UGG-DA:Uncertainty-Guided Gradual Distribution Adaptation and Dynamic Prediction with Streaming Data

Streaming data prediction has had a great deal of success in a variety of situations as sensor technology has advanced. However, efficacy declines due to the typical evolution of data distribution over time. Traditional domain adaptation entails only a minor transition. We propose a novel progressive domain adaptation method to address this issue. Our method can reduce domain discrepancy for both large and minor shifts that have accumulated. Our approach designs a dynamic alignment strategy for pseudo labels based on uncertainty and confidence. Then, we provide a new theoretical bound for a scenario of gradual domain adaptation that is more stringent than traditional DA. The experimental results confirm the superiority of our method in the wind field dataset, demonstrating its viability and potential.