Critical periods support representation learning in a model of cortical processing
Exposure of the brain to multiple stimuli drives the development of cortical representations, likely controlled by rules of synaptic plasticity. However, the type of developmental plasticity rules that lead to high-level representations of objects are unknown. Here we study a generalized Hebbian plasticity model that includes a predictive component. The learning rule uses only quantities that are locally available at the site of the synapse, is consistent with recent plasticity experiments in pyramidal neurons, and, as opposed to backpropagation learning, does not need a detailed feedback architecture. Our model shows that limiting plasticity in time to critical periods of development improves the quality and stability of sensory representation across different cortical areas described as layers of an artificial neural network. Our model achieves state-of-the-art performance for bio- plausible plasticity models on both an abstract hierarchical object database and a large image dataset designed for unsupervised learning.