜ੒ɺৄࡉͳܗঢ়ͷੜ੒ɺ৭ͷੜ੒
    ੜ੒͠ͳ͚Ε͹ͳΒͳ͍΋ͷ͕ଟ͍
    શͯͷݐ෺͕ಉ͡ݟͨ໨ʹͳ͍ͬͯͨΒෆࣗવ
    ੜ੒݁Ռʹଟ༷ੑΛ࣋ͨͤΔඞཁ͕͋Δ
    

    View Slide

63. ৚݅
    cGAN
    ཚ਺
    ৚݅ʹରԠ͢Δग़ྗ
    ֶशΛߦ͏ͨΊʹ͸
    ৚݅ͱਖ਼͍͠ग़ྗͷηοτ͕
    େྔʹඞཁ
    ৚݅෇͖ఢରతੜ੒ωοτϫʔΫͷ໰୊఺
    

    View Slide

64. GAN
    ཚ਺
    ৚݅ແ͠ఢରతੜ੒ωοτϫʔΫͷ໰୊఺
    ग़ྗΛগ͠ม͍͑ͨͱࢥͬͯ΋
    ೖྗͷཚ਺ͷͲ͕͜มΘΕ͹ཉ͍͠ग़ྗʹͳΔͷ͔
    શ͘෼͔Βͳ͍
    

    View Slide

65. ΦʔτΤϯίʔμ
    A A'
    ೖྗͱग़ྗ͕
    ࣅ͍ͯΔఔྑ͍
    ͜ͷ෦෼͕ࡉ͍ͷͰ׬શͳ৘ใΛग़ྗʹ఻͑Δࣄ͸ग़དྷͳ͍
    ݶΒΕͨ৘ใ͔ΒݩͷσʔλΛ෮ݩ͢ΔͨΊʹ
    ॏཁͳಛ௃͸ͲΕ͔Λֶश͢Δ͜ͱʹͳΔ
    

    View Slide

66. ΦʔτΤϯίʔμ
    A A'
    ॏཁͳ
    ಛ௃
    Τϯίʔμ σίʔμ
    

    View Slide

67. ཚ਺͔Β"Λੜ੒
    A'
    A
    ຊ෺ͱੜ੒෺Λ
    ࣝผͰ͖ͳ͍ఔྑ͍
    Z
    ຊ෺ͷ"ͱੜ੒͞Εͨ"Λࣝผ
    ී௨ͷ("/
    

    View Slide

68. ཚ਺͔Β"Λੜ੒
    A'
    A
    ຊ෺ͱੜ੒෺Λ
    ࣝผͰ͖ͳ͍ఔྑ͍
    Z ຊ෺ͷ"ͱੜ੒͞Εͨ"Λࣝผ
    *OGP("/
    C
    ఢରతੜ੒ωοτϫʔΫ
    C'
    ࣅ͍ͯΔఔྑ͍
    $͕"ͷॏཁͳಛ௃ʹؔ܎͍ͯ͠ΔͳΒ
    $ ;
    ˠ"ˠ$Ͱݩͷ$ʹ໭ΕΔഺ
    ΦʔτΤϯίʔμ
    $͕"ͷಛ௃Λίϯτϩʔϧ͢Δ
    ύϥϝʔλʹͳΔ
    "͔Β$Λ෮ݩ
    X. Chen, Y. Duan, R. Houthooft, J. Schulman, I.
    Sutskever, and P. Abbeel. Infogan: interpretable
    representation learning by information maximizing
    generative adversarial nets. In NIPS, 2016.
    

