Join the AI group for our regular lunch webinars.
Free to attend.
The talk
Machine learning models can make incorrect predictions with high confidence. How can we identify potentially unreliable outputs before they lead to poor decisions? In this talk, Joel Bjervig will introduce SafetyCage (https://github.com/SINTEF/safetycage), an open-source Python package developed at SINTEF for detecting potentially misclassified predictions in machine learning classification tasks. SafetyCage brings together several statistical detection methods that assess prediction reliability and flag outputs that may be incorrect. Joel will present the main ideas behind SafetyCage and show how misclassification detection can support safer and more reliable use of machine learning models.
Speaker
Joel Bjervig
Researcher at the Artificial Intelligence group, SINTEF Digital (LinkedIn)
Joel Bjervig is a researcher at the Artificial Intelligence group at SINTEF Digital. He holds a bachelor’s degree in engineering physics and a master’s degree in computational science. His master’s thesis focused on transfer learning in image classification. His work includes research on building trust in AI systems. He is one of the developers of SafetyCage.