CN113222134B - Brain-like computing system, method and computer readable storage medium - Google Patents
Brain-like computing system, method and computer readable storage medium Download PDFInfo
- Publication number
- CN113222134B CN113222134B CN202110782692.XA CN202110782692A CN113222134B CN 113222134 B CN113222134 B CN 113222134B CN 202110782692 A CN202110782692 A CN 202110782692A CN 113222134 B CN113222134 B CN 113222134B
- Authority
- CN
- China
- Prior art keywords
- computing
- brain
- node
- result
- nodes
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5061—Partitioning or combining of resources
- G06F9/5072—Grid computing
Abstract
The invention relates to a brain-like computing system, a brain-like computing method and a computer-readable storage medium. According to the method, the computing job request of the user terminal is obtained through the login node and sent to the management node, then the management node distributes computing tasks to the computing job request, the distribution result is sent to the plurality of computing nodes, then the plurality of computing nodes execute parallel computing according to the distribution result, and finally the storage node performs distributed storage on the parallel computing result, so that the problems that the existing brain-like computing capability is not strong, the brain-like computing is not clear, effective intelligent analysis cannot be performed on high-density data and the like are solved, and strong computing support is provided for complex scene application such as big data analysis and the like.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a brain-like computing system, a brain-like computing method and a computer-readable storage medium.
Background
In recent years, the computing architecture is subject to the ceiling effect of memory walls, power consumption walls, processing walls, and scaling walls, and the von neumann architecture followed by the conventional computer is facing a great challenge. Therefore, there is a need for a high performance computing technique capable of dealing with high density data and high dimensional features to meet the demands of limited computing power and fast growth of data processing.
With the development of brain science, people gradually know that the human brain is a computer with extremely high energy efficiency, and brain-like computing is carried out at the same time. The basic idea of brain-like computing is to apply the concept of biological neural network to the design of a computer system, and the brain-like computing refers to the simulation of a brain nervous system and an information processing process to realize a high-performance and low-power-consumption computing system. The ultimate goal of brain-like computing is artificial general intelligence, also known as strong artificial intelligence, which is the ultimate goal in most artificial intelligence research fields. Researchers have been working to advance this goal through continued exploration in software and hardware design for decades.
It can be said that the brain-like computing brings huge development opportunities for big data applications, and simultaneously, presents many new challenges, and the basis of the present world computer science and engineering is constructed based on the "computing theory of turing-godel-qiu" and the "architecture of von neumann", however, the "computing theory" only solves the problem of computing, does not consider how to improve the computing power of the system, and the "architecture of von neumann" only solves the engineering and practical problems of the computing theory, and fails to care about the optimization of the computing power of the system.
In summary, in the face of big data application in important fields and key industries, a new architecture and method capable of dealing with poor brain-like computing ability and unclear brain-like computing cognition is urgently needed, and high-density data is effectively and intelligently analyzed, so as to meet the requirements of the electronic information industry on continuously improved computing performance and extremely low power consumption.
Disclosure of Invention
The invention aims to solve the technical problems that the existing brain-like computing capability is not strong, the brain-like computing is not clear, and effective intelligent analysis can not be performed on high-density data in the prior art, and provides a brain-like computing system, a method and a computer readable storage medium.
According to a first aspect of the present invention there is provided a brain-like computing system comprising: the system comprises a login node, a management node, a plurality of computing nodes and a storage node;
the login node is used for acquiring a calculation job request of a user terminal and sending the calculation job request to the management node;
the management node is used for distributing the computing tasks to the computing operation requests and sending distribution results to the computing nodes, wherein the computing nodes correspond to a plurality of parallel artificial neural networks, different computing tasks correspond to different computing nodes, different computing nodes correspond to different artificial neural networks, and the distribution results comprise the computing tasks, corresponding relations among the computing nodes and the artificial neural networks and task computing parameters;
the plurality of computing nodes are used for executing parallel computing according to the distribution result, acquiring a parallel computing result from the storage node, updating the state of the corresponding computing task according to the parallel computing result, and feeding back the parallel computing result and the state updating result to the management node; the plurality of computing nodes are in communication connection through the constructed full-switching network;
the storage node is used for performing distributed storage on the parallel computing result;
the computing nodes are provided with computing support by brain-like coprocessor components supporting artificial neural network modeling, and the brain-like coprocessor components are at least one processor or a combination of a deep learning processor, a neural network processor, a tensor processor and a vector processor; and/or
The brain-like coprocessor component comprises at least one hybrid coprocessor which simultaneously supports the artificial neural network computation;
and/or the brain coprocessor component comprises at least one processor combination and at least one hybrid coprocessor which simultaneously supports the artificial neural network computation.
