Information processing system and information processing method

Abstract

The present invention makes it possible to compare knowledge databases using a quantum computer.　A plurality of sentences are acquired, a graph Laplacian obtained by converting logical relationships between propositions of the sentences is acquired on the basis of the propositions, and quantum encoding instructions for applying the graph Laplacian to a unitary matrix are output to the quantum computer.

Description

Information processing system and information processing method

[0001] The present invention relates to an information processing system and an information processing method for constructing a knowledge database suitable for processing by a quantum computer.

[0002] A knowledge base is a database that is constructed to improve the searchability and usability of data by consolidating business experience, know-how, etc. in one place. Knowledge databases are organized based on the relationships between knowledge, and various technologies have been proposed to search these relationships using computer processing.

[0003] Patent Document 1 discloses a fuzzy inference technology that learns inference rules necessary for fuzzy inference and performs fuzzy inference based on the learning results. Patent Document 1 discloses that by learning the causal relationships between knowledge information as fuzzy inference rules, it becomes possible to acquire inference rules from knowledge information that includes ambiguity in a few minutes or less, and that fuzzy inference can be performed using the truth values ​​and validity of the rules.

[0004] Japanese Patent Application Publication No. 06-202870

[0005] Kubra Yeter-Aydeniz et al (2019). “Quantum topological data analysis with continuous variables”Ankit Khandelwal and M Girish Chandra (2023). “Quantum-Enhanced Topological Data Analysis: A Peep from an Implementation Perspective”

[0006] As mentioned above, database searches that involve connections between pieces of knowledge have had the problem of taking a long time to process. In particular, in recent years, the amount of information has increased, and the relationships between these pieces of information have tended to become more complex.

[0007] Quantum computers can demonstrate superior computing power compared to classical computers by using algorithms that utilize the characteristics of quantum computing. In order to demonstrate superior computing power with quantum computers, the challenge was to quantum encode and convert the information to be calculated.

[0008] The present invention has been made in view of the above-mentioned problems, and aims to provide a technology that enables knowledge database matching using quantum encoding that can be calculated advantageously by a quantum computer, and also aims to provide a technology that generates a quantum circuit that is advantageous for searching a knowledge database using quantum encoding.

[0009] In order to solve the above-mentioned problems, the present invention comprises a sentence acquisition unit, a conversion unit, and an instruction unit, wherein the sentence acquisition unit acquires a plurality of sentences, the conversion unit acquires a graph Laplacian by converting the logical relationship between propositions based on the propositions of the sentences, and the instruction unit outputs quantum encoding instructions to the quantum computer for applying the graph Laplacian to a unitary matrix.

[0010] With this configuration, text can be converted into information that can be calculated by a quantum computer.

[0011] In a preferred embodiment of the present invention, a matching unit is provided that acquires eigenvalues ​​from the quantum computer as quantum computation results in response to the quantum encoding instructions, and determines whether or not a logical relationship exists from the eigenvalues. With this configuration, it is possible to quickly calculate eigenvalues ​​using the quantum computer, and to determine whether or not a logical relationship exists in the target sentence based on the results.

[0012] In a preferred embodiment of the present invention, the system further comprises a knowledge database for storing information necessary for the quantum encoding instruction, and the matching unit matches a sentence to be matched with information necessary for the quantum encoding instruction, including graph data or a graph Laplacian, and outputs a quantum encoding instruction using the information necessary for quantum encoding, which is the matching result. With this configuration, it is possible to issue a quantum encoding instruction using the information necessary for quantum encoding corresponding to the sentence to be matched.

