ニューラルFM音源

畳み込みニューラルネットワークを使って与えられた音に近い音色を奏でるFM音源のパラメータを探す方法を解説します
これは2017年11月18日に行われた 第8回 カーネル／VM探検隊@関西の発表資料です。

Transcript

1. χϡʔϥϧ'.Իݯ NAOMASA MATSUBAYASHI
2. '.Իݯ प೾਺มௐ '. ΛԠ༻ͯ͠
   গͳ͍ύϥϝʔλͰΞίʔεςΟοΫͳָثͷԻ৭Λ࠶ݱ͢Δ
   ిࢠָثͷԻ৭ͷੜ੒ख๏ ιϑτ΢ΣΞͰԻָΛԋ૗͢Δ͜ͱ͕೉͔ͬͨ͠
   ੈلͷ1$΍ήʔϜػʹ
   ͜ΕΛߦ͏ͨΊͷઐ༻ͷϋʔυ΢ΣΞ͕Α͘ࡌ͍ͬͯͨ
3. ຊ෺ͷָثͷΑ͏ͳԻΛ
   ࿥ԻσʔλΛϝϞϦʹ࣋ͭ͜ͱͳ͘
   खଓ͖తʹੜ੒͍ͨ͠ গͳ͍ύϥϝʔλͰ
   ༷ʑͳָثͷԻ৭ʹରԠ͍ͨ͠ '.Իݯͷૂ͍
4. ָثͷΑ͏ͳԻͱ͸
5. $ $ $ $ ඵ ඵ ඵ ඵ ඵ ඵ ϐΞϊͷ$ͷԻͷप೾਺ղੳͷ݁Ռ $ $
6. $ $ $ $ ඵ ඵ ඵ ඵ ඵ ඵ $ $ $ͷप೾਺)[ͷ੒෼ جԻ ϐΞϊͷ$ͷԻͷप೾਺ղੳͷ݁Ռ
7. $ $ $ $ ඵ ඵ ඵ ඵ ඵ ඵ $ $ )[ͷ੔਺ഒͷ੒෼ ഒԻ ϐΞϊͷ$ͷԻͷप೾਺ղੳͷ݁Ռ
8. $ $ $ $ ඵ ඵ ඵ ඵ $ $ ؚ·Ε͍ͯΔഒԻ͕มΘΔͱ
   ผͷָثͷԻʹฉ͑͜Δ $ͷप೾਺)[ͷ੒෼ ϚϦϯόͷ$ͷԻͷप೾਺ղੳͷ݁Ռ
9. $ $ $ $ ඵ ඵ ඵ ඵ ඵ $ $ Ի৭ʹؚ·ΕΔഒԻ͸࣌ؒͰมԽ͢Δ େ఍ͷָث͸جԻ͔Β཭ΕͨഒԻ͕ઌʹফ͑Δ ϐΞϊͷ$ͷԻͷप೾਺ղੳͷ݁Ռ
10. ຊ෺ͷָثͷΑ͏ͳԻΛ૗ͰΔ
    ؔ਺ʹඞཁͳಛ௃ جԻͷप೾਺Ͱ܁Γฦ͞ΕΔपظؔ਺Ͱ͋Δ ࣌ؒͱͱ΋ʹৼ෯͕มԽ͢Δ ࣌ؒͱͱ΋ʹؚ·ΕΔഒԻ͕มԽ͢Δ ؚ·ΕΔഒԻͷप೾਺ͱେ͖͞͸
    ؔ਺ͷҾ਺ͰίϯτϩʔϧͰ͖Δ
11. प೾਺มௐ '. ৴߸೾ͷ܏͖͕
    ൖૹ೾ͷप೾਺ͷมԽʹม׵͞ΕΔ प೾਺͕ߴ͍ प೾਺͕௿͍ ਖ਼ͷ܏͖ ෛͷ܏͖
12. ৴߸೾ͱൖૹ೾ͷप೾਺͕͍ۙͱ
    ৴߸೾͸ൖૹ೾Λ࿪·ͤΔಇ͖Λ͢Δ प೾਺มௐ '.
