A detailed study of recurrent neural networks used to model tasks in the cerebral cortex.
C. Jarne, Rodrigo Laje · arXiv (Cornell University) · 2019
Recurrent Neural Networks or RNN are frequently used to model different aspects of brain regions. We studied the properties of RNN trained to perform temporal and flow control tasks with temporal stimuli. We present the results regarding three aspects: inner configuration sets, memory capacity with the scale and immunity to induced damage on trained networks. Our results allow us to quantify different aspects of these physical models, which are normally used as black boxes and must be understood previous to modeling the biological response of cerebral cortex.