DISC combines time-series similarity metrics to predict transfer
Original title: DISC: Domain-Integrated Similarity Cost for Time-Series Dataset
After calibration, DISC scores pairs from metric values and tracks forecasting error better than any single metric.
DISC is a learnable aggregation that selects, weights, and combines similarity metrics using calibration transfer observations. The authors correlate source-target similarity with downstream forecasting errors in out-of-distribution and transfer-learning settings. They report that dataset similarity is task-dependent and that integrated metrics align with transfer better than one hand-chosen metric.
- The method avoids exhaustive evaluation of every source-target candidate pair.
- Once calibrated, scores use only the pair's similarity-metric values.
- Higher alignment is reported on the paper's evaluation split.
From the original article
1 min read at huggingface.co
The growth of time-series datasets and foundation models has made dataset comparison a central part of time-series workflows. Assessing source-target transfer before model adaptation requires similarity estimates that can summarize transfer-related dataset relationships without exhaustive source-target evaluation over every candidate pair.
By Yen-Ku Liu, Hongjie Chen, Franck Dernoncourt · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.