PredMod Newsflash
A PROJECT FOCUSED ON PREDICTIVE MODELLING AND UNCERTAINTY QUANTIFICATION
A part of the research environment on applied AI (AFAIR) at the School of Engineering, Jönköping University.
September 2023
More about the project
Project Objectives
The purpose is to study situations in which it is important to be confident in the decision made and to evaluate and further develop the "Prediction with Confidence" framework. A goal is to combine research and knowledge transfer between the collaborators.
Publications
Implementation of Conformal Predictive Systems in KNIME together with Redfield is now public. Tuwe presented a paper on the updates to the Conformal Prediction toolbox in KNIME at the COPA conference at Cyprus mid-September. Ulf also presented a paper on Confidence Classifiers with Guaranteed Accuracy or Precision at COPA. Finally, Nasir presented a paper on applying conformal prediction to two cases within Husqvarna, also at COPA.
We currently have two journal papers under review which introduce a new explanation method for predictive models called calibrated-explanations, utilizing Venn-Abers and Conformal Predictive Systems to provide well calibrated feature importance estimates with uncertainty quantification. It also enables creation of counterfactual explanations, illustrated by the figure.
The new method is available as a Python package and pre-prints of the introductory paper for classification and a paper introducing an extension for regression are also available. Note that the preprint papers are not yet accepted by any journals.
Flashback
We had our workshop for the second iteration at Husqvarna in late May, with a lot of interesting presentations, good food and a look at the 334 year history of Husqvarna. All companies reported their work on the cases they had been focusing on. The researchers introduced Conformal Classification, which is the focus of the third iteration, and some lessons learned so far.
Up next
Up next is the second gate meeting on the 15th of November, where most companies will present their ongoing work during iteration 3. Planning for the final year in OnTime (2024) is underway. More information on this topic will follow.
Researchers
Contact
- Senior Lecturer Computer Science
- School of Engineering
- tuwe.lofstrom@ju.se
- +46 36-10 1108
PredMod is one of several projects within the frame for the research profile AFAIR in the research and educational environment SPARK at Jönköping University’s School of Engineering.
Read more at ju.se/spark.