    View Slide

69. ཚ਺Λແࢹͯ͠Λੜ੒
    ຊ෺ͱੜ੒෺Λ
    ࣝผͰ͖ͳ͍ఔྑ͍
    Z
    ΛࣝผͰ͖ͳ͍
    Ϟʔυ่յ
    ࣝผث͕
    ݟ෼͚ΒΕͳ͍෺ͳΒ
    ͳΜͰ΋ྑ͍
    ೖྗͷཚ਺;Λແࢹͯ͠
    ৗʹಉ͡Α͏ͳσʔλΛੜ੒͢Δ
    ωοτϫʔΫ͕ग़དྷ্͕Δࣄ͕͋Δ
    Mode Collapse
    

    View Slide

70. ;͔Β"Λੜ੒
    A'
    A
    ຊ෺ͱੜ੒෺Λ
    ࣝผͰ͖ͳ͍ఔྑ͍
    ຊ෺ͷ"ͱੜ੒͞Εͨ"Λࣝผ
    7"&("/
    ఢରతੜ੒ωοτϫʔΫ
    ΦʔτΤϯίʔμ
    ࣅ͍ͯΔఔྑ͍
    ੜ੒ث͕;Λ࢖͍ͬͯΔͳΒ
    "ˠ;ˠ"Ͱݩͷ"ʹ໭ΕΔഺ
    Ϟʔυ่յΛى͜͢ͱ
    ΦʔτΤϯίʔμଆͷଛࣦ͕૿͑Δ
    "ͷॏཁͳಛ௃ΛؚΉ;Λ࡞Δ
    Rosca, Mihaela & Lakshminarayanan, Balaji
    & Warde-Farley, David & Mohamed, Shakir.
    (2017). Variational Approaches for Auto-
    Encoding Generative Adversarial Networks.
    Z
    

    View Slide

71. Z A'
    A
    ຊ෺ͱੜ੒෺Λ
    ࣝผͰ͖ͳ͍ఔྑ͍
    #JDZDMF("/
    C'
    C
    Jun-Yan Zhu, Richard Zhang, Deepak Pathak, Trevor
    Darrell, Alexei A. Efros, Oliver Wang: “Toward Multimodal
    Image-to-Image Translation”, 2017; arXiv:1711.11586
    ࣅ͍ͯΔఔྑ͍
    ࣅ͍ͯΔఔྑ͍
    D("/Ͱ৚݅Λࢦఆ
    *OGP("/͕
    ੜ੒෺ͷόϦΤʔγϣϯΛ࡞Γ
    7"&("/͕Ϟʔυ่յΛ๷͙
    

    View Slide

72. A ֶश
    ৚݅ͱͦΕΛຬͨ͢"ͷϖΞΛ࢖ͬͯ
    Τϯίʔμͱੜ੒ثΛ࡞Δ
    C
    B
    ੜ੒
    A
    "ͷॏཁͳಛ௃ͱ৚݅ͱଟ༷ੑͷૉ$͔Β
    "ͷಛ௃Λ࣋ͪ৚݅Λຬͨ͢ੜ੒෺Λ୔ࢁ࡞Δ
    #JDZDMF("/
    

    View Slide

73. ্͔Βݟͨ
    ܗঢ়
    େࡶ೺ͳ
    ݐ෺ͷܗ
    ࣮ࡍͷ
    นͷը૾
    ࣮ࡍͷ
    ԰ࠜͷը૾
    ૭౳ͷ഑ஔͷநग़
    Ԏಥ౳ͷ഑ஔͷநग़
    ԣ͔Β
    ݟͨܗঢ়
    ԰ࠜͷԜತ
    ԰ࠜͷ໛༷
    ૭ͷ഑ஔ
    นͷ໛༷
    นͷԜತ
    ߴղ૾౓ͳ
    ԰ࠜͷ໛༷
    ߴղ૾౓ͳ
    ૭ͷ໛༷
    ૭࿮ͷܗঢ়
    ߴղ૾౓ͳ
    นͷ໛༷
    ԰ࠜͷԜತ
    นͷԜತ
    ৄࡉͳ
    ݐ෺ͷ
    ܗͱ໛༷
    FrankenGAN
    

    View Slide

74. χϡʔϥϧωοτϫʔΫ͕
    ༏Εͨࣗಈੜ੒Λߦ͏ͨΊʹ͸
    ଛࣦؔ਺ͷઃܭ͕伴ʹͳΔ
    ͦΕҎ֎͍͡ΕΔͱ͜Ζ͕ແ͍ͷͰ
    ग़͖ͯͯ΄͍͠෺Λ͍͔ʹͯ͠ଛࣦؔ਺Ͱදݱ͢Δ͔ʹͳΔ
    ͱ΋ݴ͑Δ
    