[0013] In a preferred embodiment of the present invention, the conversion unit includes an analysis unit that maps text vectors obtained by vectorizing text as point cloud data and outputs the identity between text vectors that connect each point by changing its radius as an analysis result; and a graph conversion unit that generates the graph Laplacian based on the text vectors and the analysis result. In a preferred embodiment of the present invention, the conversion unit includes a comparison unit that obtains eigenvalues ​​as quantum computation results for the quantum encoding instructions and determines that there is no logical connection if the Betti number of the zero-dimensional homology group from the eigenvalue is not 1. This configuration allows the range of semantic ambiguity of propositions accumulated through geometric analysis to be adjusted. Furthermore, if the Betti number indicates that there is no geometric connection, the conversion unit determines that there is no logical connection between the texts, thereby eliminating the need for further calculations of the logical relationship, thereby reducing computational costs.

[0014] In a preferred embodiment of the present invention, the system includes a circuit generation unit that generates a quantum circuit for processing the quantum-encoded information based on a multi-way gene tree, and the circuit generation unit generates the quantum circuit based on constraints that generate a gene tree consisting of a parent node having a unitary gate and a child node that is either a quantum bit, a unitary gate for a single quantum bit, or a unitary gate having one or more control bits. In a preferred embodiment of the present invention, the circuit generation unit generates the quantum circuit based on constraints that generate a gene tree in which nodes other than terminal nodes are unitary gates. In a preferred embodiment of the present invention, the circuit generation unit evaluates the generated quantum circuit using an evaluation function that highly evaluates quantum circuits with a small number of nodes. With this configuration, it is possible to generate a quantum circuit suitable for processing quantum-encoded text.

[0015] According to the present invention, it is possible to provide a technique that enables knowledge database matching using a quantum computer.

[0016] 1. A block diagram of a system according to this embodiment. A hardware configuration diagram according to this embodiment. A processing sequence for building a knowledge database according to this embodiment. A processing sequence for knowledge database matching according to this embodiment. A schematic diagram of the logical relationship of sentences according to this embodiment. A schematic explanatory diagram of persistent homology. A processing flowchart for building a knowledge database according to this embodiment. A configuration example of graph data according to this embodiment. A schematic diagram of graph data using nodes and edges. A processing flowchart for knowledge database matching according to this embodiment. A configuration example of a quantum circuit including a controlled unitary gate according to this embodiment. A processing flowchart for generating a quantum circuit according to this embodiment.

[0017] An information processing system according to an embodiment of the present invention will be described below with reference to the accompanying drawings. Note that the embodiment shown below is merely an example of the present invention, and the present invention is not limited to the embodiment below, and various configurations can be adopted.

[0018] In this embodiment, the configurations, operations, etc. of an information processing system, an information processing device, and an information processing program are described, but similarly configured methods, computer programs, and program recording media on which the programs are recorded also achieve similar effects. For example, by using a program recording media, the programs can be installed on a computer. The series of processes according to this embodiment described below are provided as a computer-executable program, and can be provided via a non-transitory computer-readable recording medium such as a CD-ROM or a flexible disk, or even via a communication line.

[0019] The information processing system is composed of a computer device. The computer device has an arithmetic unit such as a CPU (Central Processing Unit) and a storage device. The computer device can function as an information processing device by executing an information processing program stored in the storage device using the arithmetic unit. The information processing method is realized by processing the computer device including the information processing device.

[0020] In this embodiment, sentences are interpreted as propositions. Sentences have a logical relationship based on propositions. Examples of propositions include "A is B" and "C is A." Based on such propositions, the logical relationship "C is B" is derived. By quantum encoding data representing the logical relationship of such propositions, high-speed quantum computation can be achieved using a quantum computer.

[0021] Fig. 1 shows a block diagram of an information processing system 1. The information processing system 1 includes an information processing device 2, a quantum computer 3, a user terminal 4, and a knowledge database NB, which are communicatively connected. Although Fig. 1 shows only one user terminal 4, there may be multiple user terminals 4. The quantum computer 3 may be configured to be mounted on the information processing device 2.

[0022] The information processing device 2 includes, as functional components, a sentence acquisition unit 21, a conversion unit 22, an instruction unit 23, a storage processing unit 24, a matching unit 25, and a circuit generation unit 26. The conversion unit 22 also includes a vector conversion unit 221, an analysis unit 222, and a graph conversion unit 223.

