Research map / Label-efficient and robust learning / Recognition and long-tail
Coherent Hierarchical Multi-Label Classification Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use C-HMCNN(h) instead of the strongest listed baseline | real-world HMC benchmarks · 20 datasets | AU(̅PRC) | Mixed |
| Add MCM to the underlying network (ablation) | h · Funcat validation datasets | AU(̅PRC): Cellcycle: 0.220 -> 0.229; Eisen: 0.262 -> 0.271; Expr: 0.246 -> 0.265 | Helped |
| Add MCM to the underlying network (ablation) | h · Derisi validation | AU(̅PRC): 0.179 -> 0.179 | No clear effect |
| Add MCLoss alongside MCM instead of standard BCE (ablation) | h+MCM · Funcat validation datasets | AU(̅PRC): Seq: 0.269 -> 0.274; Cellcycle: 0.229 -> 0.232 | Helped |
| Replace standard BCE with MCLoss while retaining MCM (ablation) | h+MCM · Nine-rectangle synthetic, A5 | AU(̅PRC): 0.938 -> 0.974 | Helped |
| Use MCLoss rather than standard BCE with MCM (ablation) | h+MCM · Nine-rectangle synthetic, A5 | AU(̅PRC) standard deviation: 0.038 -> 0.007 | Helped |
| Use coherent max-constraint predictions instead of post-processing only (ablation) | h · Funcat validation datasets | AU(̅PRC): Seq: 0.245 -> 0.274; Eisen: 0.262 -> 0.285 | Helped |
| Train with C-HMCNN(h) rather than the strongest competing model | HMC benchmarks · GO Expr test | AU(̅PRC): 0.452 -> 0.447 | Hurt |
| Train with C-HMCNN(h) rather than the strongest competing model | HMC benchmarks · GO Gasch2 test | AU(̅PRC): 0.465 -> 0.414 | Hurt |
| Train with C-HMCNN(h) rather than the strongest competing model | HMC benchmarks · GO Seq test | AU(̅PRC): 0.447 -> 0.446 | Hurt |
| Use one coherent model across differing class-region overlaps rather than either fixed post-processing strategy | synthetic rectangle task · Nine overlap positions | AU(̅PRC) | Helped |
Closest papers by meaning
- Hierarchical Multi-Label Classification Networks (ICML 2018 · similarity 0.92)
- Self-Paced Unified Representation Learning for Hierarchical Multi-Label Classification (AAAI 2024 · similarity 0.88)
- A no-regret generalization of hierarchical softmax to extreme multi-label classification (NeurIPS 2018 · similarity 0.87)
- Taming the Sigmoid Bottleneck: Provably Argmaxable Sparse Multi-Label Classification (AAAI 2024 · similarity 0.84)
- Enhancing Multi-Label Classification via Dynamic Label-Order Learning (AAAI 2024 · similarity 0.84)
- MLC-NC: Long-Tailed Multi-Label Image Classification Through the Lens of Neural Collapse (AAAI 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.