Concepts de base
稳态扩散模型为因果推断提供新方法,不受因果图形式限制。
Stats
我们学习随机微分方程(SDEs)以及它们引起的稳态密度。这些模型通常比传统方法更好地泛化到未见干预情况。
Citations
"Stationary SDEs induce a time-invariant stationary density µ over Rd, while they internally unroll the causal dependencies of the variables over time t, akin to real-world processes."
"Unlike inference of causal graphs and SCMs, which often exploits the statistical properties of particular functions, exogenous noise, or interventions, our learning algorithm for stationary diffusions is general and thus agnostic to the system and intervention model."