Machine learning pervades many aspects of contemporary life. In response to this situation, media artists have also started to apply and reflect on machine learning algorithms in their own work. This course introduces basic concepts of machine learning for artists in a hands-on practical way. The focus is not on a rigorous presentation of technical material but on the use of techniques for creative purposes.

The course will have two parts. First, we will introduce the fundamental concepts of machine learning and other related ideas (supervised vs. unsupervised learning, regression vs. classification, etc.) and apply classical algorithms in such areas as clustering, classification, dimensionality reduction, and manifold learning. Instead of jumping directly to advanced topics like deep neural networks, we therefore begin with classical algorithms and fundamental notions to build a strong foundation. Students will apply those techniques to the production of creative projects. The second part will then move on to neural networks and deep learning. Students will not only use pretrained models but also design simple networks to perform such tasks as image classification, object recognition, semantic segmentation, depth estimation, etc.

Assessment will be studio-based. Students will present their work and participating in critique sessions. Material will be presented in the form of  hands-on coding workshops supplemented by lectures on historical and social aspects. The course will mainly concentrate on practical techniques that artists can use. The focus of learning tasks will be on image processing rather than natural language or sound, but students can develop projects in those areas on the basis of the concepts learnt in class. Students are expected to write their own code and to reflect on the techniques that they use from technical, aesthetic, cultural and social standpoints. They will do this by presenting their projects in class and critiquing classmates projects. Students will relate their projects and the techniques that they use to social and cultural aspects. These aspects can include, for instance: the history of neural networks and machine learning in relation to eugenics, cybernetics, or warfare; the social impact of machine learning on gender, work, poverty, or race; the philosophical aspects of machine learning, such as the nature of induction; political and social aspects of ImageNet and other popular datasets; the tendency of technology to become a black box and the problem of interpretable or explainable AI; questions of resource-use (for instance energy consumption, carbon emission, or impact on climate); data collection, digital labour, and surveillance capitalism; etc. 

Workshops will be conducted using contemporary languages and frameworks, such as for instance Python, scikit-learn, scikit-image, Pytorch, or TensorFlow/Keras. This list is only indicative. The specific languages to be used will depend on the instructor.