Enabling Artistic Expression

How mith we use AI techniques like style transfer to empower people with disabilities to create art?

Artistic style transfer is a recent technique developed within deep learning for recomposing images in the style of other images. These algorithms allow for the automatic generation of images in the style of specific artists. In addition, models such as Generative Adversarial Networks (GANS) are being developed which enable seemingly technically accomplished compositions to be generated from minimal inputs (such as the generation of coloured illustrations from simple line sketches). The power of these models can be used to enable people with limited technical ability the agency to create sophisticated compositions and creative artefacts.

One group that may benefit from using such techniques to create art is people with mental and physical disabilities. However, how can we design interfaces and/ or contexts of use that will give these users a sense of creative agency and support their individual expression?

The goal of this project is to:

  1. Use techniques from deep learning to develop an implementation of style transfer.
  2. Explore a number of design choices for interfaces which will enable persons with significant accessibility needs to express individual creative styles and generate artistic works.

In order to achieve this, you will:

  • Research cutting-edge algorithms and models, such as Convolutional Neural Networks and GANs for the production of creative artefacts, using methods such as artistic style transfer.
  • Develop methods for training networks which will enable the iterative and cumulative combination of styles to individual preferences.
  • Develop simple user interfaces which will allow people with significant accessibility needs to train such networks and use them in individual creative expression.

Required Skills

  • Competent in programming, specifically python and (ideally) C++.
  • Be able to work with High Performance Computing (HPC), specifically systems such as Sun Grid Engine, for running processes on GPU clusters (YARCC).
  • Be familiar with the creation and training of deep learning models, such as CNNs and GANs, using libraries such as Tensorflow or Torch.
  • Ability to rapidly prototype user interfaces for web or mobile.

How to Apply

For more details on the summer school application process (including eligibility and funding) please see the overview page: here

Supervisors

Sebastian Deterding

Jon Hook

Davy Smith