Get Started
Research questionHow can classification models adapt to removed labels without retraining on the original data?Removing a category changes the model’s output space, while its learned representations and confidence scores still reflect the old taxonomy. Updating predictions under these conditions is difficult without the data and computation required for retraining.
AI
Machine Learning
Statistical Machine Learning
Latest papersRecent research connected to this question, newest first.OSR: Output Space Redistribution for Adaptive Label Removal in Classification ModelsThe source concerns classification models undergoing label removal in evolving taxonomies. It describes adaptation using existing labels and prior output confidences without access to original training data, and reports experiments across several classification tasks against full retraining; the evidence is limited to those reported evaluations.research paper · Sep 3, 2026
Related questions
How can we measure label-preserving adversarial vulnerability separately from ordinary misclassification in high-dimensional classifiers?How can machine-learning performance-modeling frameworks stay maintainable as models and systems evolve?How can online time-series forecasters adapt to evolving, unseen patterns without costly continual parameter updates?How can climate disclosure classifiers remain reliable when documents shift across sources?
Home
Topics
Search
Library