본문 바로가기
  • Home

Study of Making Media Art using Tensorflow - Focused on applying Magenta API -

  • Journal of Communication Design
  • Abbr : JCD
  • 2019, 69(), pp.177-184
  • DOI : 10.25111/jcd.2019.69.14
  • Publisher : CDAK Society of Communication Design
  • Research Area : Arts and Kinesiology > Design > Visual Information Design > Information Design
  • Received : September 10, 2019
  • Accepted : October 28, 2019
  • Published : October 31, 2019

No Seung Kwan 1

1한양대학교

Accredited

ABSTRACT

Machine learning is the core technology of 4th industrial revolution and brought big changes in art and design creations. Media art is a form of art practice which converge cutting edge art and design and possibilities of creating art with machine learning is active in this field. This paper investigate the pratical possibilities of machine learning in art and design through analysing the creative process of interactive web art work FVTM:From Vera to Magenta(2019). The paper surveyed the development of artificial intelligence and machine learning technology and compared the characteristics of four major machine learning frameworks. Main study illustrates the characteristic of Google’s Tensorflow, the chosen framework for project and analyse two components of Magenta API which was developed based on Tensorflow especially for artists and designers; it’s sound models and image models. In creation stage, the paper illustrates evolution of open source remixing process for FVTM and explain the two main elements of the work; generation of sound applying pre-trained VAE model of Magenta API and creation and mapping process of visual elements which utilize the process of color value overlapping. Significance of FVTM work and possible of future work using new model build, train and mapping. By presenting the process of creating generative art work using machine learning API, the paper aim to stimulate more active collaboration between machine learning and art of various disciplines.

Citation status

* References for papers published after 2023 are currently being built.