Graph coloring problem solving optimization method based on hardware memristor coupling neural network

By building a hardware memristor coupled neural network, using the output phase characteristics of the coupling resistor and capacitor, the undirected graph is mapped to the neural network, and the graph coloring results are directly obtained, solving the problem that the output characteristics are not related to the coloring results in the existing technology, and realizing the problem of the output characteristics being unrelated to the coloring results. Fast and stable convergence and simplified solution process.

CN119067165APending Publication Date: 2024-12-03HANGZHOU DIANZI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411173142.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The existing memristor neuron coupling scheme has no direct relationship with the shading results in the graph shading problem, and requires an additional transformation process, resulting in a complex solution process.

Method used

By building a hardware memristor-coupled neural network, using the output phase characteristics of the coupling resistor RC and the coupling capacitor CC, the undirected graph is mapped to the memristor-coupled neural network, and the parameters are set to automatically form clusters of neuron output phases, which can directly obtain Graph coloring results.

Benefits of technology

A fast and stable convergence graph coloring solution is realized, the solution process is simplified, the output phase automatically forms clustering, and the graph coloring results are directly obtained, and the optimal solution is obtained through the multi-input delay strategy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119067165A_ABST
    Figure CN119067165A_ABST
Patent Text Reader

Abstract

The invention discloses a graph coloring problem solving optimization method based on a hardware memristor coupling neural network, and the method comprises the following steps: obtaining an undirected graph of a graph coloring problem to be solved, and calculating the degree of each vertex and the maximum degree in all vertexes according to the connection relation of each vertex of the undirected graph; mapping the undirected graph to the memristor coupling neural network according to a mapping rule specified by the output phase characteristics of the coupling resistor RC and the coupling capacitor CC so as to construct a hardware memristor coupling neural network; setting parameters of the memristor coupling neural network according to the degree of each vertex obtained in the step 1 and the maximum degree in all vertexes, wherein the parameters comprise coupling resistance RC, coupling capacitance CC, neuron capacitance C and compensation capacitance C '; and inputting a voltage waveform, outputting a current waveform and a final coloring result by applying the memristor coupling neural network with the set parameters. The method can be used for solving a graph coloring problem and shows a relatively high solving speed and a visual network result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of hardware memristor neural networks, and in particular to a method for solving and optimizing graph coloring problems based on hardware memristor coupled neural networks. Background Art

[0002] In today's rapidly developing digital age, interdisciplinary research in neuroscience and computer science is leading scientific and technological innovation. Neuroscience reveals the operating principles of biological nervous systems, while computer science designs systems that can simulate, learn and perform complex tasks, driving technological progress. The combination of the two has deepened the understanding of the nature of human intelligence and provided a new thinking framework and technical means to solve complex problems. The remarkable learning and adaptability of biological nervous systems has inspired the development of artificial neural networks, which have achieved remarkable results in areas such as image recognition, speech processing and natural language processing. However, the traditional von Neumann architecture has exposed problems such as communication bottlenecks and transistor process limitations in the post-Moore era when dealing with complex and diverse computing needs. New computing architectures and new electronic devices are urgently needed to meet these challenges.

[0003] Memristors are considered to be strong candidates for new computing systems in the future. New computing architectures based on memristors can break through the limitations of the von Neumann architecture and show great potential in replacing traditional von Neumann architecture computing platforms. Memristor-based neurons also show more and more neuromorphic behaviors, showing broad application prospects in the field of neuromorphic computing, but the role of a single memristor neuron still has certain limitations, and it is necessary to explore how memristor neurons can play a key role in solving specific problems. If multiple memristor neurons are coupled, each neuron may show different behaviors, which can well simulate the information exchange and transmission between neurons. Using specific diverse behaviors, memristor coupled neural networks can be constructed to solve specific problems. For example, in the graph coloring problem in solution optimization, the existing memristor neuron coupling scheme has technical defects, and the output characteristics or forms of the memristor neurons have no direct relationship with the coloring results. It is necessary to use a specific method to convert the output into the coloring result. Therefore, a better solution method is needed, which does not require an additional conversion process and outputs the coloring result intuitively, thereby simplifying the solution process. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention proposes a method for solving the graph coloring problem based on a hardware memristor coupled neural network, which can be used to solve the graph coloring problem and exhibits a faster solution speed and intuitive network results.

[0005] In order to solve the above technical problems, the technical solution of the present invention is:

[0006] A method for solving and optimizing a graph coloring problem based on a hardware memristor coupled neural network comprises the following steps:

[0007] Step 1: Obtain an undirected graph of the graph coloring problem to be solved, and calculate the degree of each vertex and the maximum degree of all vertices according to the connection relationship of each vertex in the undirected graph;

[0008] Step 2: According to the coupling resistance R C and coupling capacitor C C The mapping rule specified by the output phase characteristic of the undirected graph is mapped to the memristor coupled neural network, thereby constructing a hardware memristor coupled neural network, wherein the hardware memristor coupled neural network includes NbO x Memristor neurons, coupling resistor R C and coupling capacitor C C , each NbO x Memristive neurons are similar to all other NbO x Memristive neurons are coupled to each other through a resistor R C Or coupling capacitor C C To couple;

[0009] Step 3: Set the parameters of the memristor coupled neural network according to the degree of each vertex obtained in step 1 and the maximum degree of all vertices, wherein the parameters include the coupling resistor R C , coupling capacitor C C , neuron capacitance C and compensation capacitance C′, where coupling capacitance C C Fixed to 1nF;

[0010] Step 4: Apply the memristor coupled neural network with set parameters, input voltage waveform, output current waveform and final coloring result.

[0011] Preferably, the coupling resistor R C The output phase characteristic of NbO is: x Memristive neurons are coupled to each other through a resistor R C When coupled, the neuron outputs are in phase and synchronized.

[0012] Preferably, the coupling capacitor C C The output phase characteristic of NbO is: x Memristive neurons are coupled via capacitors C C When coupled, the neuron outputs are in anti-phase synchronization.

[0013] Preferably, the mapping rule of the undirected graph to the neural network is: the vertex of the undirected graph is mapped to NbO x Memristor neurons, connected vertices are mapped to NbO x Memristive neurons are coupled via capacitance, and unconnected vertices are mapped to NbO xMemristive neurons are coupled via resistors; NbO x The nodes for coupling between memristor neurons are NbO x Output node of the memristor.

[0014] Preferably, the NbO x Memristive neurons include NbO x A memristor, a resistor, a capacitor and a voltage source; the NbO x The memristor is connected in parallel with the capacitor, and one end of it is connected to the resistor and voltage source. x The other end of the memristor and the capacitor is grounded.

[0015] Preferably, the NbO x The model expression of the memristor is:

[0016]

[0017] where i m and v m Respectively represent the flow through NbO x Memristor current and NbO x The voltage across the memristor, T is NbO x State variables of the memristor, R0, α0, α1, α2, C th , Γ th , T amb is a constant.

[0018] Preferably, the parameter setting method of the memristor coupled neural network is:

[0019] R C =(Nd max )×R Δ

[0020] C=C0+(N-3)C Δ

[0021] C′ i =(d max -d i )·(C C ||C)

[0022] Among them, C i ′ is the compensation capacitor that needs to be connected in parallel across the memristor in the memristor neuron; d i is the degree of each vertex in the undirected graph, d max Then it is the largest degree among all vertices; C C , R C are coupling capacitance and coupling resistance respectively; C is the capacitance in the memristor neuron; N is the number of vertices in the undirected graph, R Δ , C0 and CΔ is a constant.

[0023] The present invention has the following characteristics and beneficial effects:

[0024] By adopting the above technical solution, a complete hardware memristor neural network can converge quickly and stably, and the output phase of each memristor neuron automatically forms a cluster. The undirected graph vertices corresponding to the memristor neurons whose output phases are in the same cluster are colored with the same color, and finally a coloring scheme for the graph coloring problem is formed. Finally, the present invention also formulates a multi-input delay strategy for improving the network to obtain the optimal solution to the graph coloring problem, that is, randomly selecting multiple groups of memristor neuron voltage input times, multiple different coloring schemes can be obtained, and the result using the least number of colors is selected as the result of the network solving the graph coloring problem.

[0025] In order to make memristive neurons play a key role, the present invention will be based on NbO x The memristor's memristor neurons are coupled with resistors or capacitors, and the memristor neurons can show the same-phase synchronization and anti-phase synchronization of neuron output under resistor coupling and capacitor coupling respectively. This characteristic is used to construct a fully connected hardware memristor coupled neural network, which can be used to solve graph coloring problems and show a faster solution speed and intuitive network results.

[0026] By coupling multiple memristor neurons, each neuron may exhibit different behaviors, which can well simulate the information exchange and transmission between neurons. Such diverse behaviors can be used to construct memristor coupled neural networks to solve specific problems, such as solving graph coloring problems in optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0028] Figure 1 The basic structure and mapping method of the memristor coupled neural network in the present invention;

[0029] Figure 2 A schematic diagram of an undirected graph including eight vertices and a method for connecting parallel compensation capacitors when performing network mapping provided in this embodiment, wherein a solid line in the undirected graph indicates that there is a connection relationship between the vertices, and a dotted line indicates that there is no connection relationship between the vertices;

[0030] Figure 3 Schematic diagram of the circuit for memristive neurons coupled by resistors or capacitors;

[0031] Figure 4 is an input voltage waveform of a network neuron obtained by a multi-input delay strategy in one embodiment;

[0032] Figure 5 This is the final coloring result of the output waveform of all neurons and the undirected graph after the network stabilizes. DETAILED DESCRIPTION

[0033] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0034] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings.

[0035] On the contrary, the present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention as defined by the claims. Further, in order to make the public have a better understanding of the present invention, some specific details are described in detail in the detailed description of the present invention below. Those skilled in the art can fully understand the present invention without the description of these details.

[0036] The present invention provides a method for solving and optimizing a graph coloring problem based on a hardware memristor coupled neural network, comprising the following steps:

[0037] Step 1: Get the undirected graph of the graph coloring problem to be solved, and calculate the degree of each vertex and the maximum degree of all vertices based on the connection relationship of each vertex in the undirected graph.

[0038] This embodiment provides an undirected graph for a graph coloring problem to be solved, such as Figure 2 As shown, the undirected graph contains 8 vertices. The degree d of each vertex is calculated based on the connection relationship of each vertex in the undirected graph. i With the maximum degree d max :

[0039] d1=d2=d3=d4=d6=d7=3

[0040] d5=d8=2

[0041] d max =3

[0042] Step 2: According to the coupling resistance R C and coupling capacitor C C The mapping rule specified by the output phase characteristic of the undirected graph is mapped to the memristor coupled neural network, thereby constructing a hardware memristor coupled neural network, wherein the hardware memristor coupled neural network includes NbO x Memristor neurons, coupling resistor RC and coupling capacitor C C , each NbO x Memristive neurons are similar to all other NbO x Memristive neurons are coupled to each other through a resistor R C Or coupling capacitor C C To couple.

[0043] Furthermore, the coupling resistor R C The output phase characteristic of NbO is: x Memristive neurons are coupled to each other through a resistor R C When coupled, the neuron output is in phase synchronization. C The output phase characteristic of NbO is: x Memristive neurons are coupled via capacitors C C When coupled, the neuron outputs are in anti-phase synchronization.

[0044] In a further configuration of this embodiment, the mapping rule of the undirected graph to the neural network is as follows: Figure 1 As shown: The vertex mapping of the undirected graph is NbO x Memristor neurons, connected vertices are mapped to NbO x Memristive neurons are coupled via capacitance, and unconnected vertices are mapped to NbO x Memristive neurons are coupled via resistors; NbO x The nodes for coupling between memristor neurons are NbO x Output node of the memristor.

[0045] Specifically, Figure 2 As shown, the undirected graph provided in this embodiment is mapped according to the established mapping rules. Figure 2 The undirected graph shown is mapped to a memristor coupled neural network.

[0046] Capacitively couple neuron 1 to neurons 2, 4, and 6, and resistively couple neuron 1 to neurons 3, 5, 7, and 8;

[0047] Capacitively couple neuron 2 with neurons 1, 3, and 5, and resistively couple neuron 2 with neurons 4, 6, 7, and 8;

[0048] Capacitively couple neuron 3 with neurons 2, 4, and 7, and resistively couple neuron 1, 5, 6, and 8;

[0049] Capacitively couple neuron 4 to neurons 1, 3, and 8, and resistively couple neuron 4 to neurons 2, 5, 6, and 7;

[0050] Capacitively couple neuron 5 to neurons 2 and 6, and resistively couple neuron 1, 3, 4, 7, and 8;

[0051] Capacitively couple neuron 6 with neurons 1, 5, and 7, and resistively couple neuron 6 with neurons 2, 3, 4, and 8;

[0052] Capacitively couple neuron 7 to neurons 3, 6, and 8, and resistively couple neuron 1, 2, 4, and 5;

[0053] Capacitively couple neuron 8 to neurons 4 and 7, and resistively couple neuron 1, 2, 3, 5, and 6;

[0054] Finally, the construction of the memristor coupled neural network is completed.

