The rise of creative machines - GAN Technology for Fake Data as Resource
Workshop on opportunities and perils of fake data produced by the AI concept Generative Adversarial Networks (GAN) and other generative machine learning techniques.
In the last 5 years the field of so-called creative or generative Artificial Intelligence, notably Generative Adversarial Networks (GANs), has progressed in an enormous pace. What does it mean for our society that AI is gaining an increasing capacity to (re-)produce informational patterns? How does the world look like when it is populated with deep fakes and synthetic realities?
Generative AI can be used to generate images of people who do not exist, create convincing fake footage in which people do things they never did, dream up moving images based on a single picture, produce artificial influencers on Instagram, make art, take over basic creative labor (production of video game graphics, interior design, fashion, texts, music or recipes) or produce synthetic data -thus circumventing some of the problems with real data: scarce availability, labor intensity of data labelling, data biases, privacy intrusiveness!- to fuel our hungry data economy, speed up machine learning, or break or pervert existing data analytical systems.
By bringing together a variety of use cases for generative AI, this seminar offers a unique opportunity to think about some recurrent societal and legal challenges and opportunities that generative AI brings along.
When: 27 August 12.30 to 16.00
Where: Eden hörsal, Eden, floor 1, Paradisgatan 5H, Lund, Sweden [map]
Registration: The registration is now closed!
12.30 Mingle, registration and lunch
Legal and societal challenges of generative AI. How useful is it to speak of imaginative AI? [VIDEO]
In this talk Katja de Vries, postdoctoral researcher at the Department of Sociology of Law (Lund University), will introduce some of the societal and legal challenges that follow from the use of generative AI. She will show the need to find some common concepts and metaphors to discuss this type of technology with a broader audience. Generative AI is sometimes said to ‘dream up’ or ‘hallucinate’ its creative outputs. How useful is it to talk ofimaginative AI? And how does such a metaphor steer how we look at generative AI from the perspective of democracy (the need for the public debate to be informed by ‘objective facts’), intellectual property (can machines be creative?) and can there be science based on ‘fabricated’ data?
From generative models to Variational Autoencoders and Generative Adversarial Networks [VIDEO]
Presented by: Najmeh Abiri, Computational Biology and Biological Physics, Lund University and Henning Petzka, Postdoc at Mathematics. Lund University
Abstract: One of the most powerful ways of analyzing and understanding data is the use of generative models. These models learn to represent an estimation of the data distribution, and they have the advantages to:
- represent uncertainty in data,
- not require labeled data (unsupervised learning),
- generalize to new data.
Combining generative models with deep neural networks allows modeling high-dimensional and complex data. In this talk, we explore two of the most popular deep generative models, Variational Autoencoders (VAE) and Generative Adversarial Networks (GAN).
Fashion model modelling [VIDEO]
Presented by Håkan Jonsson, Zalando Operations Data Products
Abstract: A brief talk covering uses cases in the fashion retail domain, from design and content creation to time series modelling
GAN, regulation and legal challenges [VIDEO]
Presented by Katja de Vries
15.00 Fika and mingle
15:30 - 16.00 Panel discussion with the speakers modeated by Stefan Larsson [VIDEO]