Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·Aug 6·5 min read
Personalized Federated Sparse Adaptation of Time-Series Foundation Models
TL;DR
A personalized federated sparse adaptation framework for time-series foundation models (TSFMs) enhances building energy forecasting by tailoring model adaptations to individual client needs. This approach outperforms traditional global and local training methods across multiple buildings and model backbones.
✦ Why It Matters
Implement personalized federated learning strategies to improve energy forecasting accuracy in your building management systems today.
Key Takeaways
- Adopt personalized federated learning for better model performance.
- Utilize mixture-of-experts for specialized temporal adaptations.
- Evaluate model performance across different building contexts.
- Consider client-specific needs in federated model training.
How It Works
The framework employs a heterogeneous temporal mixture-of-experts (MoE) adapter that uses a sequence-level router to map each 168-hour context window to a top-k subset of experts. These experts are specialized in various aspects of time-series data, such as periodicity and long-range interactions, allowing for tailored adaptations that enhance forecasting accuracy.