    View Slide

75. ڧԽֶश
    Reinforcement Learning
    ୡ੒͍ͨ͠໨ඪʹۙͮ͘ํ๏Λֶश͢Δ
    ঢ়گΛೖྗͱͯ࣍͠ͷߦಈΛग़ྗ͢Δ
    ͨͩͦ͠ͷߦಈ͕ਖ਼͔͔ͬͨ͠͸͙͢ʹ֬ఆ͢Δͱ͸ݶΒͳ͍
    a = f(s)
    ࠓͷঢ়ଶ
    ࣍ʹ
    औΔ΂͖ߦಈ
    

    View Slide

76. ใु
    Reward
    Ձ஋
    
    ӈ୺ͷϚεʹͨͲΓண͘ͱՁ஋ͷใु͕ಘΒΕΔͱ͢Δ
    

    View Slide

77. 2ֶश
    Q Learning
    ͋ΔϚε͔ΒྡͷϚεʹҠಈͨ͠Β
    Ձ஋ͷ͓ๅ͕ஔ͍ͯ͋Δ࣌
    ͦͷҠಈʹ͸ͷՁ஋͕͋Δ
    Ձ஋
    
    

    View Slide

78. 2ֶश
    Q Learning
    Ձ஋ͷߦಈ͕Ͱ͖Δঢ়ଶʹ
    ભҠͰ͖Δߦಈʹ͸
    ΑΓগ͠গͳ͍Ձ஋͕͋Δͱߟ͑Δ
    Ձ஋
    
    Ձ஋
    
    

    View Slide

79. 2ֶश
    Q Learning
    ͍Ζ͍Ζͳঢ়ଶ͔ΒߦಈͷՁ஋ΛٻΊ͍ͯ͘
    Ձ஋
    
    Ձ஋
    
    ͜ͷૢ࡞Λे෼ͳճ਺܁Γฦ͢ͱ
    ֤ঢ়ଶͰߦ͑Δ֤ߦಈͷՁ஋͕໌Β͔ʹͳΔ
    

    View Slide

80. 2ֶश
    Q Learning
    
    ͋Δঢ়ଶ͔Β͋ΔߦಈΛߦͳͬͨ৔߹ͷՁ஋ͷද
    
    
    
    
    
    
    
    
    ঢ়ଶ ߦಈ Ձ஋
    
    
    
    
    
    
    
    
    
    ʜ
    ঢ়ଶ΍ߦಈͷ਺͕ଟ͍໰୊Ͱ͸
    ͜ͷද͸େ͖͘ͳΓ͗ͯ͢هԱͰ͖ͳ͍
    

    View Slide

81. /FVSBM'JUUFE2*UFSBUJPO
    ͦͷߦಈʹ͸
    ͜Ε͚ͩͷ
    Ձ஋͕͋Δ
    શͯͷߦಈͷՁ஋ΛهԱ͓ͯ͘͠୅ΘΓʹ
    ͋Δঢ়ଶ͔Β֤ߦಈΛߦͳͬͨ৔߹ͷՁ஋Λਪఆ͢Δ
    χϡʔϥϧωοτϫʔΫͷֶशΛߦ͏
    ࠓ͜͏͍͏ঢ়ଶͰ
    ͜͏͍͏ߦಈΛ
    ͨ͠৔߹
    Riedmiller, Martin. Neural fitted Q iteration–first experiences with a data efficient neural
    reinforcement learning method. In: European Conference on Machine Learning. Springer,
    Berlin, Heidelberg, 2005. p. 317-328.
    