In the brain-like computing system of the present invention, the login node is further configured to compile and configure parameters of an operation instruction initiated by the user terminal, so as to generate the computing job request.
In the brain-like computing system, the storage node runs a characterization database, the characterization database stores characterization information, and the characterization information is a knowledge characterization system corresponding to the brain cognitive function class, which is established by classifying and characterizing or describing cognitive contents of a physical world or a problem space by adopting a formal description method based on the brain cognitive function structure.
Another technical solution adopted by the present invention to solve the technical problem is to construct a brain-like computing method, including:
acquiring a calculation operation request of a user terminal;
distributing the computing tasks to the computing operation requests, and sending distribution results to a plurality of computing nodes, wherein the computing nodes correspond to a plurality of parallel artificial neural networks, different computing tasks correspond to different computing nodes, different computing nodes correspond to different artificial neural networks, and the distribution results comprise the computing tasks, corresponding relations among the computing nodes and the artificial neural networks, and task computing parameters;
executing parallel computation on the plurality of computing nodes according to the distribution result, acquiring a parallel computation result from the storage node, updating the state of the corresponding computing task according to the parallel computation result, and feeding back the parallel computation result and the state updating result to the management node; the plurality of computing nodes are in communication connection through the constructed full-switching network;
performing distributed storage on the parallel computing result;
the computing nodes are provided with computing support by brain-like coprocessor components supporting artificial neural network modeling, and the brain-like coprocessor components are at least one processor or a combination of a deep learning processor, a neural network processor, a tensor processor and a vector processor; and/or
The brain-like coprocessor component comprises at least one hybrid coprocessor which simultaneously supports the artificial neural network computation;
and/or the brain coprocessor component comprises at least one processor combination and at least one hybrid coprocessor which simultaneously supports the artificial neural network computation.
In the brain-like computing method according to the present invention, the acquiring of the computing job request of the user terminal includes: and compiling and configuring parameters of the operation instruction initiated by the user terminal so as to generate the calculation job request.
According to a third aspect of the present invention, there is provided a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements a brain-like computing method according to the second aspect of the present invention.
According to the brain-like computing system, the brain-like computing method and the computer readable storage medium, the computing job request of the user terminal is obtained, the computing job request is distributed with computing tasks, the distribution result is sent to the plurality of computing nodes, then parallel computing is executed according to the distribution result, and the parallel computing result is stored in a distributed mode, so that the problems that high-density data are difficult to establish a corresponding computing mode and cannot be subjected to effective intelligent analysis are solved, the limited computing power and the rapidly increased data processing are realized, and the requirements of the electronic information industry on continuously improved computing performance and extremely low power consumption are met.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a block diagram of a brain-like computing system according to an embodiment of the present invention;
FIG. 2 is a flowchart of a brain-like computing method according to an embodiment of the present invention;
fig. 3 is a flowchart of another brain calculation method according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Before the technical solutions of the neural network searching method and system provided by the embodiments of the present invention are explained in detail, the technical terms related to the present invention are briefly introduced and explained.
(1) The artificial neural network is a method for realizing artificial intelligence, is inspired by the processing process of a brain neural network system on physical world information, and realizes the fitting approximation of an input/output mode by connecting a neural network formed by a large number of neuron nodes with the same calculation function. Various artificial neural network models have been proposed so far, such as Hopfield neural networks, convolutional neural networks (CNN: constant, functional neural networks), feedback neural networks (RNN: feedback neural networks) deep neural networks (DNN: deep neural networks), deep belief networks (DBN: deep belief neural networks), impulse neural networks (spiking neural networks), and so on. The neural network model running on the computing node in the invention can be one or a combination of several of the above neural networks.
(2) The Hopfield neural network is an important milestone in the development history of neural networks, and is proposed by physicist j.j.hopfield professor of california institute of technology, usa in 1982, and is a single-layer feedback neural network, which is also called self-associative memory network, and aims to design a network, store a set of balance points, so that when a set of initial values are given to the network, the network finally converges to the designed balance points through self-operation, and the feedback neural network is a neural network system which enables output to be accessed to an input layer after one-step time shifting. The feedback neural network can show the dynamic characteristics of a nonlinear dynamical system, and has the following main characteristics:
first, the network system has several stable states. When the network starts to move from a certain initial state, the network system can always converge to a certain stable equilibrium state;
second, the stable equilibrium state of the system can be stored in the network by designing the weight values of the network.
