The article discusses Onlineensembling Network (OneNet), a new approach to time series forecasting models to address the concept drift problem. OneNet updates two models focusing on time dimension dependency and cross-variate dependency, respectively. It uses a reinforcement learning-based approach within the online convex programming framework, enabling dynamic adjustment of weights. OneNet reduces the online forecasting error by over 50% compared to the State-Of-The-Art method, proving its effectiveness in adapting to concept drifts.
Publication date: 22 Sep 2023
Project Page: https://github.com/yfzhang114/OneNet
Paper: https://arxiv.org/pdf/2309.12659