[0055] Furthermore, in this embodiment, NbO x Memristor is a current-controlled local active memristor whose resistance switching mechanism is the Poole-Frenkel heat conduction effect. Its physical model can be expressed as:

[0056]

[0057] where v m represents the voltage across the memristor; i m represents the current flowing through the memristor; T is the internal temperature of the memristor; R0, α0, α1 and α2 are the parameters of the physical model; C th With Γ th Respectively represent thermal effect capacitance and conductivity; T amb Represents the ambient temperature.

[0058] In this embodiment, the values ​​of the model parameters are R0=0.8Ω, α0=2923K, C th =1.324×10 -13 J / K, Γ th =5.953×10 -6 W / K, T amb =298K.

[0059] It should be noted that the use of the NbO x Memristor, a capacitor C, a resistor R, and a voltage source V for power supply and biasing the operating point D Constructing memristive neurons, NbO x The memristor is connected in parallel with the capacitor C, and one end of them is connected in series with the resistor R, which is then connected to the voltage source V D In series, finally NbO xThe other end of the memristor and capacitor C is grounded. Such a memristor neuron can generate current spikes and voltage oscillations. If two NbOx memristor neurons are coupled through a resistor at the node between the NbOx memristor and the resistor, there is an input delay in the voltage source of the two neurons, that is, one neuron inputs voltage at time 0, and the other neuron inputs voltage after a short delay. The outputs of the two memristor neurons initially have a certain phase difference, and finally show a state of in-phase synchronization after a period of time. If two NbOx memristor neurons are coupled through a capacitor, the outputs of the two neurons will eventually show a state of reverse synchronization. This characteristic is similar to the graph coloring problem. The connected vertices in an undirected graph cannot be colored with the same color, and can be simulated by the anti-phase synchronization state obtained by the coupling of neurons through capacitors; while any two unconnected vertices in an undirected graph can be colored with the same color, which can be simulated by the in-phase state obtained by the coupling of neurons through resistors.

[0060] Step 3: Set the parameters of the memristor coupled neural network according to the degree of each vertex obtained in step 1 and the maximum degree of all vertices, wherein the parameters include the coupling resistor R C , coupling capacitor C C , neuron capacitance C and compensation capacitance C′, where coupling capacitance C C Fixed to 1nF.

[0061] By setting the appropriate parameters:

[0062] R C =(Nd max )×R Δ

[0063] C=C0+(N-3)C Δ

[0064] C′ i =(d max -d i )·(C C ||C)

[0065] Among them, C i ′ is the compensation capacitor that needs to be connected in parallel across the memristor in the memristor neuron to prevent the network from failing to converge; d i is the degree of each vertex in the undirected graph, d max Then it is the largest degree among all vertices; C C =1nF, R C are coupling capacitance and coupling resistance respectively; C is the capacitance in the memristor neuron; N is the number of vertices in the undirected graph; R Δ , C0 and C Δ are constants, and in this embodiment are 147Ω, 40nF and 20nF respectively.

[0066] In this embodiment, after the memristor coupled neural network is constructed, the parameters are set according to the above formula: R C =735Ω、C C =1nF, C=140nF, C1′=C2′=C3′=C4′=C6′=C7′=0nF, C5′=C8′=0.99nF.

[0067] According to the above technical solution, a complete hardware memristor neural network can converge quickly and stably. The output phase of each memristor neuron automatically forms a cluster. The undirected graph vertices corresponding to the memristor neurons whose output phases are in the same cluster are colored with the same color, ultimately forming a coloring solution for the graph coloring problem.

[0068] Step 4: Apply the memristor coupled neural network with set parameters, input voltage waveform, output current waveform and final coloring result.

[0069] Specifically, in this embodiment, multiple groups of random input delays are set for the memristor coupled neural network, and inputs such as Figure 4 The voltage waveform is shown.

[0070] It should be noted that, in the present embodiment, there is a section of 0 input between each group of input delays for initializing the network in stages, and the output waveform and the final coloring result can be obtained after the memristor coupled neural network is stabilized.

[0071] It can be understood that this embodiment formulates a multi-input delay strategy for improving the network's ability to obtain the optimal solution to the graph coloring problem, that is, by randomly selecting multiple groups of memristor neuron voltage input times, multiple different coloring schemes can be obtained, and the result using the least number of colors is selected as the result of the network solving the graph coloring problem.

