From Theoretical Physics to Experimental Neuroscience and Deep Learning

Can Physics assist with key challenges in Neuroscience and Artificial Intelligence? Are current theoretical techniques of statistical mechanics capable of dealing with brain dynamics? Let us swim against the stream when scientific evidence tells you something is wrong.

Century-old assumptions regarding neurons and brain learning are disproved. According to the long-lasting computational scheme, each neuron sums the incoming electrical signals through its dendrites and when the membrane potential reaches a certain threshold the neuron typically generates a spike. We present several types of experiments, indicating that each stochastic neuron functions as a collection of independent threshold units, where the neuron is anisotropically activated. In addition, experimental and theoretical results reveal a new underlying non-local mechanism for the fast brain learning process, dendritic learning, as opposed to learning which is based solely on slow synaptic plasticity, where just one single neuron can realize deep learning algorithms. Though the brain is a very slow machine, its capabilities exceed typical state-of-the-art ultrafast artificial intelligence algorithms; hence, a revolution in deep learning must emerge.