    View Slide

82. ਪఆ
    ࣮ࡍ͸
    ਪఆ
    ࣮ࡍ͸
    ਪఆ
    ࣮ࡍ͸
    ਪఆ
    ࣮ࡍ͸
    Ձ஋ ʹḷΓண͚ͨ
    Ͱ΋ߦಈͷՁ஋͸
    ͷ༧૝ͱҧͬͨ
    ࠓ"ͷঢ়ଶ
    ͜͏͍͏ଛࣦؔ਺ͰֶशΛߦ͏
    ʹߦ͘ͱ͘Β͍ͷྑ͞
    ͷྑ͞͸Ͳ͜ΖͰ͸ͳ͔ͬͨ
    ࣮ࡍͱͷ
    ޡࠩ
    "
    

    View Slide

83. ਪఆ
    ࣮ࡍ͸
    ਪఆ
    ࣮ࡍ͸
    ਪఆ
    ࣮ࡍ͸
    Ձ஋ ʹḷΓண͚ͳ͍৔߹΋ಉ༷
    Ձ஋
    ʹߦ͘ͷ͕ߴՁ஋
    ʹߦ͘ͷ͸ྑ͘ͳ͍
    ࣮ࡍͱͷ
    ޡࠩ
    

    View Slide

84. Q(st
    , at
    ) ← Q(st
    , at
    ) + α (R(st
    , at
    ) + γmax (Q(st+1
    , at+1
    )) − Q(st
    , at
    ))
    XΛগ͠मਖ਼͢Δࡍͷগ͠۩߹
    ࠓͷߦಈͰಘΒΕΔใु ଍ݩʹ͕͋Ε͹ͦͷՁ஋
    
    मਖ਼લͷχϡʔϥϧωοτϫʔΫ͕ਪఆͨ͠
    ࣍ͷߦಈΛߦͳͬͯભҠͨ͠ઌͷߦಈͷՁ஋
    मਖ਼લͷχϡʔϥϧωοτϫʔΫ͕ਪఆͨ͠
    ࠓͷߦಈͷՁ஋
    ͜Εͱ͜ΕͷࠩΛখ͘͢͞ΔΑ͏ʹXΛमਖ਼͢Δ
    

    View Slide

85. Q(st
    , at
    ) ← Q(st
    , at
    ) + α (R(st
    , at
    ) + γmax (Q(st+1
    , at+1
    )) − Q(st
    , at
    ))
    ޡࠩΛܭࢉ͢ΔաఔʹχϡʔϥϧωοτϫʔΫͷग़ྗ͕ඞཁ
    ֶशத͸XΛαϯϓϧຖʹมߋ͍ͯ͠Δҝ͜ͷ஋͕҆ఆ͠ͳ͍
    ਪఆͨ͠Ձ஋͕
    ٻΊͨՁ஋ʹ͍ۙ΄Ͳྑ͍
    ࣌ʑίϐʔ
    Xͷमਖ਼͕Ձ஋ͷਪఆʹ
    ௚͙ʹ൓ө͞Εͳ͍Α͏ʹͯ͠
    ଛࣦؔ਺Λ҆ఆͤ͞Δ
    ਂ૚2ωοτϫʔΫ
    Deep Q Network
    Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou, Daan Wierstra,
    Martin Riedmiller Playing atari with deep reinforcement learning. arXiv preprint arXiv:1312.5602, 2013.
    

    View Slide

86. Vlad Firoiu, William F. Whitney, Joshua B.
    Tenenbaum Beating the World's Best at Super
    Smash Bros. with Deep Reinforcement
    Learning. arXiv preprint arXiv:1702.06230, 2017.
    ਂ૚QωοτϫʔΫΛ༻͍ͨ
    େཚಆεϚογϡϒϥβʔζDXͷAIͷ࡞੒
    https://youtu.be/dXJUlqBsZtE
    ڧԽֶश
    ϑΝϧίϯ
    ϓϩήʔϚʔ
    ϑΝϧίϯ
    "*͕ϓϩͱઓͬͯউͬͨࣄΛใ͡Δهࣄ
    https://www.businessinsider.com/super-smash-bros-
    pro-slox-mafia-artificial-intelligence-2017-2
    

    View Slide

87. ਓؒ͸ը໘ΛݟͯߦಈΛܾఆ͢Δ
    "*͸ϝϞϦΛݟͯߦಈΛܾఆ͢Δ
    ͜ͷҧ͍͕͠͹͠͹
    ΤϑΣΫτ͕දࣔ͞ΕΔલ͔Βߦಈͩ͢͠ͱ͍ͬͨ
    ཧෆਚʹڧ͍"*ΛੜΈग़͢
    

