Slice Discovery

Slice discovery focuses on identifying subsets of data where machine learning models underperform, aiming to improve model robustness and fairness by revealing systematic errors. Current research emphasizes developing algorithms that discover coherent and interpretable error slices, often leveraging techniques like clustering of model embeddings, influence functions, or cross-modal embeddings combined with natural language processing to generate descriptive labels. This work is crucial for enhancing the reliability and trustworthiness of machine learning models across various domains, particularly in sensitive applications like medical image analysis, by providing actionable insights into model biases and limitations.