Controlling LIME Kernel Width to Achieve Comprehensible Explanations on Tabular Data.

Hai Duong, Lam Hoang,Bac Le

Integrated Uncertainty in Knowledge Modelling and Decision Making: 10th International Symposium, IUKM 2023, Kanazawa, Japan, November 2–4, 2023, Proceedings, Part II(2023)

Cited 0|Views4
No score
Abstract
LIME [ 9 ] is an Explainable AI (XAI) method that can offer local explanation for any Machine Learning model prediction. However, the design of LIME often leads to controversial problems in the explanation, which are mostly due to the randomness of LIME’s neighborhood generating process. In this paper, we contribute a method that can help LIME deliver comprehensible explanation by optimizing feature attribution and kernel width of the generating process. Our method ensures high level of features attribution while keeping kernel width lower than the default setting to remain high locality in the explanation. The study will focus mainly on LIME for tabular data.
More
Translated text
Key words
lime kernel width
AI Read Science
Must-Reading Tree
Example
Generate MRT to find the research sequence of this paper
Chat Paper
Summary is being generated by the instructions you defined