    View Slide

88. χϡʔϥϧωοτϫʔΫʹ
    ը໘͔Βঢ়ଶΛ൑அ͢Δͱ͜Ζ͔Β
    ΍ΒͤΔ͜ͱ͸ग़དྷͳ͍͔
    

    View Slide

89. Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou,
    Daan Wierstra, Martin Riedmiller Playing atari with deep reinforcement learning. arXiv
    preprint arXiv:1312.5602, 2013.
    Atari 2600ͷը໘Λೖྗͱͯ͠
    ద੾ͳߦಈΛબ୒͢ΔڧԽֶश
    

    View Slide

90. Guillaume Lample, Devendra Singh Chaplot Playing FPS Games with Deep
    Reinforcement Learning. In: AAAI. 2017. p. 2140-2146.
    ը໘Λೖྗͱͯ͠
    DOOMΛ߈ུ͢ΔڧԽֶश
    https://github.com/fhennecker/deepdoom
    https://youtu.be/L_9F9bO3-mI
    ࿦จͷஶऀʹΑΔσϞ ஶऀͱ͸ผͷਓʹΑΔ࣮૷
    

    View Slide

91. Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou,
    Daan Wierstra, Martin Riedmiller Playing atari with deep reinforcement learning. arXiv
    preprint arXiv:1312.5602, 2013.
    ը໘͔Β͙͢ʹ࣍͢΂͖ࣄ͕Θ͔ΔήʔϜͰ͸
    ڧԽֶशͷείΞ͸ਓؒͷείΞΛ௒͑Δ
    ௕ظతͳઓུΛཁ͢ΔήʔϜͰ͸ਓ͕ؒѹউ͢Δ
    

    View Slide

92. ಛผͳ஍ܗʹԊͬͨಈ͖͸
    ͦΕʹ߹ΘͤͯΩϟϓνϟΛߦͳ͓ͬͯ͘ඞཁ͕͋Δ
    ݹయతͳϞʔγϣϯΩϟϓνϟʹΑΔ
    Ξχϝʔγϣϯͷݶք
    

    View Slide

93. Daniel Holden, Taku Komura, and Jun Saito. 2017. Phase-functioned neural networks for
    character control. ACM Trans. Graph. 36, 4, Article 42 (July 2017), 13 pages. DOI: https://
    doi.org/10.1145/3072959.3073663
    http://theorangeduck.com/page/phase-functioned-neural-networks-character-control
    ϞʔγϣϯΩϟϓνϟͱ஍ܗͷϖΞͰ
    χϡʔϥϧωοτϫʔΫͷֶशΛߦ͍
    ༩͑ΒΕͨ஍ܗΛࣗવʹา͘ಈ͖Λ
    ੜ੒͢Δ
    ஍ܗ
    ஍ܗʹԊͬͨ
    า͖ํ
    

    View Slide

94. ࣍ʹલʹग़͢΂͖଍͕पظతʹʹมԽ͍ͯ͠Δ
    + +
    =
    =
    ಉ͡஍ܗͰ΋
    λΠϛϯάʹΑͬͯ
    ద੾ͳಈ͖ํ͸มΘΔ
    w(0) w(1) w(2) w(3)
    पظతʹ੾ΓସΘΔͭͷXΛ࣋ͭωοτϫʔΫͰਪఆ
    1IBTF'VODUJPOFE/FVSBM/FUXPSL
    

    View Slide

95. =
    =
    λΠϛϯάʹΑͬͯ
    ద੾ͳಈ͖ํ͸มΘΔ
    w(0) w(1) w(2) w(3)
    पظతʹ੾ΓସΘΔͭͷXΛ࣋ͭωοτϫʔΫͰਪఆ
    े෼ͳྔͷֶशσʔλΛूΊΔͨΊʹ
    ࣮ࡍʹ༷ʑͳ஍ܗΛ࣌ؒʹΘͨͬͯา͖ճͬͨ৔߹ͷಈ͖Λ
    ϞʔγϣϯΩϟϓνϟͰूΊΔ
    ΋͏ͪΐͬͱָͰ͖ͳ͍͔
    

