eprintid: 10132169 rev_number: 14 eprint_status: archive userid: 608 dir: disk0/10/13/21/69 datestamp: 2021-08-04 16:33:52 lastmod: 2021-10-05 00:37:39 status_changed: 2021-08-04 16:33:52 type: proceedings_section metadata_visibility: show creators_name: Sensoy, M creators_name: Saleki, M creators_name: Julier, S creators_name: Aydogan, R creators_name: Reid, J title: Misclassification Risk and Uncertainty Quantification in Deep Classifiers ispublished: pub divisions: UCL divisions: B04 divisions: C05 divisions: F48 keywords: Training, Deep learning, Computer vision, Uncertainty, Conferences, Decision making, Predictive models note: This version is the author accepted manuscript. For information on re-use, please refer to the publisher's terms and conditions. abstract: In this paper, we propose risk-calibrated evidential deep classifiers to reduce the costs associated with classification errors. We use two main approaches. The first is to develop methods to quantify the uncertainty of a classifier’s predictions and reduce the likelihood of acting on erroneous predictions. The second is a novel way to train the classifier such that erroneous classifications are biased towards less risky categories. We combine these two approaches in a principled way. While doing this, we extend evidential deep learning with pignistic probabilities, which are used to quantify uncertainty of classification predictions and model rational decision making under uncertainty.We evaluate the performance of our approach on several image classification tasks. We demonstrate that our approach allows to (i) incorporate misclassification cost while training deep classifiers, (ii) accurately quantify the uncertainty of classification predictions, and (iii) simultaneously learn how to make classification decisions to minimize expected cost of classification errors. date: 2021 date_type: published publisher: Institute of Electrical and Electronics Engineers (IEEE) official_url: https://doi.org/10.1109/WACV48630.2021 oa_status: green full_text_type: other language: eng primo: open primo_central: open_green verified: verified_manual elements_id: 1873423 doi: 10.1109/WACV48630.2021.00253 isbn_13: 978-1-6654-0477-8 lyricists_name: Julier, Simon lyricists_id: SJULI23 actors_name: Julier, Simon actors_id: SJULI23 actors_role: owner full_text_status: public publication: WACV pagerange: 2483-2491 event_title: The Winter Conference on Applications of Computer Vision (WACV) 2021 book_title: Proceedings of the IEEE Winter Conference on Applications of Computer Vision (WACV) 2021 citation: Sensoy, M; Saleki, M; Julier, S; Aydogan, R; Reid, J; (2021) Misclassification Risk and Uncertainty Quantification in Deep Classifiers. In: Proceedings of the IEEE Winter Conference on Applications of Computer Vision (WACV) 2021. (pp. pp. 2483-2491). Institute of Electrical and Electronics Engineers (IEEE) Green open access document_url: https://discovery.ucl.ac.uk/id/eprint/10132169/1/Sensoy_Misclassification_Risk_and_Uncertainty_Quantification_in_Deep_Classifiers_WACV_2021_paper.pdf