BHN: A Brain-like Heterogeneous Network
share
The human brain works in an unsupervised way, and more than one brain region is essential for lighting up intelligence. Inspired by this, we propose a brain-like heterogeneous network(BHN) which can cooperatively learn distributed representations, like the cortex, and a global contextual representation, like the medial temporal lobe(MTL). By optimizing a distributed, self-supervised and gradient-isolated contrastive loss function in a discriminative adversarial fashion, our model successfully learns to extract useful representations from video data. Methods developed in this work may help to solve some key problems in pursuit of human-level intelligence.
READ FULL TEXT