Communication sparsification method for pulse neural network calculation load
A technology of spiking neural network and computing load, applied in the field of deep learning, can solve problems such as the decline of computing efficiency, and achieve the effects of reducing the generation of pulses, improving computing efficiency, and reducing communication time
Pending Publication Date: 2021-05-25
苏州蓝甲虫机器人科技有限公司
View PDF0 Cites 2 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0005] In order to solve the problem that the computing efficiency of SNN gradually decreases with the increase of computing nodes in the distributed computing platform, the present invention provides a communication sparse method for the computing load of the pulse neural network, which effectively solves the problem that the communication efficiency of the distributed platform can be extended. With the gradual increase of computing nodes in the distributed computing platform, the problem of gradually decreasing computing efficiency is effectively solved
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0046] In this patent, a communication sparsification method for the calculation load of the pulse neural network, such as figure 2 The NEST simulator is shown as the SNN load characterization simulation tool, and the implementation steps of the patented technical solution are introduced in detail.
[0047] S1: Based on the distributed system architecture, construct the pulse neural network;
[0048] When constructing a spiking neural network, there is no need to create specific synapses, only the connections between neurons need to be constructed, and the neuron connection table can be obtained.
[0049] S2: Traverse the spiking neural network, find all nodes and neurons, and the neuron connection table.
[0050] When applying this patented technical solution on NEST, Create only needs to obtain the connection between neurons; when Connecting, no specific synaptic object will be created, but a connection table will be obtained through all connections.
[0051] S3: Redistri...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention provides a communication sparsification method for a pulse neural network calculation load, which effectively solves the problem of expandability of the communication efficiency of a distributed platform, and effectively solves the problem that the calculation efficiency is gradually reduced along with the gradual increase of calculation nodes in the distributed calculation platform. In the technical scheme of the invention, the neurons are redistributed on each node based on the redistribution operation, the neurons distributed on each node have the most post-synaptic neurons in the node, the nodes are based on a non-blocking communication mode, and in each communication process, each node asynchronously sends pulse data to all target nodes of the node and waits for receiving pulses sent by a source node of the node, that is, each node only sends necessary data to the target node of the communication and does not communicate with a non-target node, so that communication with a non-adjacent process is avoided.
Description
technical field [0001] The invention relates to the technical field of deep learning, in particular to a communication sparsification method for calculation load of a pulse neural network. Background technique [0002] With the development of the deep learning field, more and more brain neurocomputing science research works have emerged, hoping to overcome the shortcomings of the existing deep learning by analyzing the working mechanism of the brain and developing brain-like computing. The basis of brain-like computing is Spiking Neural Network (SNN, Spiking Neural Network). Compared with traditional deep neural network (DNN, Deep Neural Network), the working mechanism of SNN is closer to the biological brain. [0003] In order to obtain the best computing power and to get closer to the scale of brain computing, most of the existing technologies build large-scale clusters to form distributed computing platforms to perform brain-like computing. However, with the increase of ...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More IPC IPC(8): G06F15/173G06N3/04
CPCG06F15/173G06N3/049Y02D10/00
Inventor 柴志雷刘家航王涛白云王皓洋尤佳
Owner 苏州蓝甲虫机器人科技有限公司