Core Concepts
This paper introduces FORMED, a novel foundation model for medical time series classification, repurposed from a pre-trained time series forecasting model, demonstrating superior generalizability and data efficiency compared to traditional and task-specific adaptation models.
Huang, N., Wang, H., He, Z., Zitnik, M., & Zhang, X. (2024). Repurposing Foundation Model for Generalizable Medical Time Series Classification. arXiv preprint arXiv:2410.03794.
This paper addresses the challenge of generalizing medical time series (MedTS) classification models across datasets with varying channel configurations, time series lengths, and diagnostic tasks. The authors propose a novel approach to repurpose a foundation model pre-trained on large-scale time series data for enhanced generalizability in MedTS classification.