User:Alexander Roidl/thesis-outline3

From XPUB & Lens-Based wiki

Thesis Outline

Thesis Statement

Topic

The use of machine learning algorithm for visual production

Focus

What can be learned from the use of machine learning in the fields of visual production (art & design)?

Argument

The current application of machine learning in artistic fields comes down to the output of some pre-made open source scripts without a deeper investigation on the inner workings of these tools. Why is it important to have a deeper understanding of these algorithms in order to create meaningful outputs with them? What can be learned from machine learning algorithms and what is its relation to visual culture?

Scope

This refers mostly to the current use of machine learning and how this hype is used by artists to generate more or less random art using prewritten algorithms. And also how it is referred to as the magical tool for everything, providing surprising outputs with less effort. This thought can be discussed controversially. But still I think that there is something that can be learned from those kind algorithms that is beyond »interesting« images. And that is what I want to find out with my work. I have the feeling that there is space for intervention in the very fundamental layers of neural networks and how they operate. I want to further investigate in the inner workings of them and also find out why it is important to understand them in order to work with them in a more meaningful way.

I want to write an analytical essay that will relate to my practical work and build its theoretical foundation.

Introduction

Background

With my background as a graphic designer I am interested visual culture and in algorithmic generation of visual output. Recently new kinds of algorithms found their way into digital systems: Machine Learning. They are increasingly implemented in our daily life and perform better than other kinds of calculations. At the same time machine learning raises other questions regarding e.g. opacity, bias and moral. Following this huge current attention machine learning is also being used in the arts. While recent artistic approaches are limited in their engagement with the technological understanding, I want to investigate on the use of machine learning in a more sustainable way, asking what can be left after the hype of machine learning. What makes those algorithms different from other existing ones? How is their output relevant for visual culture?

Body

PART 1: Machine Learning: Interventions in visual culture

Current State of machine learning

  • Description of machine learning in our daily life in different applications
    • What machine learning is, where it is and what it does
    • The controversial discussion on machine learning (a short overview)
  • Current situation of artists/designers using machine learning algorithms in the realm of visual production
    • Influence on visual culture: While computer vision has been used a lot by artists, artificial image generation is still less explored
    • it is mostly used to generate images from existing algorithms, there is no deeper engagement with the technique / tool itself
    • Is machine learning just the next filter that will be outdated in 2 years?

New Image Language

Images as an example for visual production

  • How we talk about images has changed. Technicians deconstruct images to their features and have invented a strange language for images.
    • Visual Culture has not yet intervened in that field
  • At the same time: Images gain new qualities: They are not only being valuable as datasets, but with algorithms we are able to add another layer of metadata to them. A layer of language, that describes the image. Images of our time are infected with texts; they visualize texts. Our image makers’ imaginations are infected with conceptual thinking, with trying to hold processes still. (Flusser, 2011, p. 13)


PART 2: Understanding Machine Learning

(maybe this needs to become smaller)

CNN / Computer Vision

Insight in one specific kind of algorithm

  • Algorithms that see and interpret our world
  • How algorithms intervene in our world
  • Visibility for Computers and Humans
    • How we help computers to help them see
    • Therefore: how computers help us see
    • Computer challenge what and how we see
    • Machines see things that remain invisible for humans
    • Relation to the invisibility of models in machine learning algorithms

GAN / Generative Algorithms

  • Algorithms that generate visual material -> images
  • Effects / surprising outputs, but why is it different from other techniques
    • Learning as engagement with material

Images and imagination of machine learning

  • The blurry understanding of machine learning creates imaginations and anthropomorphizing of machine learning
    • This insights creates a certain fascination for machine learning images


(maybe this needs to be deleted)

Reverse Engineering neural networks

  • Experimental approach in understanding
  • Black Box and how to approach this complexity

PART 3: Intervention

Ways to find a use of machine learning that is beyond effects

comparison between machine learning and other algorithms

  • The effect of machine learning as an universal tool
    • The pro and cons of this «all-eating-machine»
  • Complexity and surprise
    • Complicated statistical algorithms generate a moment of surprise
    • How any new technique has this momentum

Drawing parallel between Software Arts and the current state of machine learning

  • Software Arts can be seen as a reaction to the limited interaction with the tools in Generative Arts
  • A parallel case can be seen with machine learning: Artists are not yet interacting with the deeper source.
    • What would be an intervention in the spirit of Software Arts?

Conclusion

Summary of what can be learned from machine learning and how this can be applied. Hope to create a new discussion how we talk about machine learning and share insights.


References

Flusser, V. (2011). Into the Universe of Technical Images. (N. A. Roth, Trans.) (1 edition). Minneapolis: Univ Of Minnesota Press.

more references and readings