Announcement
Tiny recursive language model reportedly reaches state-of-the-art performance on long-context time series
The paper’s authors say they trained it with AnomalyXL, which generates synthetic time series with programmatic ground-truth answers.
TLDR
The authors say frontier time-series language models struggle to pinpoint anomalies in long recordings. Their new paper reports training a tiny recursive language model to state-of-the-art performance on long-context time series using AnomalyXL. They say the training transfers to anomaly detection in software observability, ECG and sleep data.
Combined views
36
5 Sources, first seen ago
