PDFTime: Prototype-Guided Framework for Interpretable Time Series Classification
A new framework called PDFTime (Prototype-Guided Classification Sub-Task Decoupling Framework) has been proposed to enhance both generalization and interpretability in multivariate time series classification (TSC). The framework reformulates TSC as a multi-stage decision process, using learned prototypes to approximate class-conditional feature distributions in latent space, rather than direct feature-to-label mapping common in deep learning approaches. This addresses the limitation of existing models that conflate feature extraction and decision logic into an inseparable mapping. The paper is available on arXiv under identifier 2605.22055.
Key facts
- PDFTime is a prototype-guided framework for time series classification.
- It reformulates classification as a multi-stage decision process.
- It uses learned prototypes to approximate class-conditional feature distributions.
- It avoids direct feature-to-label mapping.
- The paper is published on arXiv with ID 2605.22055.
- The framework aims to improve generalization and interpretability.
- Time Series Classification (TSC) is a long-standing research problem.
- Deep learning has enabled substantial progress in TSC.
Entities
Institutions
- arXiv