ABSTRACT
The anterior cingulate cortex (ACC) has been shown to be activated by response conflict or error in a classification task. An influential computational model of the role of ACC proposes that following the generation of the conflict signal, the signal-to-noise ratio of prefrontal circuits is raised, leading to an improvement of performance. In machine learning, the margin of classification is the distance of the input to the current classificatory boundary. A high margin implies good confidence in the prediction. It can be shown that the amount of conflict in a set of neurons coding for a response on a set of activation values corresponds to the margin of classification in a decision task. The essential feature of the model consists in activating a learning mechanism when a conflict condition ensues as defined by the classification margin. Hence, as in boosting, learning progressively concentrates on those inputs having difficult classification. The details of the model are provided in the presentation.