    View Slide

96. ڧԽֶशʹΑΔาߦϞʔγϣϯͷੜ੒
    Xue Bin Peng, Glen Berseth, Kangkang Yin, and Michiel Van De Panne. 2017. DeepLoco:
    dynamic locomotion skills using hierarchical deep reinforcement learning. ACM Trans.
    Graph. 36, 4, Article 41 (July 2017), 13 pages. DOI: https://doi.org/
    10.1145/3072959.3073602
    https://www.cs.ubc.ca/~van/papers/2017-TOG-deepLoco/
    

    View Slide

97. ʮา͘ҝͷ଍ͷಈ͔͠ํ͕ະ஌ͷঢ়ଶ͔Β໨ඪ஍఺·Ͱา͚ͨΒใुʯ
    ͜Ε͸ใु͕ԕ্͗ͯ͢ख͍͔͘ͳ͍
    ? ? ?
    2ֶश͸ใुΛۮવൃݟ͢Δ·Ͱ͸
    ϥϯμϜͳࢼߦ͔͠ग़དྷͳ͍
    ਪఆ
    ਪఆ
    ใु͸ΰʔϧ͚ͩͰͳ͘
    ͦ͜·ͰͷಓఔΛ༠ಋ͢ΔΑ͏ʹ
    ഑ஔ͞Ε͍ͯΔඞཁ͕͋Δ
    ?
    

    View Slide

98. ͜Ε͸ใु͕ԕ্͗ͯ͢ख͍͔͘ͳ͍
    ·ͣԿॲΛ౿ΜͰ໨ඪʹͨͲΓண͔͚ͩ͘Λ൑அ͢Δ
    ݱࡏ஍
    ஍ܗ
    ໨ඪ
    ࢟੎
    ԿॲΛ౿Ή͔
    ʮา͘ҝͷ଍ͷಈ͔͠ํ͕ະ஌ͷঢ়ଶ͔Β໨ඪ஍఺·Ͱา͚ͨΒใुʯ
    

    View Slide

99. ݱࡏ஍
    ஍ܗ
    ໨ඪ
    ࢟੎
    ԿॲΛ౿Ή͔ ࣍ͷҰาͷҝʹͲ͏ಈ͔͘
    ࣍ͷҰาΛ
    ࢦఆ͞ΕͨҐஔͰ౿ΊͨΒ
    ใु
    ໨ඪҐஔ·Ͱ
    า͘͜ͱ͕Ͱ͖ͨΒ
    ใु
    ҰาΛਖ਼͘͠า͘͜ͱΛֶश͢ΔڧԽֶशͱ
    ໨ඪ஍఺·ͰͲͷΑ͏ͳܦ࿏Ͱา͔͘Λֶश͢ΔڧԽֶशΛ
    ૊Έ߹ΘͤΔ͜ͱͰ
    ໨ඪ஍఺·Ͱา͘ҝͷಈ͖ํΛशಘͤ͞Δ
    ඵʹճ
    ඵʹճ
    

    View Slide

100. ڧԽֶशʹΑΔө૾͔Βͷಈ͖ͷशಘ
     Xue Bin Peng, Angjoo Kanazawa, Jitendra Malik, Pieter Abbeel, and Sergey Levine. 2018.
     SFV: reinforcement learning of physical skills from videos. In SIGGRAPH Asia 2018
     Technical Papers (SIGGRAPH Asia '18). ACM, New York, NY, USA, Article 178, 14 pages.
     DOI: https://doi.org/10.1145/3272127.3275014
     

