Abstract
Detecting a potentially threatening stimulus is a crucial operation for the survival of any organism and is often related with the expression of fear. Fear is a well studied emotion because apart from being an innate response it can also be learned by the process of conditioning, through which an otherwise neutral stimulus acquires affective properties. Although there is a growing understanding of this mechanism’s processing of visual stimuli, sufficient techniques for visual fear conditioning have not yet been produced. Further, although this is a highly evolved and efficient system it has never been explored as an alternative architecture for complex threat detection tasks. This system is comprised of a cortical and sub-cortical pathway that transmits stimuli to the amygdala, the neural structure responsible for threat evaluation. In this work, we provide an abstract yet biologically plausible representation of the participating neural structures in this process which is evaluated against abstract and real-world two-dimensional visual inputs. Our model provided computational evidence for the influence of the sub-cortical pathway on the amygdala under simulations of visual masking experiments. In addition, its use for the task of emotional face classification has verified its capability to handle complex inputs achieving a classification accuracy of 77% for the six basic emotions. In order to further strengthen the biological plausibility of our model, we have shown that its mechanics are adequate to develop synaptic formations that approximate centre-surround and orientation selective visual receptive fields. Finally, we have demonstrated that our architecture can be extended to biologically plausible simulations of neural structures such as the superior colliculus (SC). We reproduced behaviour observed in the SC, such as coordinate alignment, multisensory integration of abstract audio and visual inputs, and the effect of cortical feedback on multisensory responses.