Tingzhu Bi

We propose UnCLe, a deep learning framework that uses temporal perturbations to uncover dynamic, time-varying causal graphs from observational time series.