Prediction and control of primate amygdala via electrophysiology and deep neural networks

• Applicant: Dr. Alina Peter
• Subject Area: Cognitive, Systems and Behavioural Neurobiology ; Experimental and Theoretical Network Neuroscience
• Term: from 2021 to 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 465345441

The amygdala is a multisensory brain structure critically involved in emotional processing. In primates, it responds to complex visual stimuli including animals and faces. Vision is a dominant sense in primates, involving large parts of neocortex. Nevertheless, how the strong visual inputs from neocortex are processed in the primate amygdala is currently underexplored. I propose to combine electrophysiological recordings with deep artificial neural network (ANN) models of the ventral visual (“What”) stream involved in object recognition to improve understanding of such processing. In a first step, a systematic survey of responses to a large stimulus set will be collected, which will allow an unbiased assessment of visual response preferences for different object classes. In a second step, these data also enable the generation of a new model of the visually-driven responses of the primate amygdala. Such a deep ANN model is able to make precise predictions about the responses of the amygdala to arbitrary images, including those never before “seen” by the model (“prediction”). In a third step, the model will then be used to precisely engineer artificial images that, when viewed by the animals, finely manipulate (“control”) single-neuron responses, for the first time in the amygdala. ANN models generate images highly unintuitive for human observers. Besides providing the potential for neural control, a deep ANN model enables fine-grained comparisons between neural selectivity in the amygdala and the ventral stream. Since it makes predictions about response strength for arbitrary images, it can also guide detailed recordings of further image classes, resulting in a mutually informative, iterative procedure of data collection and modeling. This procedure may be a new avenue for a better understanding of the primate amygdala, which in time can be expanded in terms of the scope of the stimuli, tasks performed on the stimuli, brain areas, and the ability of models to explain and manipulate activity, ultimately uncovering mechanisms of emotional processing of sensory stimuli.

• DFG Programme: WBP Fellowship
• International Connection: USA
• Host: Professor James J. DiCarlo, Ph.D.