Spiking neural network operation system and method for brain-like intelligence and cognitive calculation

Abstract

The invention discloses an impulsive neural network operation system and method for brain-like intelligence and cognitive calculation. The system comprises a model description module, a parameter database, a configuration description module, a configuration manager, a rule manager, a data manager, a network builder, a network manager, an operation manager, a scheduler, a log manager, an operationmonitoring module and an imaging display module. The system provides a function of automatically executing synapse and neuron clipping and neogenesis according to certain conditions and rules; and multiple flexible rules for starting trigger conditions and execution processes of clipping and neogenesis processes are provided, so that the burden that a neural network developer needs to compile synapses and neuron clipping and neogenesis programs by himself / herself is avoided, and some problems existing in an existing brain-like pulse neural network operation framework are effectively solved.

Description

technical field [0001] The invention relates to the technical field of brain-inspired pulse neural network simulation and high-performance computing, and more specifically relates to a pulse neural network computing system and method for brain-inspired intelligence and cognitive computing. Background technique [0002] At present, brain-inspired intelligence and cognitive computing are based on spiking neural networks, combined with a variety of neurotransmitters, neuromodulators, receptors, electrical synapses, chemical synapses, dendrites, neurons, and glia in the biological brain The rich working mechanism of cells is used for computational modeling, and the constructed neural circuits, neural nuclei, brain regions and whole brain models can simulate many cognitive mechanisms and behaviors of biological brains, such as memory and learning, simulated emotions, navigation and Planning, motor control, brain-like vision and hearing, attention, decision-making, etc., provide a...

AI Technical Summary

Problems solved by technology

[0004] 1. It does not provide a unified and flexible modeling method, nor can it support flexible network topology structures, so it cannot achieve unity between modeling scale and modeling richness

For example, in terms of modeling scale, most of the existing brain-like spiking neural network computing frameworks cannot support and unify at the molecular scale, neurotransmitter and neuromodulator scale, receptor scale, synapse scale, dendrite scale, neuron scale, Modeling at scale, glial cell scale, neural circuit scale, neural nucleus scale, nerve fiber tract scale, brain region scale, whole brain scale, and behavioral scale

As another example, in terms of modeling richness, the existing brain-like spiking neural network computing frameworks often ignore the modeling of electrical synapses, the working mechanism of simulated neuromodulators, and the support of simulated dendrites. The mechanism for exchanging information and performing logical operations between multiple synapses in the network cannot support the topology of synapses and synapses directly connected, etc.

[0005] 2. Does not provide a description method that mimics the organization of the biological brain nervous system

[0006] 3. It does not integrate enough rich built-in functions and flexible user interface. If it cannot support automatic cutting and regeneration of synapses and neurons, users need to write programs to realize related functions.

[0007] 4. Impossible to effectively combine spiking neural networks, traditional neural networks, and other algorithms that use tensor as the main data representation method for mixed operations

[0008] 5. It is not compatible with CPU, ARM, GPU, DSP and other chip architectures, and most of the hardware resources are not optimized enough, so it is not suitable for efficient operation of medium to large-scale brain-like neural networks on general-purpose computers and embedded devices

[0009] The existence of the above problems has led to the limited modeling range of the existing brain-inspired neural network computing framework hardware platform, poor compatibility, low computing efficiency, inconvenient development and use, and it is difficult to achieve low-cost large-scale brain-inspired intelligence and cognitive computing Deployed in products such as smart toys, robots, drones, IOT devices, smart homes, and vehicle systems

Images