Friston Karl
The Wellcome Department of Cognitive Neurology, University College London, Queen Square, United Kingdom.
Annu Rev Neurosci. 2002;25:221-50. doi: 10.1146/annurev.neuro.25.112701.142846. Epub 2002 Mar 19.
Unsupervised models of how the brain identifies and categorizes the causes of its sensory input can be divided into two classes: those that minimize the mutual information (i.e., redundancy) among evoked responses and those that minimize the prediction error. Although these models have the same goal, the way that goal is attained, and the functional architectures required, are fundamentally different. This review describes the differences, in the functional anatomy of sensory cortical hierarchies, implied by the two models. We then consider how neuroimaging can be used to disambiguate between them. The key distinction reduces to whether backward connections are employed by the brain to generate a prediction of sensory inputs. To ascertain whether backward influences are evident empirically requires a characterization of functional integration among brain systems. This review summarizes the approaches to measuring functional integration in terms of effective connectivity and proceeds to address the question posed by the theoretical considerations. In short, it will be shown that the conjoint manipulation of bottom-up and top-down inputs to an area can be used to test for interactions between them, in elaborating cortical responses. The conclusion, from these sorts of neuroimaging studies, points to the prevalence of top-down influences and the plausibility of generative models of sensory brain function.
一类是使诱发反应之间的互信息（即冗余度）最小化的模型，另一类是使预测误差最小化的模型。尽管这些模型目标相同，但实现该目标的方式以及所需的功能架构却存在根本差异。本综述描述了这两种模型在感觉皮层层级功能解剖学方面所隐含的差异。接着，我们探讨如何利用神经成像来区分这两种模型。关键区别在于大脑是否利用反向连接来生成感觉输入的预测。要确定反向影响在经验上是否明显，需要对大脑系统之间的功能整合进行描述。本综述总结了从有效连接性角度测量功能整合的方法，并着手解决理论思考所提出的问题。简而言之，将表明对一个区域进行自下而上和自上而下输入的联合操作，可用于测试它们在阐述皮层反应时的相互作用。这些神经成像研究得出的结论表明自上而下影响的普遍性以及感觉脑功能生成模型的合理性。