13.       $$ $ $ $ ЋΛม͑ͳ͕Βप೾਺ղੳΛߦͳͬͨ݁Ռ Ћ ৴߸೾ͷৼ෯͕େ͖͍΄Ͳ೾ܗ͕࿪ΜͰഒԻ͕ग़Δ
14. ഒ ഒ ഒ ഒ ഒ ഒ $$ $ $ $ ТTΛม͑ͳ͕Βप೾਺ղੳΛߦͳͬͨ݁Ռ ТT ৴߸೾ͷप೾਺ʹΑͬͯഒԻ͕ग़ΔҐஔ͕มΘΔ
15. ͜ΕΛར༻ͯ͠
    ຊ෺ͷָثͷΑ͏ͳ೾ܗΛ࡞Ζ͏ͱ͍͏ͷ͕
    FMԻݯ
16. ΦϖϨʔλ प೾਺มௐ Τϯϕϩʔϓ ೖྗΛप೾਺มௐͯ͠
    ΤϯϕϩʔϓͰৼ෯Λ࣌ؒมԽͤͯ͞
    ग़ྗ͢ΔΑ͏ͳؔ਺Λ
    '.ԻݯͰ͸ΦϖϨʔλͱݺͿ
17. प೾਺มௐ Τϯϕϩʔϓ प೾਺มௐ Τϯϕϩʔϓ  ΦϖϨʔλͷग़ྗ͸
    ผͷΦϖϨʔλͷೖྗʹͰ͖Δ ΦϖϨʔλ
18. प೾਺มௐ Τϯϕϩʔϓ प೾਺มௐ Τϯϕϩʔϓ  ഒԻͷຊ਺͕ঃʑʹ
    ݮ͍ͬͯ͘ͷΛ
    ࠶ݱ͢Δ Իͷେ͖͕͞ঃʑʹ
    ݮ͍ͬͯ͘ͷΛ
    ࠶ݱ͢Δ ͭͷΤϯϕϩʔϓͰ
    ഒԻͷ࣌ؒมԽͱԻྔͷ࣌ؒมԽΛ
    දݱͰ͖Δ ΦϖϨʔλ
19. प೾਺มௐ Τϯϕϩʔϓ प೾਺มௐ Τϯϕϩʔϓ प೾਺มௐ Τϯϕϩʔϓ प೾਺มௐ Τϯϕϩʔϓ ࣮ࡍͷ'.ԻݯͰ͸͞Βʹଟ͘ͷ
    ΦϖϨʔλΛܨ͛Δࣄ͕ଟ͍
20. $$ $ $ $ '.Իݯ ഒԻ͕࣌ؒมԽ͢Δ
    ΞίʔεςΟοΫͳָثͷΑ͏ͳ
    Ի৭Λग़ͤΔ
21. OΦϖϨʔλͷ'.Իݯͷܭࢉࣜ O
    ΦϖϨʔλ਺
    &J
    Τϯϕϩʔϓ
    TJ
    جԻͷप೾਺ʹର͢Δഒ཰
    ЋJK
    ΦϖϨʔλK͕ΦϖϨʔλJΛมௐ͢Δڧ͞
    ЌJ
    ग़ྗʹؚ·ΕΔΦϖϨʔλJͷ৴߸ͷׂ߹
22. Ͳ͜Λ͍ͬͨ͡Β
    ཉָ͍͠ثͷԻ৭ʹۙ෇͘ͷ͔
    ͬ͞ͺΓΘ͔ΒΜ OΦϖϨʔλͷ'.Իݯͷܭࢉࣜ
23. Ҩ఻త'.Իݯ NAOMASA MATSUBAYASHI ೥લʹൃදͨ͠ωλ
24. αϯϓϦϯάԻΛप೾਺ղੳ ϥϯμϜͳҨ఻ࢠΛੜ੒ ֤Ҩ఻ࢠΛ'.ԻݯͰԋ૗ ֤Ҩ఻ࢠͷԻΛप೾਺ղੳ Ұ༷ަ伹ͱಥવมҟͰҨ఻ࢠΛੜ੒ αϯϓϦϯάԻͱͷڑ཭Λܭࢉ ϧʔϨοτબ୒ͰݸମΛݫબ ࠷େ෼ղೳͰ
    ্ҐʹมԽ͕ݟΒΕͳ͍
    ͸͍ ͍͍͑ ෼ղೳΛ্͛Δ͔Ͳ͏͔Λ൑அ Ҩ఻తΞϧΰϦζϜͰ
    ༩͑ΒΕͨԻͱ͍ۙप೾਺੒෼ΛؚΉԻ͕ग़Δ
    '.ԻݯͷύϥϝʔλΛ୳͢
25. $ cufind_fm_params -i ../Piano.mf.C4.aiff -o out -n 48 -m 6 -t 40 -w -8 -s 4 -c 1000 real 163m42.037s user 139m52.715s sys 23m34.418s $ ࣌ؒ෼ඵ 04 (FOUPP
    -JOVY
    Y@
    MJOVY ίϯύΠϥ OWDD ''5ʹ࢖༻ͨ͠ϥΠϒϥϦ $V''5 $16 *OUFM
    $PSF
    J 4LZMBLF
    ()[
    ίΞ (16 OWJEJB
    (F'PSDF(59
    ()[
    $6%"ίΞ ϝϞϦ IPTU
    (#
    EFWJDF
    (# αϯϓϦϯάσʔλͷ௕͞ ඵ L)[ ऴྃ৚݅ ੈ୅໨ΛٻΊͨΒऴྃ ໰୊఺
    ͕͔͔࣌ؒΔ
26. ͳΔ΄Ͳ
27. '.Իݯύϥϝʔλ ΦʔσΟΦσʔλ '.ԻݯͰԋ૗ ͜ͷؔ਺͕
    ཉ͍͠ '.Իݯͷٯؔ਺͕ཉ͍͠
    ղੳతʹٻΊΔͷ͸೉͍͕͠
    χϡʔϥϧωοτϫʔΫʹΑΔճؼͰ
    ٯؔ਺ͷۙࣅ͕ಘΒΕͳ͍͔
28. χϡʔϥϧωοτϫʔΫ
29. ܗࣜχϡʔϩϯ ೖྗ ೖྗ ೖྗ ॏΈX ॏΈX ॏΈX ೖྗΛෳ਺ड͚औͬͯ
    ॏΈΛ͔͚ͯ૯࿨ΛऔΓ
    ͦΕΛ׆ੑԽؔ਺ʹ௨ͨ݁͠ՌΛग़ྗ͢Δؔ਺ ׆ੑԽؔ਺
30. ׆ੑԽؔ਺ ೖྗ͕͋Δᮢ஋ΑΓେ͖͍࣌ʹҎ্ͷ஋Λฦ͠
    ᮢ஋ҎԼͷ৔߹ʹΛฦ͢Α͏ͳԿΒ͔ͷؔ਺
31. ೖྗ ೖྗ ೖྗ ग़ྗ ग़ྗ ग़ྗ ܗࣜχϡʔϩϯ͕͜Μͳ෩ʹ
    ܨ͕͍ͬͯΔؔ਺Λߟ͑Δ ֤઀ଓͷॏΈʹΑͬͯ