[0023] In this embodiment, the quantum computer 3 configures quantum bits using any hardware (for example, a superconducting quantum circuit or an optical quantum circuit) and performs quantum computation by performing quantum gate operations on the quantum bits. The quantum computer 3 performs quantum computation based on the quantum circuit.

[0024] The user terminal 4 accepts input of sentences to be stored in the knowledge database NB and sentences to be compared in the knowledge database NB. The user terminal 4 can build the knowledge database NB by storing sentences by sending the sentences to the information processing device 2. The user terminal 4 can refer to the comparison results by sending the sentences to be compared to the information processing device 2. In this embodiment, the sentences are processed as computer-processable text data.

[0025] 2(a) shows a hardware configuration diagram of the information processing device 2. The information processing device 2 includes a control device 201, a storage device 202, and a communication device 203 as its hardware configuration. In this embodiment, the information processing device 2 may be a computer device such as a server device or a personal computer. Note that the information processing device 2 may be configured by multiple computer devices, and is not limited to the configuration shown in FIG. 2(a) as long as it can realize the above-described functional components (21-26) as a whole.

[0026] The control device 201 is configured with one or more processors such as a CPU or a GPU (Graphics Processing Unit), and controls the overall processing of the information processing device 2 by executing an information processing program, an OS (Operating System), and other applications. The storage device 202 is a hard disk drive (HDD), a solid state drive (SSD), a flash memory, a RAM (Random Access Memory), or the like, and stores the information processing program and various data. The communication device 203 controls communication with a communication network and realizes data communication with the user terminal 4, the quantum computer 3, and the like.

[0027] 2B shows a hardware configuration diagram of the user terminal 4. The user terminal 4 includes, as its hardware configuration, a control device 401, a storage device 402, a communication device 403, an input device 404, and an output device 405. In this embodiment, the user terminal 4 may be a smartphone, a personal computer, a tablet terminal, or the like.

[0028] The control device 401 is composed of one or more processors such as a CPU, and controls the overall processing of the user terminal 4 by executing terminal programs, an OS, other applications, etc. The storage device 402 is an HDD, SSD, flash memory, RAM, etc., and stores a browser application and various data. The communication device 403 controls communication with the communication network and realizes data communication with at least the information processing device 2. The input device 404 is an input interface that accepts input operations by the user, and is composed of a microphone, touch panel, mouse, keyboard, etc. The output device 405 is composed of a display that displays and outputs data, etc.

[0029] <Construction of Knowledge Database> FIG. 3 shows a processing sequence of the information processing system 1 in the process of constructing a knowledge database.

[0030] The user terminal 4 transmits a sentence to the information processing device 2 (S101). The information processing device 2 stores the sentence received from the user terminal 4 in the knowledge database NB (S102). The information processing device 2 also stores sentence vectors obtained by vectorizing the sentence received from the user terminal 4 in the knowledge database NB (S103). The information processing device 2 also stores graph data and / or graph Laplacians obtained by graphing the sentence received from the user terminal 4 in the knowledge database NB (S104).

[0031] <Knowledge Database Matching> Figure 4 shows the processing sequence of the information processing system 1 in the process of matching the knowledge database constructed in the process of building the knowledge database. In this embodiment, matching the knowledge database means that when a user inputs a sentence to be matched (target sentence), the target sentence is interpreted as a proposition and matched with sentences stored in the knowledge database NB, thereby obtaining a truth / falseness determination result for the proposition. Note that, as with conventional knowledge databases, it is of course possible to search and view sentences using any search query, etc.