The network is mainly used for associative memory and optimization calculation, and in the network, each neuron simultaneously feeds back an output signal of the neuron as an input signal to other neurons, and the neuron needs to work for a period of time to be stable.
The artificial neural network is the simplest and widely applied model in the feedback neural network, has the function of associative memory, and is also a cyclic neural network, and feedback connection is formed from output to input, and under the excitation of the input, the state change can be generated continuously.
(3) Associative memory, which is to form a memory matrix in a certain way for the complete pattern to be memorized in advance, then memorize them in a computer, and later, when the incomplete pattern is input into the neural network, the complete pattern can be obtained at the output end of the network through the action of the memory matrix and a threshold function.
Associative memory has two prominent features:
first, the access of information (data) is not realized by the address of the memory as in the conventional computer, but by the content of the information itself, so it is "memory by content";
second, the information is not stored centrally in some units, but is stored in a distributed manner, and the memory unit and the processing unit are integrated in the human brain.
It should be noted that the structure or construction and the service scenario described in the embodiment of the present invention are for more clearly illustrating the technical solution of the embodiment of the present invention, and do not limit the technical solution provided in the embodiment of the present invention, and the data described in the embodiment of the present invention may be some continuous values, such as analog data of sound, image, and the like; or may be some discrete values, such as numerical data of symbols, characters, etc., where the type of the data is not specifically limited, and it is known to those skilled in the art that, as the device structure or configuration evolves and new service scenarios occur, the technical solution provided in the embodiment of the present invention is also applicable to similar technical problems.
Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
The method comprises the steps of obtaining a calculation job request of a user terminal through a login node and sending the calculation job request to a management node, then distributing calculation tasks to the calculation job request by the management node, sending distribution results to a plurality of calculation nodes, then executing parallel calculation by the plurality of calculation nodes according to the distribution results, and finally performing distributed storage on the parallel calculation results by a storage node. In the calculation process, distributed user input is adopted for calculation tasks (or operation), parallel calculation is carried out by adopting large-scale brain-like neural network calculation nodes in the calculation process, calculation results are stored by adopting distributed storage nodes, and the storage nodes mainly store representation information and support exponential data access speed; the distributed computing nodes and the storage nodes are interconnected and intercommunicated through a full-switching network, so that the information processing speed of the brain-like neural network system is achieved. The problems that the existing brain-like computing capability is not strong, the brain-like computing is unclear, effective intelligent analysis cannot be performed on high-density data and the like are solved, and powerful computing support is provided for complex scene application such as big data analysis and the like.
As shown in fig. 1, a brain-like computing system according to an embodiment of the present invention includes: a login node 10, a management node 20, a cluster of computer nodes 30 and a storage node 40. The computer node cluster 30 comprises a plurality of computing nodes 31, 32 … 3n, n being a positive integer. The storage node 40 correspondingly includes n storage areas, 41, 42 … 4 n.
The login node 10 is configured to obtain a calculation job request of a user terminal and send the calculation job request to the management node 20. The management node 20 is configured to distribute the computing tasks to the computing job request, and send a distribution result to the plurality of computing nodes 31, 32 … 3 n. The plurality of compute nodes 31, 32 … 3n are used to perform parallel computations based on the distribution results. The storage node 40 is used for performing distributed storage on parallel computing results.
In the embodiment of the present invention, first, when a user logs in a system, the login node 10, as a gateway (including security components such as a firewall and an intrusion detection system) of the entire application system, is a unique entry for external access, and intercepts an operation behavior of a user terminal, compiles and configures parameters of an operation instruction initiated by the user terminal, thereby generating a calculation job request for the user terminal, and sends the calculation job request to the management node 20.
Then, the management node 20 is responsible for monitoring the operation status of each node and the network as a control node of the whole brain-like computing system, implementing corresponding management measures by running corresponding management software, and after receiving a computing job request submitted by the login node 10, the management node 20 will distribute computing tasks to the computing job request and send the distribution result to a plurality of computing nodes (e.g. computing node 31, computing node 32, … … computing node 3 n).
