Long Tailed Classification

Long-tailed classification addresses the challenge of training machine learning models on datasets where class frequencies are highly skewed, with a few dominant "head" classes and many under-represented "tail" classes. Current research focuses on developing novel loss functions, improving model architectures (like Vision Transformers and Graph Neural Networks), and employing techniques such as data augmentation, parameter-efficient fine-tuning, and ensemble methods to mitigate the bias towards head classes and improve performance on tail classes. This field is crucial for real-world applications where data imbalance is common (e.g., medical image analysis, software vulnerability detection), impacting the reliability and generalizability of machine learning models in diverse domains.