Causal Discovery from Conditionally Stationary Time Series

Carles Balsells-Rodas, Xavier Sumba, Tanmayee Narendra, Ruibo Tu, Gabriele Schweikert, Hedvig Kjellstrom, Yingzhen Li

TL;DR: We tackle theoretical and algorithmic challenges for conditionally stationary time series.

Abstract: Causal discovery, i.e., inferring underlying causal relationships from observational data, is highly challenging for AI systems. In a time series modeling context, traditional causal discovery methods mainly consider constrained scenarios with fully observed variables and/or data from stationary time-series. We develop a causal discovery approach to handle a wide class of nonstationary time series that are _conditionally stationary_, where the nonstationary behaviour is modeled as stationarity conditioned on a set of latent state variables. Named State-Dependent Causal Inference (SDCI), our approach is able to recover the underlying causal dependencies, with provable identifiablity for the state-dependent causal structures. Empirical experiments on nonlinear particle interaction data and gene regulatory networks demonstrate SDCI's superior performance over baseline causal discovery methods. Improved results over non-causal RNNs on modeling NBA player movements demonstrate the potential of our method and motivate the use of causality-driven methods for forecasting.

Lay Summary: Understanding cause and effect from data that changes over time, e.g. weather or player movements in sports, is a big challenge for AI systems. Most current methods assume these relationships stay the same over time, but real-world systems often change in complex ways. In this work, we develop a new approach called State-Dependent Causal Inference (SDCI). It allows us to find changing cause-and-effect relationships by assuming that the system is stable when we take into account hidden "states" behind the scenes. These states help explain why and how relationships shift over time. We show that our method can correctly uncover these changing patterns, even in real-world systems. This includes predicting how genes interact in cell organisms or how basketball players move during a game. We also prove a key mathematical property: identifiability. This means our method can uniquely determine the right explanation from the data. This is very important for making scientific discoveries you can trust.

Link To Code: https://github.com/charlio23/SDCI

Primary Area: Probabilistic Methods->Structure Learning

Keywords: causal discovery, graph neural network, time series, non-stationary, probabilistic modelling, identifiability

Submission Number: 10811