Coherent Hierarchical Multi-Label Classification Networks

Hierarchical multi-label classification (HMC) is a challenging classification task extending standard multi-label classification problems by imposing a hierarchy constraint on the classes. In this paper, we propose C-HMCNN(h), a novel approach for HMC problems, which, given a network h for the underlying multi-label classification problem, exploits the hierarchy information in order to produce predictions coherent with the constraint and improve performance. We conduct an extensive experimental analysis showing the superior performance of C-HMCNN(h) when compared to state-of-the-art models.

PDF Abstract NeurIPS 2020 PDF NeurIPS 2020 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Hierarchical Multi-label Classification Cellcycle Funcat C-HMCNN AU(PRC) 0.255 # 1
Hierarchical Multi-label Classification Cellcycle GO C-HMCNN AU(PRC) 0.413 # 1
Hierarchical Multi-label Classification Derisi Funcat C-HMCNN AU(PRC) 0.195 # 1
Hierarchical Multi-label Classification Derisi GO C-HMCNN AU(PRC) 0.37 # 1
Hierarchical Multi-label Classification Eisen Funcat C-HMCNN AU(PRC) 0.306 # 1
Hierarchical Multi-label Classification Eisen GO C-HMCNN AU(PRC) 0.455 # 1
Hierarchical Multi-label Classification Expr Funcat C-HMCNN AU(PRC) 0.302 # 1
Hierarchical Multi-label Classification Expr GO C-HMCNN AU(PRC) 0.447 # 2
Hierarchical Multi-label Classification Gasch1 Funcat C-HMCNN AU(PRC) 0.286 # 1
Hierarchical Multi-label Classification Gasch1 GO C-HMCNN AU(PRC) 0.436 # 1
Hierarchical Multi-label Classification Gasch2 Funcat C-HMCNN AU(PRC) 0.258 # 1
Hierarchical Multi-label Classification Gasch2 GO C-HMCNN AU(PRC) 0.414 # 2
Hierarchical Multi-label Classification Seq Funcat C-HMCNN AU(PRC) 0.292 # 1
Hierarchical Multi-label Classification Seq GO C-HMCNN AU(PRC) 0.446 # 2
Hierarchical Multi-label Classification Spo Funcat C-HMCNN AU(PRC) 0.215 # 1
Hierarchical Multi-label Classification Spo GO C-HMCNN AU(PRC) 0.382 # 1

Methods


No methods listed for this paper. Add relevant methods here