Neural Collapse Theory Reformulated for Imbalanced and Open-Set Datasets

Researchers have theoretically reformulated neural collapse (NC) to account for imbalanced datasets and open-set scenarios. Their analysis, based on class-wise averaged losses rather than global averages, demonstrates that standard NC is achieved when these class-wise losses reach their minima, even with imbalanced data and for cross-entropy and binary cross-entropy losses. The study further extends this to open-set situations, showing that classifier vectors for unknown classes overlap and become orthogonal to known classes under minimal loss conditions. AI

IMPACT Provides a theoretical framework for understanding model behavior on challenging, real-world data distributions.

RANK_REASON Academic paper detailing theoretical advancements in machine learning concepts. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• Binary Cross-Entropy
• CatalyzeX
• cross entropy
• DagsHub
• Gotit.pub
• Hugging Face
• IArxiv
• neural collapse
• ScienceCast
• sigmoid function
• Softmax

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →