CN112149788A - Minimum overhead route generation method based on ant colony algorithm - Google Patents
Minimum overhead route generation method based on ant colony algorithm Download PDFInfo
- Publication number
- CN112149788A CN112149788A CN202011046528.4A CN202011046528A CN112149788A CN 112149788 A CN112149788 A CN 112149788A CN 202011046528 A CN202011046528 A CN 202011046528A CN 112149788 A CN112149788 A CN 112149788A
- Authority
- CN
- China
- Prior art keywords
- chip
- ant
- ant colony
- path
- colony algorithm
- 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.)
- Granted
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
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention discloses a minimum overhead route generation method based on an ant colony algorithm, which relates to the technical field of routing. The method can minimize the total length of the global routing path, thereby further minimizing the global communication load overhead.
Description
Technical Field
The invention relates to the technical field of class routing, in particular to a minimum overhead route generation method based on an ant colony algorithm.
Background
Brain-like computing is a cross research direction of multiple disciplines such as computer science, biology, information science and the like and across fields. The method aims to seek important breakthrough in many aspects such as brain-like artificial intelligence algorithm, brain-like intelligent chip, brain disease intelligent diagnosis and treatment and the like. Brain-like calculations are useful in helping to understand and understand the neural mechanisms of higher cognitive functions, understand and learn brain plasticity, help to analyze biological cerebral cortical function and neurotransmitter system regulation. And the breakthrough of intelligent algorithms based on new brain-like intelligence theory, new algorithms and new frameworks can be brought, and new generation of intelligence is created.
The impulse neural network (SNN) is called a third-generation artificial neural network, which simulates the behavior of neurons closer to actual neurons, and makes up the problem of insufficient information of the existing neural network on a time scale. The impulse neural network uses the biological nervous system for reference, adopts the impulse time sequence of the neuron to encode information, and not only uses the issuing frequency, so that the information is richer. The neurons are connected through synapses, and a certain synaptic plasticity learning algorithm is used, so that the network can realize certain judgment and reasoning functions to complete an expected task. Meanwhile, the signals of the pulse neural network are more sparse, and the information utilization rate can be higher than that of the existing neural network. The pulse neural network is realized through a certain neuron analog circuit, and compared with the existing neural network, the pulse neural network has great advantages in power consumption.
The neural mimicry calculation is honored as one of three race tracks leading into the future of artificial intelligence. It also implements complex computational networks by hardware simulating the computation of neurons. Unlike impulse neural networks, neuromorphic computations provide a high-performance, super-heterogeneous computing system framework without requiring specific and detailed functionality to be implemented. The neural mimicry computation is based on time-driven, mutually-correlated signals allowing input of multiple modalities, enabling fusion of multiple computing mechanisms. The development of the neural mimicry calculation is expected to bring a new super computer system, and the method is higher in efficiency, higher in calculation performance and higher in calculation complexity.
Large-scale brain-like computing networks rely on enormous computational support. The existing CPU and GPU do not support event-driven computational modes, exhibit redundancy in architecture, and cause additional overhead in power consumption. So customizing FPGA or ASIC chips is a better brain-like computing network solution. The method is specially designed for analog neurons, removes the part which is not beneficial to brain-like calculation in the traditional computer system structure, and eliminates excessive hardware power consumption overhead. Meanwhile, the calculation unit is highly matched with the neuron model, so that the calculation efficiency during neuron behavior simulation is greatly improved. The storage unit is also designed to be closer to the computing unit, so that the bandwidth of data access is increased, and the influence of bottleneck of a memory wall is reduced. The event-driven-based hardware system architecture can be matched with the signal transmission characteristic of the brain-like computing network sparseness, reduces unnecessary hardware operation, and is beneficial to solving the problem of ultrahigh energy consumption in large-scale computing faced by a traditional computing framework.
The large-scale brain-like computing network can support a large amount of pulse neural networks and neural mimicry computing, and provides ultrahigh computing power which is not possessed by a traditional computing platform for the large-scale brain-like computing network. However, in a large-scale brain-like computing network, the traditional routing method generates a lengthy routing path, which brings huge overall communication overhead of the system and affects communication quality between chips, so that a routing method capable of supporting the large-scale brain-like computing network needs to be developed.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a minimum overhead route generation method based on an ant colony algorithm, which heuristically minimizes the total path length in a single multicast route by using the ant colony algorithm according to the source and the destination of each multicast route and calculates a corresponding fixed route table aiming at a large-scale brain-like computing network with a two-dimensional Mesh network or a two-dimensional Torus network as a networking network, thereby realizing that the global communication load approximately reaches the minimum. The method can minimize the total length of the global routing path, thereby further minimizing the global communication load overhead.
In order to achieve the above object, the present invention provides the following technical solutions:
a minimum overhead route generation method based on an ant colony algorithm comprises at least one neuron, wherein each neuron corresponds to a single multicast route, a source chip and a destination chip are determined in the single multicast route, and at least one optional chip for forming a path is arranged between the source chip and the destination chip; in the single multicast routing, initializing ant colony algorithm parameters, customizing ant colony scale and repetition times:
step (1) releasing a first ant with the source chip as a starting point, wherein the first ant accesses a path with the probability of each optional chip being selected as a reference, and generates an access path after the destination chip is accessed, and updates the probability of each optional chip being selected;
step (2) releasing a second ant from the source chip, and operating according to the step (1) until all ants corresponding to the ant colony scale complete access;
step (3) iterative computation is carried out on the steps (1) to (2) by the user-defined repetition times, and the global load lowest access path is taken as a single multicast routing path;
and (4) traversing the steps (1) to (3) by the number of the neurons until multicast routes of all the neurons are generated.
Preferably, the ant colony algorithm parameters include: the number of repetitions K, the ant colony scale m, the pheromone concentration Q, the volatilization speed rho, the pheromone elicitation factor alpha and the distance elicitation factor beta;
preferably, the probability of the selectable chip being selected is p (i), and the formula is as follows:
p(i)＝q(i)α×d(i)-β
wherein q (i) is the accumulated pheromone concentration of the ith chip, and d (i) is the distance between the ith chip and the current position of the ant; the pheromone concentration q (i) is as follows:
q(i)＝(1-ρ)×q(i)+Q/dsum
wherein d issumΣ d (i) is the sum of the path distances traveled by the ant.
Advantageous effects
The invention provides a minimum overhead route generation method based on an ant colony algorithm, which aims at a large-scale brain-like calculation network with a two-dimensional Mesh network or a two-dimensional Torus network as a networking network, heuristically minimizes the total path length in a single multicast route by using the ant colony algorithm according to the source and the destination of each multicast route, and calculates a corresponding fixed route table, thereby realizing that the global communication load is approximately minimized. The method can minimize the total length of the global routing path, thereby further minimizing the global communication load overhead.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the structures shown in the drawings without creative efforts.
FIG. 1 is a flowchart of a method for generating a minimum overhead route based on an ant colony algorithm according to the present invention;
fig. 2 is a comparison diagram of the ant colony algorithm-based minimum overhead route generation method and other routing methods according to the present invention.
Detailed Description
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.
The invention provides a minimum overhead route generation and release method based on an ant colony algorithm, aiming at minimizing the global load in a two-dimensional Mesh or two-dimensional Torus interconnection network. Unlike the shortest path approach, the minimum cost approach does not pursue the shortest path between destination to source, but rather a greater degree of multiplexing and merging through the paths, minimizing the total path cost in single multicast routing. In the multiplexing process, a part of paths longer than the shortest path are even introduced, and the global overhead is optimized at the expense of the delay of a single path. Under certain application scenarios of time delay redundancy (such as extremely short calculation process and abundant communication time in a single iteration step of a brain-like calculation network), the global communication load can be significantly reduced, the communication stability is improved, and the energy consumption in the communication process is further optimized by using the minimum overhead route generation method.
As shown in fig. 1, a method for generating a minimum overhead route based on an ant colony algorithm includes at least one neuron, each neuron corresponds to a single multicast route, in which a source chip and a destination chip are determined, and at least one selectable chip for forming a path is included between the source chip and the destination chip; in the single multicast routing, initializing ant colony algorithm parameters, customizing ant colony scale and repetition times:
step (1) releasing a first ant with the source chip as a starting point, wherein the first ant accesses a path with the probability of each optional chip being selected as a reference, and generates an access path after the destination chip is accessed, and updates the probability of each optional chip being selected;
step (2) releasing a second ant from the source chip, and operating according to the step (1) until all ants corresponding to the ant colony scale complete access;
step (3) iterative computation is carried out on the steps (1) to (2) by the user-defined repetition times, and the global load lowest access path is taken as a single multicast routing path;
and (4) traversing the steps (1) to (3) by the number of the neurons until multicast routes of all the neurons are generated.
The ant colony algorithm parameters include: the number of repetitions K, the ant colony scale m, the pheromone concentration Q, the volatilization speed rho, the pheromone elicitation factor alpha and the distance elicitation factor beta; the parameters have a better empirical value range, but specific values under different problems need to be obtained by adjusting the parameters and probing.
Specifically, the repetition times K and the ant colony scale m are related to the times of the algorithm for carrying out the loop iteration, and the complexity of the algorithm is further influenced; the larger K is helpful for the algorithm to jump out of the local optimal solution; a larger m favors the ant colony approaching a shorter routing path in a single action, but also makes the algorithm more time consuming.
Pheromone concentration Q and volatilization speed rho influence the residue of pheromones on each path, the residual pheromones are more, the explorable paths of the ant colony are wider, and the algorithm is more difficult to converge; fewer residual pheromones will allow the ant colony to eliminate invalid paths earlier, but at the same time there is also a probability that valid paths will be discarded.
The pheromone heuristic factor alpha and the distance heuristic factor beta influence the decision-making behavior of the ant colony, and the larger pheromone heuristic factor alpha makes the ant colony more prone to selecting a previously-traveled path, so that the search randomness is weakened; a larger distance heuristic β will make it easier for the ant colony to select a locally shorter path and not easily jump out to the global optimal solution.
Thus, ant colony behavior is largely divided into two processes: path selection and pheromone updating.
In each iteration process of the algorithm, m ants are released in sequence, wherein each ant starts from a source chip, a path to be taken is selected according to pheromones and distances, the path selection process is carried out, for each optional chip, the probability of selection is p (i), and the formula is as follows:
p(i)＝q(i)α×d(i)-β
wherein q (i) is the accumulated pheromone concentration of the ith chip, and d (i) is the distance (block distance) between the ith chip and the current position of the ant.
For all the selectable chips, ants randomly select a path according to the size of the probability p (i), wherein the greater the probability that the chip is selected is, after all the destination chips are visited by the ant, a complete path selection process is considered to be completed, and then pheromones are updated. The pheromone update process follows the following equation:
q(i)＝(1-ρ)×q(i)+Q/dsum
wherein d issumΣ d (i) is the sum of the path distances traveled by the ant.
And after the pheromone is updated, the first ant completes the task, then the second ant is released, the path selection is carried out according to the pheromone left by the first ant, and the pheromone is updated again, and the process is circulated until the release of the m ants is completed.
In order to avoid trapping into local optima, the ant colony algorithm needs to repeat the process for K times, empty pheromones on all paths each time and release m ants again, namely, the steps are repeatedly executed. And recording the global load of each time in the K times of repetition, and obtaining the lowest global load in all the processes as an optimal solution after the K times of repetition is finished, wherein the optimal solution is regarded as a solution of the ant colony algorithm, namely the global load minimized routing path for the large-scale brain-like computing network obtained by solving the ant colony algorithm.
If N neurons are set in the route of the large-scale brain-like computing network, the multicast route needs to be traversed for N times until the multicast route of all the neurons is generated.
