Supervisor: Johannes Fürnkranz

Motivation:

LIME [1] is one of the best-known approaches for interpreting a learned black-box model such as a deep neural network. Its key idea is to learn a simple local model such as a linear regression model in the local neighborhood of a point that needs to be explained. Single conjunctive rules could also be tried in such a setting.

Objective:

The task of this thesis is to implement and explore this approach using inductively learned rules as a base model, starting from simple strategies such as flipping each binary attribute, to more elaborate strategies for forming a neighborhood, such as the one proposed in [2].

Literature

[1] Marco Túlio Ribeiro, Sameer Singh, Carlos Guestrin: "Why Should I Trust You?": Explaining the Predictions of Any Classifier, opens an external URL in a new window. KDD 2016: 1135-1144.
[2] Riccardo Guidotti, Anna Monreale, Salvatore Ruggieri, Dino Pedreschi, Franco Turini, Fosca Giannotti: Local Rule-Based Explanations of Black Box Decision Systems, opens an external URL in a new window. CoRR abs/1805.10820 (2018)