    ೖྗͱग़ྗͷؒʹ༷ʑͳؔ܎Λ࣋ͨͤΒΕΔ
32. ޡࠩٯ఻೻ Y Y Y ॏΈX ॏΈX ॏΈX ग़ྗZP
    㱠ظ଴͢Δग़ྗZ ͋ΔॏΈXͷܗࣜχϡʔϩϯʹ
    ஋YΛ௨ͯ͠ग़͖ͯͨ஋ZP͕ग़͖ͯͯཉ͍͠஋ͱҧͬͨ࣌ ͜ͷχϡʔϩϯͷৼΔ෣͍Λظ଴͢Δ΋ͷʹ͚ۙͮΔͷ͸
    YͱZΛఆ਺ͱΈͳͯ͠
    ZͱZPͷڑ཭͕࠷খͱͳΔΑ͏ͳ
    ॏΈͷ஋Λ୳Δ໰୊ʹͳΔ
33. ޯ഑߱Լ๏ ZͱZP͕ͣΕ͍ͯΔ΄Ͳ
    େ͖ͳ஋ΛͱΔؔ਺& ZP Λఆٛ͢Δ &ͷΑ͏ͳؔ਺ΛΤϥʔؔ਺ͱݺͿ X & ZP ͜͜ʹߦ͖͍ͨ ͍·͜͜
34. X & ZP ͜͜ʹߦ͖͍ͨ ͍·͜͜ ࠓͷҐஔͷ܏͖ʹԊͬͯ
    XΛߋ৽͢Ε͹ྑ͍ͷͰ͸ ޯ഑߱Լ๏ ZͱZP͕ͣΕ͍ͯΔ΄Ͳ
    େ͖ͳ஋ΛͱΔؔ਺& ZP Λఆٛ͢Δ
35. ޡࠩٯ఻೻ ZP Y X X Τϥʔؔ਺ʹର͢Δ
    XͷӨڹ Τϥʔؔ਺ʹର͢Δ
    XͷӨڹ ط஌ ͜ͷ෦෼Λग़ྗଆ͔Βॱ൪ʹٻΊ͍͚ͯ͹ྑ͍ ܗࣜχϡʔϩϯ͕Կ૚΋ॏͳ͍ͬͯͯ΋
    ॏΈͷमਖ਼͕Ͱ͖Δ
36. χϡʔϥϧωοτϫʔΫ ط஌ͷೖྗ ط஌ͷग़ྗ ? ະ஌ͷؔ਺͕͋ͬͯ
    ͦͷೖྗͱग़ྗͷ૊͕े෼ͳྔ͋Δͱ͖ʹ
    ະ஌ͷؔ਺Λۙࣅ͢Δؔ਺Λ
    ܭࢉͰٻΊΔࣄ͕Ͱ͖Δ
37. ΑΓৄ͍͠ޡࠩٯ఻೻ͷ਺ֶతͳղઆ͸
    ΦϥΠϦʔ͔Βྑ͍ຊ͕ग़͍ͯΔ θϩ͔Β࡞ΔDeep Learning
38. '.Իݯύϥϝʔλ ΦʔσΟΦσʔλ '.ԻݯͰԋ૗ ͜ͷؔ਺͕
    ཉ͍͠ '.Իݯͷύϥϝʔλͱ
    ͦͷԋ૗݁ՌͷΦʔσΟΦσʔλ͸
    େྔʹ༻ҙͰ͖Δط஌ͷೖग़ྗͰ͋Δ ΦʔσΟΦσʔλ͔Β
    '.ԻݯύϥϝʔλΛಘΔؔ਺͸
    ະ஌ͷؔ਺Ͱ͋Δ
39. ֶशσʔλͷ༻ҙ
40. ͜ͷͭͷԻɺฉ͖෼͚ΒΕ·͔͢
    ͦͷ Ի Ի
41. ͜ͷͭͷԻɺฉ͖෼͚ΒΕ·͔͢
    ͦͷ Ի Ի
42. ਓؒͷௌ֮͸
    ֤प೾਺ͷ೾ͷৼ෯͚ͩΛฉ͍͍ͯͯ
    Ґ૬Λฉ͍͍ͯͳ͍ ྘ͷഒԻͷҐ૬͕ͣΕ͍ͯΔ͕
    ͦͷࣄΛ஌֮͢Δࣄ͸ग़དྷͳ͍
43. प೾਺ ࣌ࠁ '.ԻݯΛ
    ΦΫλʔϒ໨ͷϥ)[Ͱ໐Β͠
    ''5ʹ͔͚ͯεϖΫτϧը૾ΛಘΔ
44. ΢ΣʔόʔɾϑΣώφʔͷ๏ଇ ਓؒͷޒײ͕஌֮͢Δྔ͸
    ࣮ࡍͷܹࢗʹର͓ͯ͠Αͦର਺ͷؔ܎ʹ͋Δ εϖΫτϧը૾ͷ͋Δ఺ͷ஋͕
    ͔͔ͩͬͨͩͬͨ͸
    ͔͔ͩͬͨͩͬͨൺ΂ͯ
    ૬ରతʹॏཁͰ͸ແ͍
45. ͜ͷը૾ͷεϖΫτϧʹ͸
    ҎԼͷม׵͕͔͚ΒΕ͍ͯΔ
46. ༗༻ͳԻ͕ग़Δൣғ '.Իݯύϥϝʔλͷۭؒͷ͏ͪ
    ΄ͱΜͲͷ෦෼͸Ի֊͕͸͖ͬΓ͠ͳ͍
    ϊΠζͷΑ͏ͳԻʹͳ͍ͬͯΔ
47. χϡʔϥϧωοτϫʔΫʹૹΔԻ৭ͷੜ੒ʹ͸
    ҎԼͷ੍໿Λઃ͚Δ ൪໨ͷΦϖϨʔλ͸ඞͣجԻͷप೾਺ʹ͢Δ ग़ྗஈʹܨ͕͍ͬͯΔΦϖϨʔλͷ
    ΤϯϕϩʔϓͷϐʔΫλΠϜΛἧ͑Δ ൪໨Ҏ֎ͷΦϖϨʔλͷप೾਺͸
    جԻͷ੔਺ഒ͔Β͋·Γ֎Εͳ͍Α͏ʹ͢Δ ΦϖϨʔλ΁ͷೖྗͷ૯࿨͸Λ௒͑ͳ͍Α͏ʹ͢Δ ϑΟʔυόοΫೖྗͷৼ෯͸Λ௒͑ͳ͍Α͏ʹ͢Δ Τϯϕϩʔϓ͸ඵ͔Βඵͷؒʹճ͚ͩ࠷େ஋ͱͳΓ
    ͦͷޙ͸ݮগ͢Δ ൪໨ͷΦϖϨʔλͷԻ͸ඞͣग़ྗஈʹग़Δ ༗༻ͳԻ͕ग़Δൣғ
48. ೖग़ྗͷؔ܎ΛҰҙʹ͢Δ ΦϖϨʔλ ΦϖϨʔλ ΦϖϨʔλ ग़ྗ ΦϖϨʔλ ΦϖϨʔλ ΦϖϨʔλ ग़ྗ ΦϖϨʔλͷฒͼॱ͕มΘ͚ͬͨͩͰ
    ಉ͡ܭࢉ͕ߦΘΕΔΑ͏ͳ৔߹
    ෳ਺ͷύϥϝʔλ͔Βಉ͡Ի͕ग़ͯ͘Δ ೖग़ྗͷؔ܎͕ҰҙͰͳ͍Α͏ͳσʔλ͸
    ֶश͕ऩଋ͠ͳ͍ݪҼͱͳΔ
49. ಉ͡Ի͕ग़Δύϥϝʔλ͕།ҰʹͳΔΑ͏ʹ
    ҎԼͷ੍໿ΛՃ͑Δ ग़ྗஈʹܨ͕͍ͬͯΔΦϖϨʔλ͸
    ग़ྗஈͷϛΩαʔͷԻྔͰιʔτ͢Δ ग़ྗஈʹܨ͕͍ͬͯͳ͍ΦϖϨʔλ͸
    جԻʹରͯ͠Կഒͷप೾਺Ͱಈ͍͍ͯΔ͔ͱ