Specifically, the management node 20 divides the computing nodes according to the data in the computing job request and the whole data processing stage, so that different processing stages correspond to different computing nodes, and different computing nodes correspond to different computing tasks (e.g., computing node 31 corresponds to computing task 1, computing node 32 corresponds to computing task 2, and computing node 3n corresponds to computing task n of … …), and sends the allocation result to the multiple computing nodes 31, 32 … 3n in the computing node cluster 30.
It should be noted that, before sending the distribution result to the plurality of computing nodes, the plurality of computing nodes 31, 32 … 3n may be utilized to perform artificial neural network model training, so that each computing node may serve as a neuron in the artificial neural network model, the plurality of computing nodes may correspond to a plurality of parallel artificial neural networks, and since each artificial neural network corresponds to one computing node, in order to enable the artificial neural networks to be connected with each other for performing data interactive transmission, an OVS (english name: Open vSwitch) full-switching network may be constructed among the plurality of computing nodes for performing Communication connection based on a full-Switch Communication protocol FSCP (full-Switch Communication protocol), so that the artificial neural networks may be connected with each other for performing data interactive transmission, where the full-Switch Communication protocol FSCP is a full-loop-free network, The direct two-layer exchange communication transmission protocol which is fully redundant, has no root bridge, low cost and is relatively stable can establish a computing system without center cooperation based on the constructed OVS full-exchange network, realize the elastic expansion of computing nodes, provide strong computing power, support high computing power required by safety guarantee, lay a foundation for the high parallelism of subsequent computing resources and reduce network delay. It can be understood that different computing nodes correspond to different artificial neural networks, and configure corresponding task computing parameters, and then associate each computing task with the corresponding computing node, the artificial neural network corresponding to the computing node, and the task computing parameters configured correspondingly to form a corresponding relationship between the computing task, the computing node, the artificial neural network, and the task computing parameters, and send the corresponding relationship and the task computing parameters as distribution results to the corresponding computing nodes in the whole computing node cluster.
Next, each computing node 31, 32 … 3n finds its corresponding computing task, artificial neural network and task computing parameter according to the above allocation result, and performs computation by using the configured task computing parameter, specifically, performs associative memory of data by using each artificial neural network according to the configured task computing parameter, and at the same time, in order to provide computational support for the whole computation, preferably, in an embodiment of the present invention, the process of performing computation by the whole computing node may be provided by a brain-like coprocessor component supporting artificial neural network modeling, the brain-like coprocessor component may provide computational support for at least one processor combination of a deep learning processor DPU, a neural network processor NPU, a tensor processor TPU, and a vector processor VPU, such as at least one DPU, or at least one NPU, or at least one TPU, or at least one VPU, or the brain coprocessor component may further include at least one hybrid coprocessor which simultaneously supports the artificial neural network computation, such as a DPU and an NPU, or an NPU and a TPU, or a TPU and a DPU, or a VPU and a TPU, and even the brain coprocessor component may further include at least one of the above processor combinations and at least one hybrid coprocessor which simultaneously supports the artificial neural network computation, such as a combination of at least one DPU plus a DPU and an NPU hybrid coprocessor, or a combination of at least one NPU plus an NPU and a TPU hybrid coprocessor, and so on, which are not listed here, under the condition that the above brain coprocessor component provides operational support, the system can rapidly acquire contents of associative memory from each artificial neural network, the association memory content output by each artificial neural network is the calculation result of the corresponding calculation node, and the association memory content obtained from all the artificial neural networks is used as the parallel calculation result of the whole calculation operation request.
Then, the storage node 40 performs distributed storage on the parallel computation results of the multiple computation nodes, specifically, a distributed storage system is built on itself by using a representation database in a manner of deploying a memory Cloud (RAM Cloud), a storage area is allocated to the computation result corresponding to each computation node for storage, and the storage areas obtained by each computation result are not repeated (for example, the storage area 41 allocated to the computation node 31, the storage area 42 allocated to the computation node 32, and the storage area 4n allocated to the computation node 3n by the … …) so as to ensure that subsequent data reading is not mistaken, thereby achieving high-performance computation and storage.
The characterization database stores characterization information which is a knowledge characterization system corresponding to the human brain cognitive function class and is established by classifying, characterizing or describing cognitive contents of a physical world or a problem space by adopting a formal description method based on a human brain cognitive function structure.