[0032] The user terminal 4 transmits a sentence to be compared to the information processing device 2 (S201). The information processing device 2 vectorizes the received sentence and then transmits a matching request to the knowledge database NB (S202). The information processing device 2 may store the received sentence or sentence vector in the knowledge database NB. The information processing device 2 receives the matching result as graph data or a graph Laplacian (S203). If the information processing device 2 cannot obtain a corresponding matching result, it outputs a result indicating that matching is not possible to the user terminal 4. Furthermore, if the information processing device 2 obtains graph data as the matching result, it performs graph Laplacianization. The information processing device 2 outputs a quantum encoding instruction to the quantum computer 3 for applying the graph Laplacian to the unitary matrix (S204). The quantum computer 3 performs a quantum phase estimation calculation in accordance with the quantum encoding instruction (S205). The information processing device 2 receives the quantum calculation result of the quantum phase estimation corresponding to the quantum encoding instruction (S206). The information processing device 2 performs a matching determination on the quantum phase estimation result (S207). The information processing device 2 transmits an output based on the collation determination result to the user terminal 4 (S208). Here, the output based on the collation determination includes a truth / false result of the proposition. The truth / false result of the proposition may be that the proposition has a logical connection and is true, that the proposition has a logical connection but is false, that the proposition has no logical connection and is likely to be false, or that the proposition is considered to be false.

[0033] In this embodiment, sentences are represented as a knowledge graph. Sentences are treated as nodes. Logical relationships between sentences are treated as edges. Specifically, edges represent logical operators such as AND, OR, NOT, and IMP (→).

[0034] FIG. 5 shows an overview of the logical relationships between sentences in this embodiment. In FIG. 5, sentences A, B, and C have a logical relationship of "A AND B ⇒ C." Sentences A to C are represented as nodes N1 to N3, respectively. Edge E1 represents the logical relationship of AND. Edge E2 represents the logical relationship of conclusion. Here, node N4 of sentence D is not connected to sentences A to C by an edge, and is therefore treated as having no logical relationship.

[0035] Sentences and logical relationships between sentences can be represented by sentence vectors, which are vectorized sentences. Sentence vectorization can use an existing natural language processing model, and sentences are input into the natural language processing model to output sentence vectors. Sentence vectors are multidimensional vectors, and similar sentences have similar vectors, so they are mapped onto a close vector space. Furthermore, multiple sentence vectors having a logical relationship between sentences will each be mapped onto a relatively close vector space corresponding to the logical relationship. For example, the logical relationship between similar sentence vectors can be expressed as an inclusion relationship IMP.

[0036] When a proposition in one sentence is similar to a proposition in another sentence, it is desirable to interpret these propositions as identical. In addition, regarding the logical relationship between sentences, it is desirable to expand the interpretation of similar propositions so that propositions that previously had no logical relationship can be interpreted as having a logical relationship.

[0037] In this embodiment, a geometric analysis method is adopted to provide a broader interpretation of the propositions in a sentence. More specifically, sentence vectors are interpreted using an analysis method that applies persistent homology. For a detailed explanation of persistent homology, see Non-Patent Document 1. In the analysis method using persistent homology, for example, n-th order homology groups can be considered for a vector space. This analysis allows for information such as the number of connected components in the vector space to be obtained from the zero-th order homology groups, the number of one-dimensional holes (rings) in the vector space to be obtained from the first order homology groups, and the number of two-dimensional holes (voids) in the vector space to be obtained from the second order homology groups. The information obtained by this geometric analysis is called Betti numbers. Betti numbers are used in analyses such as classifying structures that are considered identical through continuous transformation as identical.

[0038] The analysis method of this embodiment is applied to interpreting logical relationships between sentence vectors mapped to a vector space based on Betti numbers. Specifically, a zero-dimensional Betti number represents the number of connected components, and if the zero-dimensional Betti number is 1, it can be analyzed that the object of analysis has one structural connected component. In other words, it can be analyzed that the object of analysis may have a series of connections in terms of logical relationships. Conversely, if the zero-dimensional Betti number is 2 or greater, it can be analyzed that the object of analysis is divided in terms of logical relationships and has no connections. Therefore, by obtaining the zero-dimensional Betti number, it is possible to determine whether or not a logical relationship exists.

[0039] 6 is a diagram illustrating an overview of persistent homology. Here, text vectors are considered to be mapped as points in a vector space. That is, text vectors are considered to be point cloud data in the vector space.