    ଞͷΦϖϨʔλ΁ೖྗϛΩαʔͷԻྔͷ૯࿨Ͱιʔτ͢Δ ϦϦʔελΠϜ͸ద౰ͳ࣌ؒʹݻఆ͢Δ ग़ྗʹର͢ΔӨڹ͕ஶ͘͠খ͍͞ΦϖϨʔλ͸θϩΫϦΞ͢Δ ೖग़ྗͷؔ܎ΛҰҙʹ͢Δ
50. ೖྗ૚͔ΒεϖΫτϧը૾ΛಥͬࠐΜͰ
    ͦͷԻΛग़͢ͷʹ࢖ͬͨ
    '.Իݯύϥϝʔλͱͷڑ཭ͰॏΈΛमਖ਼ ೖྗ૚ '.ԻݯͰ໐Βͯ͠εϖΫτϧը૾Λ࡞Δ ཚ਺Ͱ'.ԻݯύϥϝʔλΛੜ੒ ӅΕ૚ ग़ྗ૚ Τϥʔؔ਺ ָثͷԻ ָثͷԻΛ
    ࠶ݱ͢Δ
    '.Իݯ
    ύϥϝʔλ ֶश ධՁ χϡʔϥϧωοτϫʔΫ
51. ৞ΈࠐΈ
    χϡʔϥϧωοτϫʔΫ
52. ৞ΈࠐΈ૚ શ݁߹ ৞ΈࠐΈ ը૾ͷΑ͏ʹྡΓ߹ͬͨσʔλʹؔ࿈ੑ͕͋Δ৔߹
    ۙ๣਺ϐΫηϧΛ·ͱΊͨϑΟϧλ͕ॏཁͳ໾ׂΛՌͨ͢ࣄ͕ଟ͍ શ݁߹ͷωοτϫʔΫ͔ΒֶशʹΑͬͯ
    ӈͷΑ͏ͳ઀ଓ͕ੜ·ΕΔͷΛ଴ͭΑΓ
    ࠷ॳ͔ΒӈͷΑ͏ͳݶΒΕͨཁૉ͔͠
    ઀ଓ͞Ε͍ͯͳ͍૚Λ࡞Ζ͏
53. ϓʔϦϯά૚           ϓʔϦϯά ݶΒΕͨྖҬͷதͰ࠷΋େ͖ͳ஋Λฦͨ͠
    χϡʔϩϯͷ஋͚ͩΛ࣍ͷ૚ʹ఻͑Δ ݶΒΕͨྖҬͷதͰ࠷΋େ͖ͳ஋Λฦͨ͠
    χϡʔϩϯͷ஋͚ͩΛ࣍ͷ૚ʹ఻͑Δ
54. ਂ૚ֶश
55. Shallow vs. Deep Sum-Product Networks [Olivier Delalleau and Yoshua Bengio, 2011] https://papers.nips.cc/paper/4350-shallow-vs-deep-sum-product-networks ಉ͡χϡʔϩϯΛ૿΍͢ͳΒ
    ૚Λ޿͘͢ΔΑΓ૚ͷ਺Λ૿΍͢ํ͕
    ΑΓෳࡶͳؔ਺ΛදݱͰ͖Δͱ͢Δ࿦จ
56. ޯ഑ফࣦ໰୊ ۃΊͯʹ͍ۙ ੵͳͷͰ͔͜͜Βઌ͕Ͳ͏ͳ͍ͬͯͯ΋
    ۃΊͯʹ͍ۙޯ഑ʹͳͬͯ͠·͏ ޡࠩٯ఻೻ͷաఔͰޯ഑͕ແ͘ͳͬͯ͠·͍
    ͦΕΑΓ΋্ͷ૚ͷֶश͕ਐ·ͳ͘ͳΔ໰୊ ଟ૚ͷχϡʔϥϧωοτϫʔΫͷ
    ֶश͸ࠔ೉ͱߟ͑ΒΕ͍ͯͨ
57. ޯ഑ফࣦ໰୊ ݹయతͳχϡʔϥϧωοτϫʔΫ͸
    ඍ෼Ͱ͖Δεςοϓؔ਺ͷ୅ΘΓͱͯ͠
    ׆ੑԽؔ਺ʹγάϞΠυؔ਺Λ࢖͍ͬͯͨ ͜͏͍͏ͱ͜ΖͰඍ෼͠Α͏ͱ͢Δͱ
    ܏͖͕ۃΊͯখ͘͞ͳΔ
58. ReLU ׆ੑԽ͍ͯ͠ΔݶΓେ͖ͳޯ഑ ׆ੑԽ͍ͯ͠Δχϡʔϩϯ͕ͭͰ΋͋Ε͹
    ׆ੑԽؔ਺͕ݪҼͰޯ഑͕ࣦΘΕΔ͜ͱ͸
    ແ͘ͳͬͨ Rectified linear units improve restricted boltzmann machines [Vinod Nair and Geoffrey E. Hinton, 2010] https://dl.acm.org/citation.cfm?id=3104425
59. GoogLeNet ෳ਺Օॴͷग़ྗΛΤϥʔؔ਺ʹ௨ͯ͠
    ௕͍ωοτϫʔΫͷ్தͰ
    ޯ഑Λิڅ͢Ε͹ྑ͍ ͔͜͜Βޡࠩٯ఻೻ ͔͜͜Β΋ ͬͪ͜΋ Going Deeper with Convolutions [Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, Andrew Rabinovich, 2015] https://research.google.com/pubs/pub43022.html
60. ϒϩοΫΛ௨͖ͬͯͨ஋ͱ
    ϒϩοΫͷೖྗΛࠞͥͯग़ྗ͢Δ ResNet Deep Residual Learning for Image Recognition [Kaiming He and Xiangyu Zhang and Shaoqing Ren and Jian Sun, 2015] https://arxiv.org/abs/1512.03385
61. ResNet Deep Residual Learning for Image Recognition [Kaiming He and Xiangyu Zhang and Shaoqing Ren and Jian Sun, 2015] https://arxiv.org/abs/1512.03385 ৞ΈࠐΈ
