
4 minute read
HOW CERTAIN IS UNCERTAINTY?
ONDERZOEK IN DE KIJKER
We know how much certainty the Covid-19 vaccines offer in terms of protection against the virus. But who can confirm whether this percentage will still be valid next month? At that point, today’s fixed uncertainty suddenly becomes imprecise or undetermined. Researchers from the campuses Bruges and De Nayer are joining forces to find out how artificial intelligence can deal with this kind of uncertainty. Keivan Shariatmadar, senior researcher and mathematician at Bruges Campus, tells the story.
e to the π is known as the first part of the most remarkable and beautiful identity in mathematics, named after the Swiss scholar Leonhard Euler, who invented it in 1748. However, E-pi also stands for Epistemic AI. That is the name of the European research project of Prof. Hans Hallez and Keivan Shariatmadar from the M-Group at Bruges Campus and Prof. David Moens from the LMSD group at De Nayer Campus. The other project partners are Oxford Brookes University and TU Delft.

Keivan Shariatmadar
©Tom Talloen
FET Open European project
The Epistemic AI project is part of the prestigious European FET program, where FET stands for Future and Emerging Technologies. “FET’s mission is not only visionary but also very concrete”, explains Keivan. “How to turn Europe’s excellent science base into competitive advantage? FET projects are expected to initiate radically new lines in technology through unexplored collaborations between advanced multidisciplinary science and cutting-edge engineers. It all helps Europe grasp leadership in those promising future technology areas able to renew the basis for future European competitiveness and growth”.
The FET program has three complementary lines of action to address different methodologies and scales, from new ideas to long-term challenges. “Our project belongs to the FET Open category”, Keivan continues. “FET Open funds projects on new ideas for radically new future technologies at an early stage. FET Open calls for collaborative research and innovation actions (RIA), satisfy the FET Open ‘gatekeepers’, which are: radical vision, breakthrough technological target and ambitious interdisciplinary research. This may involve a wide range of new technological possibilities, inspired by cutting-edge science, unconventional collaboration or new research and innovation practices”.
Next-generation AI
“Reality is all about uncertainties”, says Keivan. “What if sensors are drifting away from calibration? What if a production process is very prone to the uncertainty of manual intervention? What if a trained model is not accurate enough and hard to learn from the uncertain data? Our project will investigate how the novel indeterministic uncertainty models can cope with these uncertainties”. “The main goal of E-pi is to create new methodologies and paradigms for a next-generation artificial intelligence, providing specific guaranties on its predictions through proper modelling of real-world uncertainties. Although artificial intelligence has improved remarkably over the last years, its inability to deal with fundamental uncertainty severely limits its applications. Our project will re-imagine AI with proper treatment of the uncertainty stemming from our forcibly partial knowledge of the world”.
“As currently practiced, AI cannot confidently make predictions robust enough to stand the test of data generated different from those studied at training time. While recognizing this issue under different names - e.g., ‘overfitting’ - traditional machine learning seems unable to address it in non-incremental ways. As a result, AI suffers from brittle behaviour and finds it difficult to operate in new situations, e.g., adapting to driving in heavy rain or to other road users’ different style of driving e.g., deriving from cultural traits. Our objective is to create a new paradigm providing worst-case guaranties on its predictions, thanks to a proper modelling of real-world uncertainties”.
Ultimate Factory lab
The Epistemic AI project has a duration of four years and will focus on the lab ‘The Ultimate Factory” to implement the results and techniques. To this end, the project has hired one post-doc and two PhD students within the departments of Computer Sciences and Mechanical Engineering at KU Leuven”.
Yves Persoons