[0040] Figure 6(a) shows point cloud data corresponding to sentence vectors mapped onto a vector space. In persistent homology, each point in the point cloud is gradually expanded. Figure 6(b) shows the process of expanding each point in the point cloud data. Here, some points are connected to each other, which is interpreted as a decrease in the number of connected components. In Figure 6(c), the points in the point cloud data are further expanded, so that all points are connected and form a torus. By expanding each point as shown in Figures 6(a) to 6(c), a torus is finally formed, with the number of holes reduced to 1, and the Betti number changes. As the Betti number changes, the interpretation of the proposition of the sentence and the logical relationship between sentences also changes, which can be understood as adjusting the range of ambiguity in the semantics of the sentence.

[0041] FIG. 7 is a flowchart showing the detailed process of constructing the knowledge database NB in ​​the information processing device 2.

[0042] The text acquisition unit 21 acquires text related to knowledge to be stored in the knowledge database NB from the user terminal 4 (S301).

[0043] The vector conversion unit 221 converts the acquired sentences into sentence vectors and maps them to a vector space (S302). The sentence vectors mapped to the vector space are called distributed representations. The distributed representations are stored in the knowledge database NB.

[0044] The analysis unit 222 determines the similarity between the text vectors. In this embodiment, the similarity between the text vectors is determined by cosine similarity. In cosine similarity, two text vectors are normalized to a value ranging from -1 to 1. If the cosine similarity is 1, the angle between the text vectors is 0 degrees, and the similarity between the text vectors is determined to be high. If the cosine similarity is -1, the angle between the text vectors is 180 degrees, and the similarity between the text vectors is determined to be low. Note that the determination of the similarity between text vectors is not limited to cosine similarity, and methods such as Euclidean distance and KL divergence may also be used.

[0045] As in the analysis method described above, the analysis unit 222 expands each point by a radius r, using the sentence vectors mapped to a vector space as point cloud data. Here, the radius r is defined as a numerical value used to adjust the degree of similarity between the sentences described above, and can be set arbitrarily depending on the extent to which the breadth of sentence interpretation is to be broadened. For example, if there are two similar target propositions that are to be considered identical, a radius r is set that connects the points of the sentence vectors corresponding to those propositions in the vector space. If the radius r is zero, only sentences that exactly match are allowed to be identified. However, if the distance between the sentence vectors is equal to or greater than a certain value (this number is set as the radius r), the sentences can be identified, even if there is ambiguity in their meanings.

[0046] In this embodiment, the radius r is set as a constant, but an arbitrary radius r may be set via the user terminal 4. Furthermore, if a matching result cannot be obtained in response to a matching request, the radius r may be set to a larger value so that a matching request is made again. The radius r may be set to a different constant depending on the attributes and use of the knowledge stored in the knowledge database NB.

[0047] The analysis unit 222 can obtain the Betti number, which changes as a result of adjusting the radius r, as the analysis result. The Betti number represents the geometric characteristics of a structure made up of points constituting the point cloud data. The analysis unit 222 can obtain the identity of the proposition as the analysis result by the geometric transformation of the point cloud data.

[0048] The graph conversion unit 223 performs graph conversion based on the propositions of the sentences to graph the logical relationships between the propositions (S303). Specifically, the graph conversion unit 223 generates graph data based on the sentence vectors mapped to the vector space. The graph conversion unit 223 also generates a graph Laplacian by further converting the graph data. Here, the graph conversion unit 223 can generate graph data by treating two similar propositions as the same proposition using the analysis result of the analysis unit 222.

[0049] In this embodiment, graph data is composed of start point data and end point data. FIG. 8A shows an example of the configuration of graph data. The logical connection between propositions A to C has, for example, the logical relationship "A≒B" and the logical relationship "B is C." When proposition A is used as the start point data, the first row of the graph data has proposition A as the start point data and proposition B as the end point data. When proposition B is used as the start point data, the second row of the graph data has proposition B as the start point data and propositions A and C as the end point data. Note that the logical relationship "B is C" does not have an inverse logical relationship, so a row of graph data having proposition C as the start point data is not generated.