In addition, after the storage node 40 performs distributed storage on the results of parallel computation, the computing node may also obtain the results of the parallel computation from the storage node 40, updating the state of the corresponding computing task according to the parallel computing result, for example, after the computing node 1 performs the computation on the computing task 1 to obtain the computing result 1, after the computing node 1 obtains the computing result 1 from the storage area 1, the computation status in the computation task 1 is updated from initialization to completion, and so on, which is not described in detail herein, and then obtains the parallel computation result and the status update result, and feeds back the parallel computation result and the status update result to the management node 20, so that the management node 20 sends the received parallel computation result and the state updating result to the login node 10, and the login node 10 feeds back the user terminal.
An embodiment of the present invention provides a brain-like computing system, including: the login node acquires a calculation job request of a user terminal and sends the calculation job request to the management node, then the management node distributes calculation tasks to the calculation job request and sends a distribution result to the plurality of calculation nodes, then the plurality of calculation nodes execute parallel calculation according to the distribution result, and finally, the storage node performs distributed storage on the parallel calculation result. The problems that the existing brain-like computing capability is not strong, the brain-like computing is unclear, effective intelligent analysis cannot be performed on high-density data and the like are solved, and the limited computing power and the rapidly-increased data processing are realized, so that the requirements of the electronic information industry on continuously-improved computing performance and extremely low power consumption are met.
As shown in fig. 2, a brain-like calculation method provided in an embodiment of the present invention includes the following steps.
The acquiring of the computing job request of the user terminal includes: and compiling and configuring parameters of the operation instruction initiated by the user terminal so as to generate the calculation job request.
In this step, when the user logs in the system, the operation behavior of the user terminal is intercepted, and the operation instruction initiated by the user terminal is compiled and parameter configured, so as to generate a calculation job request for the user terminal.
The method comprises the following steps that a plurality of computing nodes correspond to a plurality of parallel artificial neural networks, different computing tasks correspond to different computing nodes, different computing nodes correspond to different artificial neural networks, and distribution results comprise computing tasks, corresponding relations among the computing nodes and the artificial neural networks and task computing parameters. And, communication among the plurality of computing nodes can be enabled by constructing an OVS (english name: Open vSwitch) full-switching network.
In this step, the computing job request may be assigned computing tasks and the assignment may be sent to multiple computing nodes (e.g., computing node 1, computing node 2, … …, computing node n). Specifically, according to data in the calculation job request, the calculation nodes are divided according to the whole data processing stage, so that different processing stages correspond to different calculation nodes, different calculation nodes correspond to different calculation tasks (for example, calculation node 1 corresponds to calculation task 1, calculation node 2 corresponds to calculation task 2, and … … calculation node n corresponds to calculation task n), and the distribution result is sent to a plurality of calculation nodes in the calculation node cluster.
It should be noted that, before sending the distribution result to the plurality of computing nodes, the plurality of computing nodes may be utilized to perform artificial neural network model training, so that each computing node may serve as a neuron in the artificial neural network model, the plurality of computing nodes may correspond to a plurality of parallel artificial neural networks, and each artificial neural network corresponds to one computing node, in order to enable the artificial neural networks to be connected with each other for performing data interactive transmission, an OVS (english name: Open vSwitch) full-switching network may be constructed among the plurality of computing nodes for performing Communication connection based on a full-Switch Communication protocol FSCP (full-Switch Communication protocol), so that the artificial neural networks may be connected with each other for performing data interactive transmission, where the full-Switch Communication protocol FSCP is a full loop-free, full-redundancy, or a full-redundancy network, A direct two-layer exchange communication transmission protocol which is free of root bridges, low in cost and relatively stable can be established on the basis of a constructed full-exchange network OVS, a computing system without center cooperation can be established, elastic expansion of computing nodes is achieved, strong computing power can be provided, high computing power required by safety guarantee is supported, high parallelism of subsequent computing resources is achieved, and a foundation is laid for reducing network delay. It can be understood that different computing nodes correspond to different artificial neural networks, and configure corresponding task computing parameters, then associate each computing task with the corresponding computing node, the artificial neural network corresponding to the computing node, and the task computing parameters configured correspondingly to form a corresponding relationship among the computing tasks, the computing nodes, the artificial neural networks, and the task computing parameters, and send the corresponding relationship and the task computing parameters as distribution results to the corresponding computing nodes in the whole computing node cluster, it should be noted that the artificial neural network model can be trained by using the existing mechanism.
And 203, executing parallel computation on the plurality of computing nodes according to the distribution result.
