Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
AI Research
arXiv CS.AI·7d ago·5 min read
Label-Efficient Time Series Classification at Scale: A Dual-Stream OSSE-LSTM with Counterfactual Attribution
TL;DR
A novel Dual-Stream OSSE-LSTM framework enables efficient time series classification with minimal labeled examples, achieving high accuracy across various datasets. It maintains an impressive accuracy range of 96.36-96.72% with just a few labeled instances per class.
✦ Why It Matters
Implement the Dual-Stream OSSE-LSTM framework to improve your time series classification tasks with fewer labeled examples.
Key Takeaways
- Utilize Dual-Stream OSSE-LSTM for efficient time series classification.
- Leverage Counterfactual Integrated Gradients for explainable AI decisions.
- Achieve high accuracy with minimal labeled data in your projects.
- Test the framework on diverse datasets to validate performance.
How It Works
The Dual-Stream OSSE-LSTM framework integrates an Omni-Scale CNN for multi-scale motif extraction and a Bidirectional LSTM to capture temporal context. This dual-stream approach normalizes and fuses outputs into a prototype-oriented embedding, allowing for effective classification with limited labeled examples.
Counterfactual Integrated Gradients (C-IG) are used to explain decisions by analyzing the margins between class prototypes, enhancing interpretability.