Reproduce “Hierarchy-Consistent Learning and Adaptive Loss Balancing for Hierarchical Multi-Label Classification” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ruobing Jiang, Mengzhe Liu, Haobing Liu, Yanwei Yu
HCAL targets hierarchical image classification by linking levels through child-to-parent feature aggregation and class prototypes, while combining level-specific losses with weights derived from their softmax-normalized values. On FGVC-Aircraft, using levels 2–4, it reaches 0.6829 level-2 accuracy versus 0.6328 for CAGrad and HVR 0.1012 versus 0.1116, indicating fewer measured parent–child mismatches in this comparison, not robustness beyond the tested datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.