In this step, according to the allocation result, each computing node may find its corresponding computing task, artificial neural network and task computing parameter, and perform computation by using the configured task computing parameter, specifically, perform associative memory of data by using each artificial neural network according to the configured task computing parameter, and at the same time, to provide computational support for the whole computation, preferably, in an embodiment of the present invention, the process of performing computation by the whole computing node may be provided by a brain-like coprocessor component supporting artificial neural network modeling, where the brain-like coprocessor component may provide computational support for at least one processor combination of a deep learning processor DPU, a neural network processor NPU, a tensor processor TPU, and a vector processor VPU, such as at least one DPU, or at least one NPU, or at least one TPU, or at least one VPU, or the brain coprocessor component may further include at least one hybrid coprocessor which simultaneously supports the artificial neural network computation, such as a DPU and an NPU, or an NPU and a TPU, or a TPU and a DPU, or a VPU and a TPU, and even the brain coprocessor component may further include at least one processor combination and at least one hybrid coprocessor which simultaneously supports the artificial neural network computation, such as a combination of at least one DPU plus a DPU and an NPU hybrid coprocessor, or a combination of at least one NPU plus an NPU and a TPU hybrid coprocessor, and so on, which are not listed here, under the condition that the brain coprocessor component provides operational support, the system can rapidly acquire contents of associative memory from each artificial neural network, and the association memory content output by each artificial neural network is the calculation result of the corresponding calculation node, and the association memory content acquired from all the artificial neural networks is used as the parallel calculation result of the whole calculation operation request.
And step 204, performing distributed storage on the parallel calculation result.
In combination with step 203, in this step, the parallel computation results of the multiple compute nodes are stored in a distributed manner, specifically, a distributed storage system is built on itself by deploying a memory Cloud (RAM Cloud), a storage area is allocated to the computation result corresponding to each compute node for storage, and the storage areas obtained by each computation result are not repeated (for example, storage area 1 allocated to compute node 1, storage area 2 allocated to compute node 2, and storage area n allocated to compute node n by … …) so as to ensure that subsequent data reading is not mistaken, thereby implementing high-performance computation and storage.
As shown in fig. 3, in addition to steps 201 to 204 shown in fig. 2, another brain calculation method provided for the embodiment of the present invention further includes, after the performing distributed storage on parallel calculation results, the method further includes:
and step 205, updating the state of the corresponding computing task according to the parallel computing result, and feeding back the parallel computing result and the state updating result to the user terminal.
In this step, after the result of the parallel computation is stored in a distributed manner, the result of the parallel computation may be obtained from the storage area, and the state of the corresponding computation task is updated according to the result of the parallel computation, for example, after the computation node 1 performs the computation on the computation task 1 to obtain the computation result 1, and after the computation result 1 is obtained from the storage area 1, the computation state in the computation task 1 is updated from initialization to completion, and so on, which is not described here one by one, so that the result of the parallel computation and the result of the state update are obtained, and the result of the parallel computation and the result of the state update are fed back to the user terminal.
The implementation principle and the generated technical effect of the brain-like computing method provided by the embodiment of the present invention are the same as those of the foregoing method embodiments, and for brief description, reference may be made to corresponding contents in the product embodiment shown in fig. 1, where no part of the corresponding method embodiments is mentioned, and further description is omitted here.
The brain-like calculation method provided by the embodiment of the invention comprises the following steps: the method comprises the steps of obtaining a computing job request of a user terminal, distributing computing tasks to the computing job request, sending distribution results to a plurality of computing nodes, executing parallel computing according to the distribution results, and performing distributed storage on the parallel computing results. The problems that the existing brain-like computing capability is not strong, the brain-like computing is unclear, effective intelligent analysis cannot be performed on high-density data and the like are solved, and the limited computing power and the rapidly-increased data processing are realized, so that the requirements of the electronic information industry on continuously-improved computing performance and extremely low power consumption are met.
In addition, an embodiment of the present invention further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a computer, the computer-readable storage medium executes the brain-like calculation method according to any one of the embodiments shown in fig. 2.
It should be noted that the computer readable medium shown in the embodiments of the present application may be a computer readable signal medium or a computer readable storage medium or any combination of the two. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples of the computer readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a Read-Only Memory (ROM), an Erasable Programmable Read-Only Memory (EPROM), a flash Memory, an optical fiber, a portable Compact Disc Read-Only Memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present application, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In this application, however, a computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the foregoing.
