PiXTime: A Model for Federated Time Series Forecasting with Heterogeneous Data across Nodes
Researchers have developed PiXTime, a new Transformer-based framework for federated time series forecasting that can handle heterogeneous data across different nodes. Unlike previous methods requiring uniform model architectures, PiXTime uses a parameter-decoupling approach with localized modules and a shared backbone to adapt to diverse data structures. This allows for collaborative learning and generalization even when nodes have varying temporal resolutions or variable channels, achieving state-of-the-art performance in heterogeneous environments. AI
IMPACT Enables collaborative forecasting on distributed, heterogeneous datasets, overcoming previous limitations in federated learning for time series.