arXivSajjad Pakdamansavoji, Mohammad Reza Rezaei, Ilija Subasic, Pedro Garzon, Radu Paradovschi, Zhao ChenMon, Nov 2, 2026, 9:37 AM PST
score 16.9
Method Shrinks Training Data by Matching Model Predictions
Original: Online Label-Free Gradient Matching for Continual Language Model Training
Source: arxiv.org ↗
Writing ELI5 summary…