[0072] According to the above embodiment records, the present invention constructs a memristor coupled neural network and formulates corresponding network mapping rules, that is, the undirected graph vertex mapping is NbO x The connection properties of memristor neurons and vertices are mapped to NbO x Memristive neurons are mapped to NbO through capacitive coupling and the disconnected properties of vertices. x Memristor neurons are coupled through resistors. The input of the network is the voltage input of each neuron. The input of each pair of memristor neurons in the constructed memristor neural network should have an input delay, and the input delay is randomly distributed, that is, the time of input voltage of each memristor neuron is random.

[0073] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions and variations of these embodiments including components are made without departing from the principles and spirit of the present invention, and still fall within the scope of protection of the present invention.

Claims

1. A method for solving graph coloring problems based on hardware memristor coupled neural networks. It is characterized in that The steps include: Step 1: Obtain an undirected graph of the graph coloring problem to be solved, and calculate the degree of each vertex and the maximum degree of all vertices according to the connection relationship of each vertex in the undirected graph; Step 2: According to the coupling resistance R C and coupling capacitor C C The mapping rule specified by the output phase characteristic of the undirected graph is mapped to the memristor coupled neural network, thereby constructing a hardware memristor coupled neural network, wherein the hardware memristor coupled neural network includes NbO x Memristor neurons, coupling resistor R C and coupling capacitor C C , each NbO x Memristive neurons are similar to all other NbO x Memristive neurons are coupled to each other through a resistor R C Or coupling capacitor C C To couple; Step 3: Set the parameters of the memristor coupled neural network according to the degree of each vertex obtained in step 1 and the maximum degree of all vertices, wherein the parameters include the coupling resistor R C , coupling capacitor C C , neuron capacitance C and compensation capacitance C′; Step 4: Apply the memristor coupled neural network with set parameters, input voltage waveform, output current waveform and final coloring result.

2. According to claim 1, the method for solving and optimizing the graph coloring problem based on hardware memristor coupled neural network, It is characterized in that The coupling resistor R C The output phase characteristic of NbO is: x Memristive neurons are coupled to each other through a resistor R C When coupled, the neuron outputs are in phase and synchronized.

3. According to claim 1, the method for solving and optimizing graph coloring problems based on hardware memristor coupled neural network, It is characterized in that The coupling capacitor C C The output phase characteristic of NbO is: x Memristive neurons are coupled via capacitors C C When coupled, the neuron outputs are in anti-phase synchronization.

4. According to claim 1, the method for solving and optimizing graph coloring problems based on hardware memristor coupled neural network, It is characterized in that The mapping rule of the undirected graph to the neural network is: the vertex mapping of the undirected graph is NbO x Memristor neurons, connected vertices are mapped to NbO x Memristive neurons are coupled via capacitance, and unconnected vertices are mapped to NbO x Memristive neurons are coupled via resistors; NbO x The nodes for coupling between memristor neurons are NbO x Output node of the memristor.

5. According to claim 1, the method for solving and optimizing graph coloring problems based on hardware memristor coupled neural network, It is characterized in that The NbO x Memristive neurons include NbO x A memristor, a resistor, a capacitor and a voltage source; the NbO x The memristor is connected in parallel with the capacitor, and one end of it is connected to the resistor and voltage source. x The other end of the memristor and the capacitor is grounded.

6. The method for solving and optimizing graph coloring problems based on hardware memristor coupled neural network according to claim 5, It is characterized in that The NbO x The model expression of the memristor is: where i m and v m Respectively represent the flow through NbO x Memristor current and NbO x The voltage across the memristor, T is NbO x The state variable of the memristor, R 0 , α 0 , α 1 , α 2 , C th , Γ th , T amb is a constant.

7. The method for solving and optimizing graph coloring problems based on hardware memristor coupled neural network according to claim 1, It is characterized in that The parameter setting method of the memristor coupled neural network is: R C =(N-d max )×R Δ C=C 0 +(N-3)C Δ C′ i =(d max -d i )·(C c |||C) Among them, C′ i is the compensation capacitor that needs to be connected in parallel across the memristor in the memristor neuron; d i is the degree of each vertex in the undirected graph, d max Then it is the largest degree among all vertices; C C , R C are coupling capacitance and coupling resistance respectively; C is the capacitance in the memristor neuron; N is the number of vertices in the undirected graph, R Δ , C 0 and C Δ is a constant.

8. The method for solving and optimizing graph coloring problems based on hardware memristor coupled neural network according to claim 7, It is characterized in that The coupling capacitor C C Fixed to 1nF.