Saved in:
Bibliographic Details
Main Authors: Chen, Lu, Lou, Siyu, Huang, Benhao, Zhang, Quanshi
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2401.16318
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910602273751040
author Chen, Lu
Lou, Siyu
Huang, Benhao
Zhang, Quanshi
author_facet Chen, Lu
Lou, Siyu
Huang, Benhao
Zhang, Quanshi
contents Faithfully summarizing the knowledge encoded by a deep neural network (DNN) into a few symbolic primitive patterns without losing much information represents a core challenge in explainable AI. To this end, Ren et al. (2024) have derived a series of theorems to prove that the inference score of a DNN can be explained as a small set of interactions between input variables. However, the lack of generalization power makes it still hard to consider such interactions as faithful primitive patterns encoded by the DNN. Therefore, given different DNNs trained for the same task, we develop a new method to extract interactions that are shared by these DNNs. Experiments show that the extracted interactions can better reflect common knowledge shared by different DNNs.
format Preprint
id arxiv_https___arxiv_org_abs_2401_16318
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Defining and Extracting generalizable interaction primitives from DNNs
Chen, Lu
Lou, Siyu
Huang, Benhao
Zhang, Quanshi
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Faithfully summarizing the knowledge encoded by a deep neural network (DNN) into a few symbolic primitive patterns without losing much information represents a core challenge in explainable AI. To this end, Ren et al. (2024) have derived a series of theorems to prove that the inference score of a DNN can be explained as a small set of interactions between input variables. However, the lack of generalization power makes it still hard to consider such interactions as faithful primitive patterns encoded by the DNN. Therefore, given different DNNs trained for the same task, we develop a new method to extract interactions that are shared by these DNNs. Experiments show that the extracted interactions can better reflect common knowledge shared by different DNNs.
title Defining and Extracting generalizable interaction primitives from DNNs
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2401.16318