Learning structure of sensory inputs with synaptic plasticity leads to interference

摘要: Synaptic plasticity is often explored as a form of unsupervised adaptation in cortical microcircuits to learn the structure complex sensory inputs and thereby improve performance classification prediction. The question whether specific input patterns encoded neural networks has been largely neglected. Existing studies that have analyzed input-specific structural used simplified, synthetic contrast noisy found real-world data. In this work, changes are for three empirically derived models applied temporal tasks include complex, visual auditory Two forms spike-timing dependent (STDP) Bienenstock-Cooper-Munro (BCM) rule adapt recurrent network during training process before tested on pattern recognition tasks. It shown synaptic highly sensitive classes pattern. However, does not tasks, partly due interference between consecutively presented samples. strength produced by one stimulus reversed presentation another, thus preventing from being retained network. To solve problem interference, we suggest be extended restrict activity modification subset circuit, which increasingly case experimental neuroscience.