Incremental Learning Methods for Metaverse Intelligent Models with Distributed Data Awareness

Federated learning(FL) is an emerging machine learning method that enables users to coordinate distributed model training while protecting user data privacy. However, in the metaverse scenario, the data distribution possessed by metaverse clients changes rapidly over time, causing models to gradually forget old knowledge while learning new knowledge, a phenomenon known as catastrophic forgetting. To address this challenge, we propose an Incremental Learning Method Based on Distributed Data Awareness (ILDDA).ILDDA integrates distributed data classification, model expansion, and block aggregation mechanisms, as well as data-free knowledge distillation techniques based on coarse class centers, to optimize the federated incremental learning process, thereby enhancing the accuracy of streaming data classification. Simulation results indicate that ILDDA can significantly improve model performance compared to baseline algorithms and effectively reduce catastrophic forgetting.