The invention mainly provides a large-scale brain-like computing network routing method based on an ant colony algorithm, which is superior to the existing routing generation method in the total length of a global routing path and brings lower system communication overhead. As shown in fig. 2, where S denotes a source chip in the routing process, D denotes a destination chip in the routing process, and each diagram shows a multicast routing path:
101 is a DOR route generation method, which is a shortest path route method;
102, the RTO route generation method improves the load unevenness problem compared with the DOR, but the global load (route path length) has higher cost;
103, the method for generating the minimum cost fixed route based on the ant colony algorithm of the present invention, where the path is not necessarily the shortest path, but the global load is significantly reduced compared with other methods, is a route generation method with optimal global load.
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.
Claims (3)
1. A minimum overhead route generation method based on an ant colony algorithm comprises at least one neuron, wherein each neuron corresponds to a single multicast route, a source chip and a destination chip are determined in the single multicast route, and at least one optional chip for forming a path is arranged between the source chip and the destination chip; the method is characterized in that: in the single multicast routing, initializing ant colony algorithm parameters, customizing ant colony scale and repetition times:
step (1) releasing a first ant with the source chip as a starting point, wherein the first ant accesses a path with the probability of each optional chip being selected as a reference, and generates an access path after the destination chip is accessed, and updates the probability of each optional chip being selected;
step (2) releasing a second ant from the source chip, and operating according to the step (1) until all ants corresponding to the ant colony scale complete access;
step (3) iterative computation is carried out on the steps (1) to (2) by the user-defined repetition times, and the global load lowest access path is taken as a single multicast routing path;
and (4) traversing the steps (1) to (3) by the number of the neurons until multicast routes of all the neurons are generated.
2. The method for generating a minimum overhead route based on the ant colony algorithm according to claim 1, wherein: the ant colony algorithm parameters include: the number of repetitions K, the ant colony size m, the pheromone concentration Q, the volatilization speed rho, the pheromone elicitation factor alpha and the distance elicitation factor beta.
3. The method for generating a minimum overhead route based on the ant colony algorithm according to claim 2, wherein: the probability of the selectable chip being selected is p (i), and the formula is as follows:
p(i)＝q(i)α×d(i)-β
wherein q (i) is the accumulated pheromone concentration of the ith chip, and d (i) is the distance between the ith chip and the current position of the ant; the pheromone concentration q (i) is as follows:
q(i)＝(1-ρ)×q(i)+Q/dsum
wherein d issumΣ d (i) is the sum of the path distances traveled by the ant.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011046528.4A CN112149788B (en) | 2020-09-28 | 2020-09-28 | Minimum overhead route generation method based on ant colony algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011046528.4A CN112149788B (en) | 2020-09-28 | 2020-09-28 | Minimum overhead route generation method based on ant colony algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112149788A true CN112149788A (en) | 2020-12-29 |
| CN112149788B CN112149788B (en) | 2023-04-07 |
Family
ID=73896055
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011046528.4A Active CN112149788B (en) | 2020-09-28 | 2020-09-28 | Minimum overhead route generation method based on ant colony algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112149788B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1738291A (en) * | 2005-08-26 | 2006-02-22 | 电子科技大学 | Ad hot network subsequent multi-path route method based on load balance |
| CN101043444A (en) * | 2007-04-26 | 2007-09-26 | 浙江大学 | Distributed quality of service multicast routing process based on ant group optimization |
| EP2159974A1 (en) * | 2008-08-27 | 2010-03-03 | Alcatel, Lucent | Method and equipment for establishing a connection |