     View Slide

101. Xue Bin Peng, Angjoo Kanazawa, Jitendra Malik, Pieter Abbeel, and Sergey Levine. 2018.
     SFV: reinforcement learning of physical skills from videos. In SIGGRAPH Asia 2018
     Technical Papers (SIGGRAPH Asia '18). ACM, New York, NY, USA, Article 178, 14 pages.
     DOI: https://doi.org/10.1145/3272127.3275014
     ө૾͔Β࢟੎Λਪఆ͠
     ͓खຊͱͳΔಈ͖Λ࡞Δ
     ΩϟϥΫλʔ͕෺ཧతʹՄೳͳൣғͰ͓खຊʹ͍ۙಈ͖Λֶश͢Δ
     ෺ཧγϛϡϨʔγϣϯԼͰ
     ಈ͘ΩϟϥΫλʔ͕
     ֤ϑϨʔϜʹ͓͍ͯ
     ͓खຊʹ͍ۙϙʔζΛ͍ͯ͠Δఔ
     ߴ͍ใु
     

     View Slide

102. ڧԽֶशʹΑΔ෰ΛணΔಈ͖ͷशಘ
     Alexander Clegg, Wenhao Yu, Jie Tan, C. Karen Liu, and Greg Turk. 2018. Learning to
     dress: synthesizing human dressing motion via deep reinforcement learning.
     In SIGGRAPH Asia 2018 Technical Papers (SIGGRAPH Asia '18). ACM, New York, NY, USA,
     Article 179, 10 pages. DOI: https://doi.org/10.1145/3272127.3275048
     ෰ʹ͸෍γϛϡϨʔγϣϯΛߦ͍͍ͨ
     ࣄલʹΩϟϓνϟͨ͠ಈ͖Ͱ͸
     ର৅ͷΩϟϥΫλʔ͸
     ਖ਼͘͠෰ΛணΕͳ͍͔΋͠Εͳ͍
     https://www.cc.gatech.edu/~aclegg3/projects/LearningToDress.html
     ڧԽֶशͰΩϟϥΫλʔʹ
     ࣗྗͰ෰ΛணΔํ๏Λֶशͯ͠΋Β͏
     

     View Slide

103. r(s) = w1
     rp
     (s) + w2
     rd
     (s) + w3
     rg
     (s) + w4
     rt
     (s) + w5
     rr
     (s)
     ෰ΛணΔڧԽֶशͷใु
     rp
     (s) ਐḿใुକΛ௨աͨ͠࿹΍಄ͷ௕͞ʹൺྫ͢Δใु
     rd
     (s)
     rg
     (s)
     rr
     (s)
     ͱΓ͋͑ͣକʹ࿹Λ௨ͦ͏ͱ͢ΔΑ͏ʹ༠ಋ
     มܗใु෰͕ҾͬுΒΕͯ৳ͼ͍ͯΔ΄Ͳݮগ͢Δใु
     ෰Λഁ͘Α͏ͳಈ͖Λආ͚ΔΑ͏ʹ༠ಋ
     

     View Slide

104. r(s) = w1
     rp
     (s) + w2
     rd
     (s) + w3
     rg
     (s) + w4
     rt
     (s) + w5
     rr
     (s)
     ෰ΛணΔڧԽֶशͷใु
     rp
     (s) ਐḿใुକΛ௨աͨ͠࿹΍಄ͷ௕͞ʹൺྫ͢Δใु
     rd
     (s) มܗใु෰͕ҾͬுΒΕͯ৳ͼ͍ͯΔ΄Ͳݮগ͢Δใु
     rg
     (s) ଌ஍ใु෰͕ഽʹ઀৮͍ͯ͠Δׂ߹ʹൺྫ͢Δใु
     rr
     (s)
     ͱΓ͋͑ͣକʹ࿹Λ௨ͦ͏ͱ͢ΔΑ͏ʹ༠ಋ
     ෰Λഁ͘Α͏ͳಈ͖Λආ͚ΔΑ͏ʹ༠ಋ
     ෰Λ਎ʹ͚͍ͭͯΔঢ়ଶΛ໨ࢦ͢Α͏ʹ༠ಋ
     ΰʔϧใु෰Λணͨঢ়ଶʹࣅ͍ͯΔ΄Ͳ૿Ճ͢Δใु
     କ͔Β಄Λग़ͦ͏ͱͨ͠Γ͠ͳ͍Α͏ʹ༠ಋ
     ෰ͷछྨʹΑͬͯͷׂ߹Λௐ੔
     w
     

