Hierarchical Multi-Label Classification Networks
One of the most challenging machine learning problems is a particular case of data classification in which classes are hierarchically structured and objects can be assigned to multiple paths of the class hierarchy at the same time. This task is known as hierarchical multi-label classification (HMC), with applications in text classification, image annotation, and in bioinformatics problems such as protein function prediction. In this paper, we propose novel neural network architectures for HMC called HMCN, capable of simultaneously optimizing local and global loss functions for discovering local hierarchical class-relationships and global information from the entire class hierarchy while penalizing hierarchical violations. We evaluate its performance in 21 datasets from four distinct domains, and we compare it against the current HMC state-of-the-art approaches. Results show that HMCN substantially outperforms all baselines with statistical significance, arising as the novel state-of-the-art for HMC.
PDF AbstractResults from the Paper
Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
---|---|---|---|---|---|---|
Hierarchical Multi-label Classification | Cellcycle Funcat | HMCN-F | AU(PRC) | 0.252 | # 2 | |
Hierarchical Multi-label Classification | Cellcycle GO | HMCN-F | AU(PRC) | 0.400 | # 2 | |
Hierarchical Multi-label Classification | Derisi Funcat | HMCN-F | AU(PRC) | 0.193 | # 2 | |
Hierarchical Multi-label Classification | Derisi GO | HMCN-F | AU(PRC) | 0.369 | # 2 | |
Hierarchical Multi-label Classification | Eisen Funcat | HMCN-F | AU(PRC) | 0.298 | # 2 | |
Hierarchical Multi-label Classification | Eisen GO | HMCN-F | AU(PRC) | 0.440 | # 2 | |
Hierarchical Multi-label Classification | Expr Funcat | HMCN-F | AU(PRC) | 0.301 | # 2 | |
Hierarchical Multi-label Classification | Expr GO | HMCN-F | AU(PRC) | 0.452 | # 1 | |
Hierarchical Multi-label Classification | Gasch1 Funcat | HMCN-F | AU(PRC) | 0.284 | # 2 | |
Hierarchical Multi-label Classification | Gasch1 GO | HMCN-F | AU(PRC) | 0.428 | # 2 | |
Hierarchical Multi-label Classification | Gasch2 Funcat | HMCN-F | AU(PRC) | 0.254 | # 2 | |
Hierarchical Multi-label Classification | Gasch2 GO | HMCN-F | AU(PRC) | 0.465 | # 1 | |
Hierarchical Multi-label Classification | Seq Funcat | HMCN-F | AU(PRC) | 0.291 | # 2 | |
Hierarchical Multi-label Classification | Seq GO | HMCN-F | AU(PRC) | 0.447 | # 1 | |
Hierarchical Multi-label Classification | Spo Funcat | HMCN-F | AU(PRC) | 0.211 | # 2 | |
Hierarchical Multi-label Classification | Spo GO | HMCN-F | AU(PRC) | 0.376 | # 2 |