DISC: Domain-Integrated Similarity Cost for Time-Series Dataset
Abstract
DISC is a learnable framework that calibrates and combines time-series dataset similarity metrics to better predict downstream forecasting transfer performance.
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. Many metrics have been proposed for measuring similarity between time-series datasets, but they capture different dataset relationships and vary in how well they align with downstream utility across tasks. We evaluate this gap by measuring the correlation between source-target similarity scores and downstream forecasting errors in out-of-distribution and transfer-learning settings. We propose Domain-Integrated Similarity Cost (DISC), a learnable aggregation framework that uses calibration transfer observations to select, weight, and combine similarity metrics for different downstream tasks. By integrating complementary metric values, DISC better distinguishes source-target dataset differences that matter for downstream behavior. After calibration, DISC scores source-target pairs using only their similarity-metric values and yields higher alignment than single-metric references on the reported split. These results show that time-series dataset similarity is task-dependent, and that learned metric integration provides a stronger proxy for downstream transfer behavior than manually selecting one similarity metric.
Community
Before I'd trust a transfer-similarity score I'd want it benchmarked against the dumbest thing that works: fine-tune on 10% of the target for twenty minutes and see what comes out. If DISC can't beat that probe on cost, it's a nicer number, not a better decision. The catch is labels — a probe needs them, and on day one you have none, so the metric has to run on unlabeled target data or it's answering a question nobody asks at deploy time. That also shapes the honest eval: rank a pool of candidate sources, then check whether the top pick actually transfers better than a random pick, not whether the ranking correlates with some proxy label. I'd want to see it hold when the target is small — a few thousand windows, one sensor, gaps — because that's the regime where you're actually shopping for a pretrained checkpoint. Does the domain-integration term still earn its place there, or does it flatten into a plain distribution distance once the target is thin?
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- Loss-Guided Pretraining Data Selection for Time-Series Foundation Models (2026)
- CoRe: Coherence and Relational Alignment for Multivariate Time Series Forecasting (2026)
- TStruct: Learning Shared Temporal Structures for Long-Term Time-Series Forecasting (2026)
- Local Reference Geometry Residual Augmentation for Imbalanced Time Series Classification (2026)
- Generalist Representation, Specialist Detection: TS-Router for Time-Series Anomaly Detection (2026)
- GRAM: Correcting Frozen Time-Series Foundation Models via Graph-Retrieved Amplitude Memory (2026)
- TSAE: Structured Sparse Autoencoders for Interpreting Time-Series Forecasting Models (2026)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: @librarian-bot recommend
Get this paper in your agent:
hf papers read 2610.31288 Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper
Collections including this paper 0
No Collection including this paper