arxivcs.LG2026-07-01
LeNEPA: No-Augmentation Next-Latent Prediction for Time-Series Representation Learning
Alexander Chemeris, Ming Jin, Randall Balestriero
Time series are central to modern data mining applications, from industrial telemetry and server metrics to finance and physiology, yet time-series self-supervised learning often depends on view and augmentation choices that encode domain-specific invariances. We study how an SSL…