    Y
     ৞ΈࠐΈ
    Y
     3F-6 3F-6 Ճࢉ Yͷ৞ΈࠐΈͭΛ
    ϒϩοΫͱ͢Δ Լͷ৞ΈࠐΈͷޙ
    ׆ੑԽؔ਺ʹ௨͢લʹ
    ೖྗΛࠞͥΔ 3F-6͸׆ੑԽ͍ͯ͠ΔݶΓ
    େ͖ͳޯ഑ΛऔΔͨΊ
    ੺ͷϧʔτͷޯ഑͸্૚Ͱ΋େ͖͘ͳΔ
62. ωοτϫʔΫ͸௕͍ํ͕ෳࡶͳؔ਺ΛදͤΔ ResNetΛ࢖͑͹ޯ഑ফࣦ͸໰୊ʹͳΒͳ͍ 1୆ͷϚγϯʹ৐ΔൣғͰ΍͍ͬͯΔݶΓ ৳͹ͤΔͱ͜·Ͱ௕͍ResNetΛ࢖ͬͯ ௕͗͢Δͱ͍͏ࣄ͸ͳ͍Μ͡Όͳ͍͔ͳ '.Իݯ͸ଟ෼ෳࡶͳؔ਺ͳΜ͡Όͳ͍͔ͳ
63. ࣮૷
64. Caffe Φʔϓϯιʔεͳ σΟʔϓϥʔχϯάϑϨʔϜϫʔΫͷ1ͭ http://caffe.berkeleyvision.org/
65. CUDAΛ࢖ͬͯGPUͰ χϡʔϥϧωοτϫʔΫͷ ֶशͱධՁΛߦ͏ࣄ͕Ͱ͖Δ Caffeͱιϑτ΢ΣΞFMԻݯΛ૊Έ߹ΘͤΔ GPUΛ࢖ͬͨ৔߹ CPUଆ͸΄΅σʔλͷసૹʹઐ೦͢ΔࣄʹͳΓ Ϛϧνϓϩηοαͷ৔߹ 1ͭΛআ͍ͯ༡ΜͰ͍Δঢ়ଶʹͳΔ ࢒ΓͷϓϩηοαʹεϖΫτϧը૾ͷੜ੒Λͤͯ͞ ग़དྷͨ෺͔ΒGPUʹ౤͛Ε͹ ࣄલʹେྔͷεϖΫτϧը૾Λ༻ҙ͢Δඞཁ͕ແ͍
66. void training_data_generator<T,operator_count,resource_type>::poll() { bool active=true; constexpr size_t batch_size=15; constexpr size_t image_size=400*length; constexpr size_t config_size=config_length; const auto window=softfm::generate_window(1<<13); std::random_device seed_gen; resource_type resources(batch_size,192000,seed_gen()); generate_training_data<T,operator_count,resource_type> generator; while(1) { if(active) { std::vector<T> pixels_(image_size*batch_size,0); std::vector<T> configs_(config_size*batch_size,0); generator(window,pixels_.data(),configs_.data(),batch_size,resources); std::lock_guard<std::mutex> lock(guard); configs.emplace(std::move(configs_)); pixels.emplace(std::move(pixels_)); if(pixels.size()==100)active=false; if(end)break; } else { std::lock_guard<std::mutex>lock(guard); if(pixels.size()< 100)active=true; if(end) break; } } } FMԻݯͷεϖΫτϧը૾ͱઃఆͷϖΞΛ ແݶʹ࡞ΔεϨου
67. template <typename Dtype> void FMAudio8Layer<Dtype>::DataLayerSetUp( const vector<Blob<Dtype>*>& bottom,const vector<Blob<Dtype>*>& top){ CHECK_GT(batch_size_*size_,0)<<"oops"; labels_.resize(batch_size_,0); auto &gen=softfm::training_data_generator<Dtype,8,r>::get(); data_[pos_]=std::move(gen.get_pixels()); vector<int> label_shape(1,batch_size_); top[0]->Reshape(batch_size_,channels_,height_,width_); top[1]->Reshape(label_shape); added_data_.Reshape(batch_size_,channels_,height_,width_); added_label_.Reshape(label_shape); added_data_.cpu_data(); added_label_.cpu_data(); } template <typename Dtype> void FMAudio8Layer<Dtype>::Forward_cpu( const vector<Blob<Dtype>*>& bottom,const vector<Blob<Dtype>*>& top){ top[0]->Reshape(batch_size_,channels_,height_,width_); top[1]->Reshape(batch_size_,1,1,1); top[0]->set_cpu_data(data_[pos_].data()); top[1]->set_cpu_data(labels_.data()); pos_=(pos_+1)%n_; auto &gen=softfm::training_data_generator<Dtype,8,r>::get(); data_[pos_]=std::move(gen.get_pixels()); } ੜ੒͞ΕͨFMԻݯͷεϖΫτϧը૾Λर͏ೖྗ૚
68. layer { name: "data" type: "FMAudio8" top: "data" top: "dummy1" } layer { name: "label" type: "FMConfig8" top: "label" top: "dummy2" } layer { name: "conv_1" type: "Convolution" bottom: "data" top: "conv_1" ࣮૷ͨ͠ೖྗ૚Λ ωοτϫʔΫఆٛϑΝΠϧʹ ฒ΂Δ εϖΫτϧը૾͕
    ແݶʹ༙͍ͯ͘Δ૚ '.Իݯͷઃఆ͕
    ແݶʹ༙͍ͯ͘Δ૚
69. layer { name: "conv_1" type: "Convolution" bottom: "data" top: "conv_1" param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } convolution_param { num_output: 64 pad: 3 kernel_size: 7 stride: 2 #4%ϥΠηϯεͰ഑෍͞Ε͍ͯΔ3FT/FUͷఆ͕ٛ
    طʹ͋ΔͷͰ
    ໯͖ͬͯͯઌ΄Ͳͷೖྗ૚ʹ͚ͬͭ͘Δ https://github.com/jay-mahadeokar/pynetbuilder
70. layer { name: "loss/classifier_" type: "InnerProduct" bottom: "conv_1by1_1000" top: "loss/classifier_" param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0 } inner_product_param { num_output: 104 weight_filler { type: "xavier" } ग़ྗͷ਺͕
    '.Իݯύϥϝʔλͷ਺ʹ
    ͳΔΑ͏ʹ಺ੵΛͱͬͯ
71. ࠷ޙʹ'.Իݯͷઃఆͱͷ
    ϢʔΫϦουڑ཭ΛٻΊΔ num_output: 104 weight_filler { type: "xavier" } bias_filler { type: "constant" value: 0 } } } layer { name: "loss/loss" type: "EuclideanLoss" bottom: "loss/classifier_" bottom: "label" top: “loss/loss" loss_weight: 1 }