In addition, in the embodiments provided in the present invention, it should be understood that the disclosed apparatus and method can be implemented in other ways. The apparatus embodiments described above are merely illustrative, and for example, the flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, the functional modules in the embodiments of the present invention may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and all the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the appended claims.
Claims (6)
1. A brain-like computing system, comprising: the system comprises a login node, a management node, a plurality of computing nodes and a storage node;
the login node is used for acquiring a calculation job request of a user terminal and sending the calculation job request to the management node;
the management node is used for distributing the computing tasks to the computing operation requests and sending distribution results to the computing nodes, wherein the computing nodes correspond to a plurality of parallel artificial neural networks, different computing tasks correspond to different computing nodes, different computing nodes correspond to different artificial neural networks, and the distribution results comprise the computing tasks, corresponding relations among the computing nodes and the artificial neural networks and task computing parameters;
the plurality of computing nodes are used for executing parallel computing according to the distribution result, acquiring a parallel computing result from the storage node, updating the state of the corresponding computing task according to the parallel computing result, and feeding back the parallel computing result and the state updating result to the management node; the plurality of computing nodes are in communication connection through the constructed full-switching network;
the storage node is used for performing distributed storage on the parallel computing result;
the computing nodes are provided with computing support by brain-like coprocessor components supporting artificial neural network modeling, and the brain-like coprocessor components are at least one processor or a combination of a deep learning processor, a neural network processor, a tensor processor and a vector processor; and/or
The brain-like coprocessor component includes at least one hybrid coprocessor that supports the artificial neural network computations.
2. The brain-like computing system of claim 1, wherein the login node is further configured to compile and parameter configure operational instructions initiated by the user terminal to generate the computing job request.
3. The brain-like computing system according to claim 2, wherein the storage node runs a characterization database, the characterization database stores characterization information, and the characterization information is a knowledge characterization system corresponding to the brain cognitive function class, which is established by classifying, characterizing or describing cognitive contents of the physical world or the problem space by a formal description method based on the brain cognitive function structure.
4. A brain-like computation method, comprising:
acquiring a calculation operation request of a user terminal;
distributing the computing tasks to the computing operation requests, and sending distribution results to a plurality of computing nodes, wherein the computing nodes correspond to a plurality of parallel artificial neural networks, different computing tasks correspond to different computing nodes, different computing nodes correspond to different artificial neural networks, and the distribution results comprise the computing tasks, corresponding relations among the computing nodes and the artificial neural networks, and task computing parameters;
executing parallel computation on the multiple computation nodes according to the distribution result, acquiring a parallel computation result from a storage node, updating the state of the corresponding computation task according to the parallel computation result, and feeding back the parallel computation result and the state update result to a management node; the plurality of computing nodes are in communication connection through the constructed full-switching network;
performing distributed storage on the parallel computing result;
the computing nodes are provided with computing support by brain-like coprocessor components supporting artificial neural network modeling, and the brain-like coprocessor components are at least one processor or a combination of a deep learning processor, a neural network processor, a tensor processor and a vector processor; and/or
The brain-like coprocessor component includes at least one hybrid coprocessor that supports the artificial neural network computations.
5. The brain-like computing method according to claim 4, wherein the acquiring of the computing job request of the user terminal includes: and compiling and configuring parameters of the operation instruction initiated by the user terminal so as to generate the calculation job request.
6. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out a brain-like computing method according to any one of claims 4 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110782692.XA CN113222134B (en) | 2021-07-12 | 2021-07-12 | Brain-like computing system, method and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110782692.XA CN113222134B (en) | 2021-07-12 | 2021-07-12 | Brain-like computing system, method and computer readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113222134A CN113222134A (en) | 2021-08-06 |
| CN113222134B true CN113222134B (en) | 2021-10-26 |
Family
ID=77081289
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110782692.XA Active CN113222134B (en) | 2021-07-12 | 2021-07-12 | Brain-like computing system, method and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113222134B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113792649A (en) * | 2021-09-13 | 2021-12-14 | 广州广电运通金融电子股份有限公司 | Rapid authentication method, device and medium based on finger vein biological identification technology |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108171117A (en) * | 2017-12-05 | 2018-06-15 | 南京南瑞信息通信科技有限公司 | Electric power artificial intelligence visual analysis system based on multinuclear heterogeneous Computing |
| CN108446764A (en) * | 2018-03-12 | 2018-08-24 | 清华大学 | A kind of new type nerve form chip architecture |
| CN112364988A (en) * | 2020-11-12 | 2021-02-12 | 天津大学 | Hierarchical heterogeneous brain computing system based on FPGA |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150206051A1 (en) * | 2012-08-02 | 2015-07-23 | Max-Planck-Gesellschaft zur Föderung der Wissenschaften E.V. | Method and computing system for modelling a primate brain |
| CN103473111A (en) * | 2013-08-16 | 2013-12-25 | 运软网络科技(上海)有限公司 | Brain-imitation calculation virtualization method and brain-imitation calculation virtualization system |
| CN109754060B (en) * | 2017-11-06 | 2023-08-25 | 阿里巴巴集团控股有限公司 | Training method and device for neural network machine learning model |
| CN109858620B (en) * | 2018-12-29 | 2021-08-20 | 北京灵汐科技有限公司 | Brain-like computing system |
| CN109901878B (en) * | 2019-02-25 | 2021-07-23 | 北京灵汐科技有限公司 | Brain-like computing chip and computing equipment |
| CN110322010B (en) * | 2019-07-02 | 2021-06-25 | 深圳忆海原识科技有限公司 | Pulse neural network operation system and method for brain-like intelligence and cognitive computation |
| CN112511592B (en) * | 2020-11-03 | 2022-07-29 | 深圳市中博科创信息技术有限公司 | Edge artificial intelligence computing method, Internet of things node and storage medium |
-
2021
- 2021-07-12 CN CN202110782692.XA patent/CN113222134B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108171117A (en) * | 2017-12-05 | 2018-06-15 | 南京南瑞信息通信科技有限公司 | Electric power artificial intelligence visual analysis system based on multinuclear heterogeneous Computing |
| CN108446764A (en) * | 2018-03-12 | 2018-08-24 | 清华大学 | A kind of new type nerve form chip architecture |
| CN112364988A (en) * | 2020-11-12 | 2021-02-12 | 天津大学 | Hierarchical heterogeneous brain computing system based on FPGA |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113222134A (en) | 2021-08-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gholami et al. | Integrated model, batch, and domain parallelism in training neural networks | |
| CN106953862B (en) | Sensing method and device for network security situation and sensing model training method and device | |
| Zhang et al. | Efficient federated learning for cloud-based AIoT applications | |
| Grzonka et al. | Artificial neural network support to monitoring of the evolutionary driven security aware scheduling in computational distributed environments | |
| Aouiti et al. | Fixed-time synchronization of competitive neural networks with proportional delays and impulsive effect | |
| Shadroo et al. | The two-phase scheduling based on deep learning in the Internet of Things | |
| Furber | Brain‐inspired computing | |
| Zhang et al. | DiGA: Population diversity handling genetic algorithm for QoS-aware web services selection | |
| Chaterji et al. | Resilient cyberphysical systems and their application drivers: A technology roadmap | |
| CN113222134B (en) | Brain-like computing system, method and computer readable storage medium | |
| Sarathambekai et al. | Task scheduling in distributed systems using heap intelligent discrete particle swarm optimization | |
| CN113240100B (en) | Parallel computing method and system based on discrete Hopfield neural network | |
| Saguil et al. | A layer-partitioning approach for faster execution of neural network-based embedded applications in edge networks | |
| Qin et al. | Reinforcement-learning-empowered MLaaS scheduling for serving intelligent internet of things | |
| Gand et al. | A fuzzy controller for self-adaptive lightweight edge container orchestration | |
| Fiorini | A cybernetics update for competitive deep learning system | |
| Carlson et al. | GPGPU accelerated simulation and parameter tuning for neuromorphic applications | |
| Rossant et al. | Playdoh: a lightweight Python library for distributed computing and optimisation | |
| CN110012021B (en) | Self-adaptive computing migration method under mobile edge computing | |
| Huang et al. | An efficient parallel method for batched OS-ELM training using MapReduce | |
| Alhafidh et al. | FPGA hardware implementation of smart home autonomous system based on deep learning | |
| CN113515341A (en) | Flexible distributed AI training cloud platform deployment method and related platform | |
| Kinneer et al. | Building reusable repertoires for stochastic self-* planners | |
| Plotnikova et al. | Gravitation search training algorithm for asynchronous distributed multilayer perceptron model | |
| CN110750591A (en) | Artificial intelligence knowledge management system and method for forming distributed records by using same |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |