CISAT Working Group on Adversarial Artificial Intelligence and Adversarial Machine Learning
Find Out MoreDenmark is widely seen as one of the most digitized countries in the world, managing huge amounts of data in digital form. Ongoing efforts to improve welfare among all Danish residents are increasingly based on artificial intelligence and machine learning techniques with the goal to provide algorithmic solutions for future decision-making challenges of the State. In sharp contrast, artificial intelligence systems are known to be sensitive to small errors in the input data, making them vulnerable to attack, intentional misconfiguration, and misuse. This can have detrimental effects on society, threatening the ambition to maintain equality and trust within Danish society.
Example
Deep neural network models used for image classification, are most often trained to solely maximize accuracy. As a result, they tend to use any available signal or feature that is even slightly correlated with the phenomenon being observed, even those that look incomprehensible to humans. This makes them susceptible to misclassifying an input image that has a small added perturbation.
Sponsors and MembersProfessor, Head of ITU CISAT
Senior Business Unit Manager
ITU CISAT is part of the IT University of Copenhagen (ITU), located at Rued Langgaards Vej 7, DK-2300
