Dussap, Bastien
Blanchard, Gilles
Chérief-Abdellatif, Badr-Eddine
Chapter History
First Online: 18 September 2023
Ethical Statement
: Label shift quantification has uses in a number of application domains; the results in this paper are chiefly oriented towards theory and general methodology so that we don’t discuss an application in particular. We only mention that the flow cytometry data used for proof-of-concept experiments is publicly available from a reputable scientific consortium and has been up to our knowledge collected following all established ethical standards.The original <i>Classify & Count</i> [CitationRef removed] method for label shift quantification is known to inherit the potential biases of the classification method it is based on (i.e. the misclassification errors can be very unevenly distributed across classes and “favor” majority classes). The <i>Adjusted Classify & Count (ACC)</i> approach and related methods [CitationRef removed,CitationRef removed] aim at rectifying this bias. In the present paper, we aim at going one step further and analyze certain robustness properties of the proposed label shift quantification methods, and introduce the contaminated label shift (InternalRef removed) setting with the goal of investigating trustworthiness of such methods under mild violations of the standard Label Shift model. Certainly the robustness property is desirable for improved reliability in practice, but does not mean immunity against biases; additionally, the user should always be wary of stronger model violations between reference and test data, in particular class-conditional distribution shifts. We therefore recommend established good practice of regularly checking on control data possible biases or drifts from the model, in particular for sensitive applications.
Conference Information
Conference Acronym: ECML PKDD
Conference Name: Joint European Conference on Machine Learning and Knowledge Discovery in Databases
Conference City: Turin
Conference Country: Italy
Conference Year: 2023
Conference Start Date: 18 September 2023
Conference End Date: 22 September 2023
Conference Number: 23
Conference ID: ecml2023
Conference URL: https://2023.ecmlpkdd.org/
Peer Review Information (provided by the conference organizers)
Type: Double-blind
Conference Management System: CMT
Number of Submissions Sent for Review: 829
Number of Full Papers Accepted: 196
Number of Short Papers Accepted: 0
Acceptance Rate of Full Papers: 24% - The value is computed by the equation "Number of Full Papers Accepted / Number of Submissions Sent for Review * 100" and then rounded to a whole number.
Average Number of Reviews per Paper: 3.63
Average Number of Papers per Reviewer: 4.5
External Reviewers Involved: Yes
Additional Info on Review Process: Applied Data Science Track: 239 submissions, 58 accepted papers; Demo Track: 31 submissions, 16 accepted papers.