DISC: Learnable Metric for Time-Series Dataset Similarity
October 21, 2026
DISC is a learnable aggregate metric designed to predict downstream forecasting error in transfer learning settings. It measures how well a source dataset will perform on a target task without requiring exhaustive model evaluation.
HOW THIS AFFECTS YOU
●
builderYou can use this to automate the selection of pre-trained models for specific time-series tasks.
●
researcherThis provides a more reliable way to quantify domain shift in time-series foundation models.