MeshUp #20 – Aesthetic Images and AI: What Does the Machine See?
Mohamed-Bachir Belaid will present Aesthetic Images and AI: What Does the Machine See?
Abstract
For nearly two decades, the machine learning community has tried to teach machines to judge beauty, moving from hand engineered rules to a foundational crowd rated photography dataset, then through waves of deep learning models, general purpose vision-language systems and, most recently, a live and unresolved debate over whether a universal aesthetic standard is achievable, or even desirable, at all. That journey is the main subject of this talk. It is examined here through NILU, an environmental research institute and a partner in a centre for AI and creativity. HIBE, a proposed NILU project, offers a concrete case for where this history could go next: rather than chasing a universal aesthetic model, it would score satellite imagery across three axes, beauty, interestingness, and harmfulness, against a single named expert rather than a crowd, testing whether a machine can be tailored to one person’s judgment and made to explain it rather than just score it.
Bio
Dr. Mohamed-Bachir Belaid is a Senior Data Scientist at NILU, the Climate and Environmental Research Institute, and NILU’s representative in the MishMash Centre for AI and Creativity, Work Package 7. He studied Computer Science and Optimisation at the University of Oran, Algeria, before completing a PhD in Computer Science and Artificial Intelligence at the University of Montpellier on constraint based data mining. From 2020 to 2022 he was a postdoctoral researcher at Simula Research Laboratory, working on constraint acquisition and symbolic AI, followed by a postdoctoral position at OsloMet University researching Tsetlin Machine methods for interpretable pattern recognition. Since 2022 he has worked at NILU, building machine learning systems for EU scale environmental programmes, including probabilistic pollen forecasting, deep learning for Earth observation, and explainable AI for the ESA funded AETHER project. His work has been published at AAAI, IJCAI, and in the Journal of Artificial Intelligence Research.
Access
MishMash MeetUps are short, informal meetings in the consortium where both early career and established researchers present ongoing projects. The events are open for everyone, but, for security reasons, Zoom links are only provided to people that are affiliated with a MishMash Work Package. If not, please ask for access.
Subscribe to all events (webcal)
Download calendar (ICS)
If the webcal link doesn't open in your calendar (e.g. Outlook), use "Download calendar (ICS)" and open the file in your calendar app.