The paper presents a new approach to causal inference, utilizing stationary diffusion models. The method is based on learning stochastic differential equations (SDEs) that model a system’s behavior under interventions. Unlike traditional causal graphs, these models do not require acyclicity. The authors argue that this approach can generalize to unseen interventions on variables, often more effectively than classical methods. The method is based on a new theoretical result that expresses a stationarity condition on the diffusion’s generator in a reproducing kernel Hilbert space. The resulting kernel deviation from stationarity (KDS) is an objective function of independent interest.
Publication date: 26 Oct 2023
Project Page: https://arxiv.org/abs/2310.17405v1
Paper: https://arxiv.org/pdf/2310.17405
