Researchers have developed a method called CJSD to help streaming systems decide whether to reuse an existing model, create a new one, or delay the decision. This is done by using two discriminators that can detect changes in the system's mechanism and covariate shift. The approach has been tested on synthetic data streams and shown to achieve zero false spawns and reuses after switches, outperforming a previously used heuristic. --- Why it matters: This matters because streaming systems with multiple models must often make decisions about when to reuse or replace them, which can be challenging due to the need to balance accuracy and resource usage. CJSD provides a statistically meaningful approach to making these decisions, which could improve the performance of such systems. Source: https://arxiv.org/abs/2608.19885