| CN104092615A (en) * | 2014-06-10 | 2014-10-08 | 西安电子科技大学 | Network on chip with network coding function, network topology of the network on chip, and route algorithm of the network topology |
| CN104901892A (en) * | 2015-06-03 | 2015-09-09 | 安徽理工大学 | QoS multicasting route optimizer based on ant colony algorithm and realization method of QoS multicasting route optimizer |
| CN107094115A (en) * | 2017-05-19 | 2017-08-25 | 重庆邮电大学 | A kind of ant group optimization Load Balance Routing Algorithms based on SDN |
| CN111310906A (en) * | 2018-12-11 | 2020-06-19 | 清华大学 | Method, device and equipment for calculating layout of cores in neuromorphic chip |
-
2020
- 2020-09-28 CN CN202011046528.4A patent/CN112149788B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1738291A (en) * | 2005-08-26 | 2006-02-22 | 电子科技大学 | Ad hot network subsequent multi-path route method based on load balance |
| CN101043444A (en) * | 2007-04-26 | 2007-09-26 | 浙江大学 | Distributed quality of service multicast routing process based on ant group optimization |
| EP2159974A1 (en) * | 2008-08-27 | 2010-03-03 | Alcatel, Lucent | Method and equipment for establishing a connection |
| CN104092615A (en) * | 2014-06-10 | 2014-10-08 | 西安电子科技大学 | Network on chip with network coding function, network topology of the network on chip, and route algorithm of the network topology |
| CN104901892A (en) * | 2015-06-03 | 2015-09-09 | 安徽理工大学 | QoS multicasting route optimizer based on ant colony algorithm and realization method of QoS multicasting route optimizer |
| CN107094115A (en) * | 2017-05-19 | 2017-08-25 | 重庆邮电大学 | A kind of ant group optimization Load Balance Routing Algorithms based on SDN |
| CN111310906A (en) * | 2018-12-11 | 2020-06-19 | 清华大学 | Method, device and equipment for calculating layout of cores in neuromorphic chip |
Non-Patent Citations (2)
| Title |
|---|
| TAMANNA ARORA ET AL: "Using Ant Colony Optimization for Routing in VLSI Chips", 《AIP CONFERENCE PROCEEDINGS》 * |
| 李冕: "脉冲神经网络模拟多核计算机的互连网络的研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112149788B (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9697462B1 (en) | Synaptic time multiplexing | |
| Saadat et al. | Training echo state neural network using harmony search algorithm | |
| CN111783320B (en) | Brain processor-like NoC design space exploration method, system and medium for specified applications | |
| Folli et al. | Effect of dilution in asymmetric recurrent neural networks | |
| Fang et al. | Scalable noc-based neuromorphic hardware learning and inference | |
| Ghanem et al. | Using hybrid artificial bee colony algorithm and particle swarm optimization for training feed-forward neural networks. | |
| Shen et al. | Esl-snns: An evolutionary structure learning strategy for spiking neural networks | |
| Wang et al. | A genetic timing scheduling model for urban traffic signal control | |
| Mak et al. | On the improvement of the real time recurrent learning algorithm for recurrent neural networks | |
| CN114205251B (en) | Switch link resource prediction method based on space-time characteristics | |
| Xu et al. | Living with artificial intelligence: A paradigm shift toward future network traffic control | |
| Irmak et al. | Training of the feed-forward artificial neural networks using butterfly optimization algorithm | |
| CN107169561B (en) | Power consumption-oriented hybrid particle swarm pulse neural network mapping method | |
| CN112149788B (en) | Minimum overhead route generation method based on ant colony algorithm | |
| Zhang et al. | Brain-inspired experience reinforcement model for bin packing in varying environments | |
| Qu et al. | A new algorithm for finding the shortest paths using PCNNs | |
| CN115470889A (en) | Network-on-chip autonomous optimal mapping exploration system and method based on reinforcement learning | |
| Wu et al. | An algorithm for solving travelling salesman problem based on improved particle swarm optimisation and dynamic step Hopfield network | |
| CN115150335A (en) | Optimal flow segmentation method and system based on deep reinforcement learning | |
| Morales | Deep Reinforcement Learning | |
| Clarkson et al. | From wetware to hardware: reverse engineering using probabilistic RAMs | |
| Shokouhifar et al. | Feature selection using supervised fuzzy C-means algorithm with ant colony optimization | |
| KR20210157826A (en) | Method for sturcture learning and model compression for deep neural netwrok | |
| You et al. | An ABC algorithm with recombination | |
| El-Mahalawy et al. | Applying Hierarchal Clusters on Deep Reinforcement Learning Controlled Traffic Network |
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 |