Through interdisciplinary collaboration, Liu was mentored by Dr. Ruogu Fang and Dr. Mingzhou Ding. Liu presented his research on deep neural network model for perceiving emotions in affective images that includes both a visual cortex module and an amygdala/orbital frontal cortex (OFC) module. The model was trained to predict the valence from affective images (training data) and tested against the normative valence ratings of similar images (testing data). The activities of the amygdala/OFC module of the fully trained model were also compared with fMRI data recorded from human subjects viewing images using the representational similarity analysis. The results show that (1) the model was able to predict valence with high fidelity (R>0.6) for all four datasets, (2) the model’s performance was considerably worse without the “ILF” mediated pathway, and (3) activities in the second-to-last and last layer of the amygdala/OFC module were comparable with activities in the human amygdala and OFC respectively. These results demonstrated that our DNN model has the capability to recognize the emotional content of affective images and can become a platform for formulating and testing neuroscientific hypotheses.
The goal of MAIN is to bring together trainees, researchers including world leaders in Neuroscience and AI from across the globe, and build a more inclusive and diverse Neuro-AI community.