Deep Gated Hebbian Predictive Coding Accounts for Emergence of Complex Neural Response Properties Along the Visual Cortical Hierarchy

Predictive coding provides a computational paradigm for modeling perceptual processing as the construction of representations accounting causes sensory inputs. Here, we developed scalable, deep network architecture predictive that is trained using gated Hebbian learning rule and mimics feedforward feedback connectivity cortex. After training on image datasets, models formed latent in higher areas allowed reconstruction original images. We analyzed low- high-level properties such orientation selectivity, object selectivity sparseness neuronal populations model. As reported experimentally, increased systematically across ascending model hierarchy. Depending strength regularization factors, also from lower to areas. The results suggest rationale why experimental cortical hierarchy have been inconsistent. Finally, different classes became more distinguishable Thus, neural networks formulation can reproduce several associated with responses along visual