[0050] When the graph conversion unit 223 can obtain an analysis result that identifies similar propositions A and B as the same proposition, it can generate the graph data shown in FIG. 8(b). In FIG. 8, the first line of the graph data has proposition B as the starting data and proposition C as the ending data, indicating the logical relationship "B is C." Here, since propositions A and B are considered to be the same proposition, the graph data in the first line of FIG. 8(a) does not exist. As shown in FIG. 8(b), the analysis result by the analysis unit 222 is available, and the graph conversion unit 223 can generate graph data equivalent to FIG. 8(a) even if there is no sentence vector indicating the logical relationship "A ≒ B." This makes it possible to perform matching by expanding and interpreting the propositions even if the number of sentences stored in the knowledge database NB is small.

[0051] Quantum encoding of data represented by nodes and edges will be described with reference to Non-Patent Document 2. Here, data that can be represented by nodes and edges shown in FIG. 9 is considered. In this embodiment, this data corresponds to graph data. According to FIG. 9, the target data is composed of nodes N1 to N5, and nodes N1 to N3 constitute a partial structure k. This entire structure K is expressed as equation (1).

[0052]

[0053] Equations (2) and (3) show the boundary operator ∂2 of the two-dimensional simplex of the structure. Each column of the boundary operator ∂1 of the one-dimensional simplex represents the relationship between nodes N1 to N5; for example, the first column represents the relationship between nodes N1 and N2. The reinforcement operator ∂2 of the two-dimensional simplex represents the partial structure k.

[0054]

[0055] From these boundary operators, the graph Laplacian Δ1 shown in equation (4) is obtained.

[0056]

[0057] The graph Laplacian Δ1 is padded to have the data length required for quantum computation. Here, since the graph Laplacian Δ1 is a 6x6 matrix, it is padded to an 8x8 matrix, which is a power of 2. Note that padding is performed appropriately according to the matrix size of the target data. The padded graph Laplacian Δ1′ is shown in Equation (5).

[0058]

[0059] The graph Laplacian Δ1 is a Hermitian matrix. The dimension of the eigenspace of the zero eigenvalue in a Hermitian matrix is ​​the Betti number. The zero eigenvalue is the kernel dimension of the matrix and is the multiplicity of the matrix eigenvalues ​​of 0. In equation (5), the zero eigenvalue is 1, and the one-dimensional Betti number calculated in this example is 1. However, the calculation generally becomes more difficult as the dimension of the matrix increases. A quantum computer can quickly calculate eigenvalues ​​using a quantum phase estimation algorithm.

[0060] Next, the graph Laplacian Δ1 is transformed into a Hermitian matrix H, which is a linear combination of tensor products of Pauli operators shown in equation (6).

[0061]

[0062] By using the Hermitian matrix H, it is possible to express the unitary matrix U shown in equation (7). In this transformation, the kernel dimension information of the Hermitian matrix is ​​preserved, so by finding the kernel dimension of this unitary matrix, the eigenvalues, i.e., Betti numbers, can be derived. Quantum computers are known to have a high-speed eigenvalue estimation algorithm (also called a phase estimation algorithm) for the unitary matrix U, which enables high-speed calculations by the quantum computer by converting the target data into a unitary matrix. In the controlled unitary gate of the quantum circuit, a controlled unitary operator U is placed. In the quantum circuit, multiple power gates of the controlled unitary gate are placed.

[0063]

[0064] As described above, the graph transformation unit 223 can further transform the graph data into a graph Laplacian, thereby graphing the text and generating information necessary for quantum encoding by the quantum computer 3.

