notesum.ai

Published at November 7

Interplay between Federated Learning and Explainable Artificial Intelligence: a Scoping Review

cs.LG

cs.AI

Released Date: November 7, 2024

Authors: Luis M. Lopez-Ramos, Florian Leiser, Aditya Rastogi, Steven Hicks, Inga Strümke, Vince I. Madai, Tobias Budig, Ali Sunyaev, Adam Hilbert

Arxiv: http://arxiv.org/abs/2411.05874v1

Refer to caption

Technique	Description	Manifestations in sample
Feature relevance	Calculate relevance scores for model variables.	e.g., SHAP [36, 41, 56],
		Custom builds [55, 63]
Local explanations	Estimate whole model through less complex subsystems.	e.g., GradCAM [64, 52, 65],
		heatmaps [66]
Simplification	Facilitate model while maintaining performance.	e.g., simpler models [67, 60]
Text explanations	Generate symbols that explain the results of the model.	-
Visual explanations	Visualize the inference process of the model.	-
Explanations by example	Provide representative examples that allow insight into the model.	-
Algorithmic transparency	Enable users to follow and understand the processes by the model.	e.g., Linear regression [68],
		Decision trees [69, 70]
Decomposability	Explain each model part separately for full comprehension.	e.g., Inference splits [71]
Simulatability	The inference of models could be simulated by a human.	e.g., Rule-based systems [72]