Brief

Researchers have developed new sampling bounds for regularized classification, achieving optimal $(1\pm\varepsilon)$-relative error for a wide range of Lipschitz continuous loss functions. The study presents improved sampling complexity bounds, specifically $k^2/\varepsilon^2$ for L2 regularization and $k/\varepsilon^2$ for L1 regularization. These findings rely on simple uniform or norm sampling and offer a significant improvement over previous sensitivity sampling bounds, utilizing refined arguments to avoid overcounting issues. AI

Researchers have investigated why low-dimensional representations, typically around 1000 dimensions, do not hinder the scalability of modern embedding-based retrieval models to trillions of data points. Their study focuses on maximal-margin embeddings, establishing that a near-optimal margin can be achieved with a dimension dependent on the logarithm of the data size. The findings resolve a previous setup concerning k-sparse rows and suggest that sigmoid loss outperforms InfoNCE for generating large-margin embeddings. AI