EnJoi: Ensemble Joint Score Filter for Generative Data Assimilation

Recent research introduces EnJoi, a diffusion-based assimilation algorithm that recovers dynamical system states by learning the joint distribution of past and future states to balance confidence and observations. Simultaneously, the Conformal Adversarial Generative Ensemble (CAGE) framework improves time series forecasting. CAGE integrates generative modeling, adversarial discrimination, and conformal prediction. By utilizing p-values from nonconformity scores, CAGE dynamically adjusts model weights to reduce the influence of extreme forecasts, thereby enhancing overall reliability and accuracy.