AWX: An Integrated Approach to Hierarchical-Multilabel Classification
The recent outbreak of works on artificial neural networks (ANNs) has reshaped the machine learning scenario. Despite the vast literature, there is still a lack of methods able to tackle the hierarchical multilabel classification (HMC) task exploiting entirely ANNs. Here we propose AWX, a novel appr...
Saved in:
Published in | Machine Learning and Knowledge Discovery in Databases pp. 322 - 336 |
---|---|
Main Authors | , |
Format | Book Chapter |
Language | English |
Published |
Cham
Springer International Publishing
|
Series | Lecture Notes in Computer Science |
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The recent outbreak of works on artificial neural networks (ANNs) has reshaped the machine learning scenario. Despite the vast literature, there is still a lack of methods able to tackle the hierarchical multilabel classification (HMC) task exploiting entirely ANNs. Here we propose AWX, a novel approach that aims to fill this gap. AWX is a versatile component that can be used as output layer of any ANN, whenever a fixed structured output is required, as in the case of HMC. AWX exploits the prior knowledge on the output domain embedding the hierarchical structure directly in the network topology. The information flows from the leaf terms to the inner ones allowing a jointly optimization of the predictions. Different options to combine the signals received from the leaves are proposed and discussed. Moreover, we propose a generalization of the true path rule to the continuous domain and we demonstrate that AWX’s predictions are guaranteed to be consistent with respect to it. Finally, the proposed method is evaluated on 10 benchmark datasets and shows a significant increase in the performance over plain ANN, HMC-LMLP, and the state-of-the-art method CLUS-HMC. Code related to this paper is available at: https://github.com/lucamasera/AWX. |
---|---|
ISBN: | 9783030109240 3030109240 |
ISSN: | 0302-9743 1611-3349 |
DOI: | 10.1007/978-3-030-10925-7_20 |