72. ResNet͸૚ͷ਺Λൺֱతॊೈʹ ૿ݮͤ͞Δࣄ͕Ͱ͖Δ ݩͷ࿦จʹ͸ CIFAR-10(Ұൠ෺ମೝࣝϕϯνϚʔΫ)ʹ͓͍ͯ 152૚·Ͱ͸૚Λ૿΍͢΄Ͳ ྑ͍݁Ռ͕ಘΒΕͨ͢Δ࣮ݧ݁Ռ͕ࡌ͍ͬͯΔ ͳͥResNet-36ͳͷ͔
73. GeForce GTX1070 ࠓճ࢖ͬͨGPU
74. $ nvidia-smi Thu Nov 16 00:54:35 2017 +-----------------------------------------------------------------------------+ | NVIDIA-SMI 384.90 Driver Version: 384.90 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 GeForce GTX 1070 Off | 00000000:01:00.0 Off | N/A | | 41% 69C P2 118W / 151W | 6557MiB / 8114MiB | 99% Default | +-------------------------------+----------------------+----------------------+ +-----------------------------------------------------------------------------+ | Processes: GPU Memory | | GPU PID Type Process name Usage | |=============================================================================| | 0 28991 C ./resume 6547MiB | +-----------------------------------------------------------------------------+ ϝϞϦ͕8GB͔͠ͳ͍ೈऑͳGPUͰ͸ ResNet-36ͷ࣍ʹখ͍͞ResNet-50͸ ϝϞϦʹ৐Βͳ͔ͬͨ(όοναΠζ15ͷ৔߹) ΋͏73".ͳ͍ ͳͥResNet-36ͳͷ͔
75. CaffeͷSolverͰֶश void run(){ caffe::SolverParameter solver_param; caffe::ReadProtoFromTextFileOrDie(FLAGS_solver, &solver_param); std::shared_ptr<caffe::Solver<float>> solver(caffe::SolverRegistry<float>::CreateSolver(solver_param)); if (FLAGS_snapshot.size()) { LOG(INFO) << "Resuming from " << FLAGS_snapshot; solver->Restore(FLAGS_snapshot.c_str()); } else if (FLAGS_weights.size()) { CopyLayers(solver.get(), FLAGS_weights); } const auto net = solver->net(); LOG(INFO) << "Solve start."; solver->Solve(); }
76. return; } if( FLAGS_output.empty() ) { LOG(INFO) << "Required parameter output is not set." << FLAGS_snapshot; return; } const auto net = solver->net(); float dummy = 0.f; constexpr size_t image_size = 400*400; std::vector< float > image( image_size, 0.f ); softfm::get_spectrum_image_flat( std::vector< std::pair< int, std::string > >{ { FLAGS_note, FLAGS_input } }, image.data(), FLAGS_threshold ); const auto input_layer = boost::dynamic_pointer_cast< caffe::MemoryDataLayer< float > >( net->layer_by_name("data") ); input_layer->Reset( image.data(), &dummy, 1 ); const auto result = net->Forward(); const auto data = result[ 1 ]->cpu_data(); const std::vector< float > values( data, std::next( data, result[ 1 ]->count() ) ); namespace karma = boost::spirit::karma; std::string serialized; std::ofstream out( FLAGS_output, std::ios::out|std::ios::binary ); std::ostreambuf_iterator< char > obuf( out.rdbuf() ); karma::generate( obuf, karma::float_ % ',' , values ); } ຊ෺ͷָثͷԻΛ ֶशͨ͠ωοτϫʔΫʹ௨ͯ͠ධՁ
77. ޡૢ࡞ʹΑΓ
    ͜ͷظؒͷΤϥʔϩά͸ࣦΘΕͨ 200epoc͋ͨΓ·Ͱ͸ٸܹʹޡ͕ࠩݮ͕ͬͨ ͔ͦ͜Β͸ඇৗʹΏͬ͘Γݮগ͍ͯͬͨ͠
78. 102MBͷFMԻݯ৬ਓͷΧϯ͕஀ੜͨ͠ $ ls 8op_iter_300000.caffemodel -lhaS -rw-r--r-- 1 root root 102M 11݄ 16 17:42 8op_iter_300000.caffemodel
79. ࣮ࡍͷϐΞϊͷԻ χϡʔϥϧFMԻݯ Ҩ఻తFMԻݯ ໐Γ࢝Ίʹग़͍ͯΔप೾਺͸େମ߹͍ͬͯΔ͕ ഒԻ͕શ͘ݮਰ͍ͯ͠ͳ͍ ݁Ռ ϐΞϊ
80. ࣮ࡍͷϐΞϊͷԻ χϡʔϥϧFMԻݯ Ҩ఻తFMԻݯ Ҩ఻తFMԻݯʹΑΔϐΞϊ গͳ͘ͱ΋͜ͷ͘Β͍͸ग़ͤΔ͜ͱ͸Θ͔͍ͬͯΔ ϐΞϊ ݁Ռ
81. ࣮ࡍͷϐΞϊͷԻ χϡʔϥϧFMԻݯ Ҩ఻తFMԻݯ ϐΞϊͬΆ͘ฉ͑͜Δ෦෼΋͋Δ͕ Ҩ఻తFMԻݯͱൺ΂Δͱ͠ΐͬͺ͍ ϐΞϊ ݁Ռ
82. ࣮ࡍͷϏΦϥͷԻ χϡʔϥϧFMԻݯ ϏΦϥ ݁Ռ
83. ࣮ࡍͷαοΫεͷԻ χϡʔϥϧFMԻݯ ΞϧταοΫε ݁Ռ
84. ֶशࡁΈͷϞσϧ͕͋Δঢ়ଶ͔ΒͳΒ Ҩ఻తΞϧΰϦζϜͱൺֱͯ͠ѹ౗తʹ୹࣌ؒ ܭࢉʹཁͨ࣌ؒ͠ 04 (FOUPP
    -JOVY
    Y@
    B MJOVY (FOUPP
    -JOVY
    Y@
    B MJOVY ίϯύΠϥ OWDD HDD ''5ʹ࢖༻ͨ͠ $V''5 ''58 ୯ਫ਼౓ɺ0QFO.1ରԠɺ"79ରԠ %//ϑϨʔϜϫʔΫ  $B⒎F $16 *OUFM
    $PSF
    J 4LZMBLF
    ()[
    ίΞ (16 OWJEJB
    (F'PSDF(59
    ()[
    $6%"ίΞ ϝϞϦ IPTU
    (#
    EFWJDF
    (# σʔλͷ௕͞ ඵ L)[ ऴྃ৚݅ ੈ୅໨ΛٻΊͨΒऴྃ ॱํ޲ͷܭࢉΛճߦͬͯऴྃ ܭࢉʹཁͨ࣌ؒ͠ ࣌ؒ෼ඵ ඵ ݁Ռ
85. χϡʔϥϧωοτϫʔΫͰૉૣ͘ٻΊͨ େࡶ೺ͳղΛॳظ஋ͱͯ͠ Ҩ఻తΞϧΰϦζϜͰਖ਼֬ͳղʹ͚ۙͮΕ͹ ୹͔࣌ؒͭߴ඼࣭ͳύϥϝʔλ͕ಘΒΕΔ͔΋ ߟ࡯
86. ৞ΈࠐΈχϡʔϥϧωοτϫʔΫΛ࢖ͬͯ ༩͑ΒΕͨԻ৭ʹ͍ۙԻ͕ग़Δ FMԻݯͷύϥϝʔλΛٻΊͨ Ҩ఻తΞϧΰϦζϜΛ࢖ͬͨ৔߹ͱൺֱͯ͠ ඼࣭Ͱ͸ྼΔ͕ܭࢉ࣌ؒͰ͸༗ར ྆ऀΛ૊Έ߹ΘͤΔͱྑ͍͔΋͠Εͳ͍ ·ͱΊ े෼ͳ࣮ݧΛߦΘͣʹબΜͩ ϋΠύʔύϥϝʔλ͕ଟ͍ͨΊ ΋ͬͱ඼࣭Λ্͛Δํ๏͕ӅΕ͍ͯΔ͔΋