SC3D: Dynamic and Differentiable Causal Discovery for Temporal and Instantaneous Graphs
This work addresses a key challenge in causal discovery for time series data, offering a method that could benefit fields like economics or neuroscience, though it appears incremental as it builds on existing temporal baselines.
The paper tackles the problem of discovering causal structures from multivariate time series, which involves lagged and instantaneous dependencies, and proposes SC3D, a two-stage differentiable framework that achieves improved stability and more accurate recovery of causal structures compared to existing baselines.
Discovering causal structures from multivariate time series is a key problem because interactions span across multiple lags and possibly involve instantaneous dependencies. Additionally, the search space of the dynamic graphs is combinatorial in nature. In this study, we propose \textit{Stable Causal Dynamic Differentiable Discovery (SC3D)}, a two-stage differentiable framework that jointly learns lag-specific adjacency matrices and, if present, an instantaneous directed acyclic graph (DAG). In Stage 1, SC3D performs edge preselection through node-wise prediction to obtain masks for lagged and instantaneous edges, whereas Stage 2 refines these masks by optimizing a likelihood with sparsity along with enforcing acyclicity on the instantaneous block. Numerical results across synthetic and benchmark dynamical systems demonstrate that SC3D achieves improved stability and more accurate recovery of both lagged and instantaneous causal structures compared to existing temporal baselines.