Nykamp Duane Q
School of Mathematics, University of Minnesota, 127 Vincent Hall, 206 Church Street, Minneapolis, MN 55455, USA.
Math Biosci. 2007 Feb;205(2):204-51. doi: 10.1016/j.mbs.2006.08.020. Epub 2006 Sep 5.
We describe an approach for determining causal connections among nodes of a probabilistic network even when many nodes remain unobservable. The unobservable nodes introduce ambiguity into the estimate of the causal structure. However, in some experimental contexts, such as those commonly used in neuroscience, this ambiguity is present even without unobservable nodes. The analysis is presented in terms of a point process model of a neuronal network, though the approach can be generalized to other contexts. The analysis depends on the existence of a model that captures the relationship between nodal activity and a set of measurable external variables. The mathematical framework is sufficiently general to allow a large class of such models. The results are modestly robust to deviations from model assumptions, though additional validation methods are needed to assess the success of the results.
我们描述了一种用于确定概率网络节点之间因果联系的方法，即使许多节点仍不可观测。不可观测节点给因果结构的估计带来了模糊性。然而，在某些实验情境中，比如神经科学中常用的那些情境，即使没有不可观测节点，这种模糊性也依然存在。尽管该方法可推广到其他情境，但这里是根据神经网络的点过程模型进行分析的。该分析依赖于一个能够捕捉节点活动与一组可测量外部变量之间关系的模型的存在。数学框架足够通用，能涵盖一大类这样的模型。尽管需要额外的验证方法来评估结果的成功与否，但结果对偏离模型假设的情况具有一定的稳健性。
