GrAI Matter Labs (GML) invented a novel intelligent dataflow compute architecture in which we combined principles of biological systems, near memory compute and massively parallel cores. We call it NeuronFlow™. The computations within the NeuronFlow architecture are event-based, with execution of a neuron operation only occurring when new data arrives at that neuron. New data or events arrive at the neuron only if there are changes significant enough from the previous time instance. GML’s approach of NeuronFlow technology is enabling to avoid unnecessary computations and proceed only the key aspects of a deep neutral network.
