ANNA: An artificial neural network for attention to emotional recognition

Abstract

Emotional experience has two distinct components in human beings: 'automatic' and 'attended'. The former of these is based more heavily on the ventral and limbic areas of the brain; the attention part is concerned with cognitive aspects of experience, and involves more dorsal components. A rapidly increasing body of knowledge on these two separate components of human experience is being developed through brain imaging, single cell recording and deficit analyses under emotional as compared to neutral inputs. We start by summarizing this data. We then incorporate the data into a recently developed engineering control model of attention and motor responses. The crucial extension of this model involves a ventral/limbic brain network building representations of salience and valence. A simulation of a simple paradigm is used to demonstrate the considerable dissociation possible between the cognitive and emotional components. The system is developed to give an extension of standard artificial neural network architectures to a new class, in which attention effects are explicitly included through adaptive feedback modulation. Learning laws are developed which extend BEP to the attention case. An artificial emotion recognition system is developed as part of this architectural analysis.