Today from 3-6pm Buzzhouse will be used for the final poster presentations of the projects developed during the Digital Methods annual Summer School on 'Vision methodologies'. The summer school is intended for advanced Master's students, PhD candidates and motivated scholars who would like to work on (and complete) a digital methods project in an intensive workshop setting. For a preview of what the event is like, you can view short video clips from previous editions of the School.
Vision methodologies: Research affordances and critique of new visual analysis onlineKate Crawford and Trevor Paglen have produced a withering critique of the data behind computer vision software and other AI applications. In Excavating AI, the project they describe as dataset archaeology, they question the enterprise of image labelling, particularly in the category of ‘persons’, and discuss how sets of labels encompass problematic worldviews. Among the training data sets they scrutinise is ImageNet, the very large set of tagged images with a variety of shocking (and sensible) labels. In 2019 ImageNet removed images of people, together with their labels, leading to questions of why the images were sourced as they were and labelled as they had been. Despite the removal (and the questions surrounding it) these images and their labels already have downstream effects, they argue, having served to train vision software, among other uses. The critique has opened a debate both within the research communities behind the image sets as well as outside them, asking how to ‘de-bias’ both the training data and the applications that use them as well as whether to label at all (or how to do so fairly).Similar scrutiny has not as yet been made of other aspects of computer vision. In the dual effort to both critique and repurpose, the Summer School inquires into the affordances of computer vision for media research, especially the study of image circulation via reverse image search as well as the contextual tagging of images, otherwise known as web entities. We also discuss a series of methods to augment automated image analysis through such data enrichment strategies as emoji and hashtag linkage.
During the use of BuzzHouse by the Summer School both the Round Room and Open Space will be closed for the general public.
