AIProject 2015 å T Z o M { å S x5 D t ± ï Ñ å ï ³ µ ¯ p æ hGoogle I/O w ... M ^ ¹ Ñ Ä...

6
SFC ڀݚॴελʔτΞοϓॿϓϩδΣΫτใ(2019 ) AI Λ༻৽දݱग़ͷΊͷϑϨʔϜϫʔΫͷఆతͳධՁͱζ Ϩͷಋೖͷ ڀݚಙҪੜ ܚጯक़େେӃ ɾϝσΟΞڀݚՊ।ڭ(ظ) [email protected] 2020 2 10 2019 4 ͷਃॻͷ༰ʹجɺػցशΛ༻ (ਓ·ͶͰͳ) ৽දݱΛੜ ΔϑϨʔϜϫʔΫʹɺज़తͳجڀݚͱදݱͷફͷ໘ΒڀݚΛߦͳɻ 1 ਓͱ AI ͷʹΑΔදݱҬͷେ චɺਓͱ AI ͷ DJ ʹޓҰۂۂΔͱʹΑύϑΥʔϚϯεΛߦ AI DJ Project Λ 2015 ΒଓΔɻຊ 5 ʹαϯϑϥϯγείͰߦΘΕ Google I/O ͷ Keynote Preshow ʹটΕɺ Google CEO ͷΩʔϊʔτεϐʔνͷલͷҰɺ 15,000 ͷௌͷલͰύϑΥʔϚϯεΛ൸࿐ (1)ɻ 1. Google I/O 2019 Keynote Preshow ʹΔ AI DJ Project 1

Transcript of AIProject 2015 å T Z o M { å S x5 D t ± ï Ñ å ï ³ µ ¯ p æ hGoogle I/O w ... M ^ ¹ Ñ Ä...

  • SFC (2019 )

    AI —

    ( )

    [email protected]

    2020 2 10

    2019 4 ( )

    1 AI

    AI DJ AI DJ

    Project 2015 5 Google I/O

    Keynote Preshow Google CEO 15,000

    ( 1)

    1. Google I/O 2019 Keynote Preshow AI DJ Project

    1

  • 2. Algo-Rhythm

    ”Algo-Rythm” *1

    MIDI

    AI AI

    ( 2) ORF2019

    Asia Digital Art Award

    2

    Google Magenta

    AI

    (Ableton Live)

    AI (Ableton Live Max for Live )

    *2 M4L.RhythmVAE/M4L.MelodyVAE

    Variational Autoencoder(VAE)[4]

    *1 https://cclab.sfc.keio.ac.jp/projects/algo-rhythm/*2 https://cclab.sfc.keio.ac.jp/projects/m4l-rhythmvae/

    2

  • ab

    c d

    e f

    3. M4L.RhythmVAE - AI

    4. SFC MUTEK AI Music Lab

    VAE [2]

    MIDI &

    11 MUTEK

    SFC X-Music Lab

    ”Towards democratizing music production with AI Design of

    Variational Autoencoder-based Rhythm Generator as a DAW plugin”

    New Interface for Musical Expressions 2020 ( )

    3

    2.

    2

    3

  • 2

    Newural Beatbox Web

    (human beatbox) Web *3( 5)

    Barbican Center AI: More than Human *4( 6) 2020

    5. Neural Beatbox 6. Barbican Centre AI:More than

    Human

    4

    (Generative Adversarial Networks/GANs)[3]

    AI

    GANs

    ( ) (

    )

    GANs

    GANs

    — = —

    GANs

    *3 http://www.neuralbeatbox.net/*4 https://www.barbican.org.uk/whats-on/2019/event/ai-more-than-human

    4

  • Generated Data

    Training Data

    Random Noise

    GAN Discriminator

    Generator

    Real

    Fake

    Real or Fake?

    G

    D1

    Input

    Distance Discriminator

    D3

    Genre/Style Discriminator

    Jazz

    Hip Hop

    Rock. . .

    Accuracy

    Distance

    Ambiguity

    Which Genre/Style?

    How Different?

    D2

    Training

    7. Creative Adversarial Network

    GANs

    (Hiphop, Jazz, Rock ) ( 7 D2

    ” ”

    Elgammal Creative Adversarial Network

    [1]

    Web

    *5

    2020 3 International Computational Creativity

    Conference

    5

    AI

    GANs

    AI

    AI

    *5 https://soundcloud.com/naotokui/sets/rhythms-generated-by-creative-adversarial-network

    5

  • ( )

    ( AI DJ Project

    ) AI

    [1] A. Elgammal, B. Liu, M. Elhoseiny, and M. Mazzone. CAN: Creative Adversarial Networks,

    Generating ”Art” by Learning About Styles and Deviating from Style Norms. jun 2017.

    [2] J. Gillick, A. Roberts, J. Engel, D. Eck, and D. Bamman. Learning to Groove with Inverse

    Sequence Transformations. Proceedings of the 36th International Conference on Machine

    Learning, 2019.

    [3] I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair,

    A. Courville, and Y. Bengio. Generative adversarial nets. In Advances in Neural In-

    formation Processing Systems, volume 3, pages 2672–2680, jun 2014.

    [4] D. P. Kingma and M. Welling. Auto-encoding variational bayes. In 2nd International

    Conference on Learning Representations, 2014.

    6