     View Slide

105. ڧԽֶश͕༏ΕͨߦಈΛੜ੒͢ΔͨΊ͸
     ใुͷઃܭ͕伴ʹͳΔ
     ͦΕҎ֎͍͡ΕΔͱ͜Ζ͕ແ͍ͷͰ
     ใुΛͲͷΑ͏ʹઃఆͯ͠ձಘͯ͠ཉ͍͠ߦಈʹ༠ಋ͢Δ͔ʹͳΔ
     ͱ΋ݴ͑Δ
     

     View Slide

106. ϦΞϧλΠϜϨΠτϨʔγϯά
     ޫͷ෺ཧతͳৼΔ෣͍ΛγϛϡϨʔτ͢ΔࣄͰ
     ֤ϐΫηϧͷ৭Λܾఆ͢Δ
     ͋Δ఺͕์ࣹ͢ΔޫΤωϧΪʔ͸
     ͦͷ఺͕ࣗΒൃͨ͠ޫΤωϧΪʔͱ
     ͋ΒΏΔํ޲͔Βͦͷ఺ʹ౸ୡͨ͠ޫΤωϧΪʔͷ૯࿨͔Β
     ͦͷ఺͕ٵऩͨ͠ޫΤωϧΪʔΛҾ͍ͨ΋ͷʹͳΔ
     ͜ͷܭࢉʹ͸͋ΒΏΔํ޲ʹ͍ͭͯͷੵ෼͕܁Γฦ͠ݱΕΔ
     ϞϯςΧϧϩੵ෼Ͱղ͘ͷ͕Ұൠత͕ͩ
     ϦΞϧλΠϜͱͳΔͱे෼ͳαϯϓϧ਺Λ֬อͰ͖ͳ͍ͨΊ
     ϊΠζͩΒ͚ͷϨϯμϦϯά݁Ռ͕ग़དྷ্͕Δ
     

     View Slide

107. ϊΠζͩΒ͚ͷϨϯμϦϯά݁ՌΛΦʔτΤϯίʔμʹ௨ͯ͠
     ϊΠζ͕ແ͔ͬͨ৔߹ͷը૾Λਪఆͤ͞Δ
     Chakravarty R. Alla Chaitanya, Anton S. Kaplanyan, Christoph Schied, Marco Salvi, Aaron Lefohn, Derek Nowrouzezahrai, and Timo
     Aila. 2017. Interactive reconstruction of Monte Carlo image sequences using a recurrent denoising autoencoder. ACM Trans.
     Graph. 36, 4, Article 98 (July 2017), 12 pages. DOI: https://doi.org/10.1145/3072959.3073601
     https://research.nvidia.com/publication/interactive-reconstruction-monte-carlo-image-sequences-using-recurrent-denoising
     

     View Slide

108. ࣗಈੜ੒͸༩͑ΒΕͨ৚݅ʹͰ͖Δ͚͍ͩۙ෼෍Λอͱ͏ͱ͢Δ
     ਓ͕࡞඼Λ໘ന͍ͱײ͡Δͷ͸༧૝Λཪ੾ΒΕͨ࣌Ͱ͋Δ
     ݱࡏͷػցֶश͸ήʔϜ։ൃΛܶతʹָʹ͢Δ͕
     ήʔϜͷ໘നͦ͞ͷ΋ͷΛੜΈग़͢ʹ͸ࢸ͍ͬͯͳ͍
     

     View Slide

109. ήʔϜΛߏ੒͢Δཁૉ͕
     ήʔϜʹݱ࣮ຯΛ༩͑ΔͨΊʹ࡞Δ෺ͳͷ͔
     ήʔϜΛ໘ന͘͢ΔͨΊʹ࡞Δ΋ͷͳͷ͔
     Λߟ͑Α͏
     લऀ͸ػցֶशͰੜ੒ͨ͠ํ͕ྑ͍͔΋͠Εͳ͍
     ޙऀ͸ਓ͕ؒ࡞ͬͨํ͕ྑ͍͔΋͠Εͳ͍
     

     View Slide