Institutional Repository of Vilnius University
Influence functions - comparison of existing methods /
Abstract
dc:descriptionIn order to solve the problem of explainability of machine learning (ML) models, many solutions have been suggested in the recent years. One of them is a technique from robust statistics - influence functions (IF). However, many papers indicate that IF return outliers and mislabelled data as the most influential points from training set. In order to identify relevant examples that are more understandable for end users, alternative method was suggested: relative influence functions (RelatIF). In this paper, we compare how this method works on models where influence functions fail.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Balkytė, Austėja,
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD146235357&prefLang=en_US