[0065] The storage processing unit 24 stores information necessary for quantum encoding in the knowledge database NB (S304). The information necessary for quantum encoding is graph data or graph Laplacian. The knowledge database NB stores at least sentences, sentence vectors corresponding to the sentences, and graph data. Preferably, the knowledge database NB also stores corresponding graph Laplacians. The knowledge database NB stores graph data or graph Laplacians that are the matching results in response to a matching request based on the sentence or sentence vector to be matched. In this embodiment, if graph data can be obtained as the matching result, it can then be converted into a graph Laplacian and a quantum encoding instruction can be output. However, obtaining a graph Laplacian as the matching result is more preferable because it allows the process of converting the graph data into a graph Laplacian in the matching stage to be omitted.

[0066] In one aspect, the vector conversion unit 221 generates sentence vectors by vectorizing sentences stored in the knowledge database NB. The graph conversion unit 223 generates graph data by graphing sentence vectors stored in the knowledge database NB. The instruction unit 23 generates a graph Laplacian from the graph data stored in the knowledge database NB, and generates quantum-encoded information by having the quantum computer 3 process quantum encoding based on the graph Laplacian. The timing of each data process is not limited to when new sentences are acquired, but may be performed as appropriate when a certain amount of data is accumulated in the knowledge database NB. This allows the logical relationships between sentences to be updated at any time.

[0067] FIG. 10 is a flowchart showing the detailed process of checking the knowledge database NB in ​​the information processing device 2.

[0068] First, the text acquisition unit 21 acquires a target text to be checked via the user terminal 4 (S401). In this embodiment, the target text is information the user wants to know the truth of, input as a text. For example, if the knowledge database NB stores knowledge related to in-house business, by inputting an in-house business whose truth or falsity is unknown as a text, a checking result indicating whether the text is correct or not is obtained as an output.

[0069] The vector conversion unit 221 converts the target sentence into a sentence vector (target sentence vector) (S402). The collation unit 25 requests the knowledge database NB to collate the converted target sentence vector (S403).

[0070] The analysis unit 222 may perform an analysis process in which the target sentence vectors are treated as point cloud data on persistent homology, and each point is inflated by the reciprocal r of the distance between the vectors to determine the ambiguity of the meaning. The analysis unit 222 may also receive an instruction input regarding the abstraction level of the interpretation of the proposition via the user terminal 4. The analysis unit 222 determines the radius r of the point cloud data according to the input instruction regarding the abstraction level.

[0071] The matching unit 25 acquires a matching result from the knowledge database NB (S404). The matching result is graph data or a graph Laplacian of a sentence vector corresponding to the target sentence vector stored in the knowledge database NB. If graph data is acquired as the matching result, the graph conversion unit 223 converts the acquired graph data into a graph Laplacian.

[0072] The instruction unit 23 outputs a quantum encoding instruction to the quantum computer 3 based on the acquired graph Laplacian (S405). Specifically, the instruction unit 23 generates a quantum encoding instruction for applying the graph Laplacian to the unitary matrix. The quantum computer 3 can perform quantum phase estimation in accordance with the quantum encoding instruction.

[0073] The quantum computer 3 performs quantum phase estimation from the quantum encoding instruction to be checked, and transmits the calculation result to the information processing device 2. The calculation result is the eigenvalues ​​of the quantum-encoded unitary matrix.

[0074] The collation unit 25 obtains the calculation result including the eigenvalues ​​of the unitary matrix from the quantum computer 3 (S406).

[0075] The collation unit 25 determines whether or not a logical relationship exists from the eigenvalues ​​(S407). Specifically, the collation unit 25 obtains a zero-dimensional Betti number from the eigenvalues, and if the zero-dimensional Betti number is 1, it determines that there is a logical connection (YES in S407), and then performs logical calculation in S408. If the zero-dimensional Betti number is 2 or more, the collation unit 25 determines that there is no logical connection (NO in S407), and the logical calculation can be omitted. Logical calculations using classical computers are computationally expensive, but quantum phase estimation using the quantum computer 3 makes it possible to exclude collation targets that have no logical relationship from the logical calculation, thereby reducing computational costs.

[0076] The matching unit 25 performs logical calculations on the matching targets determined to have a logical connection (S408). The logical calculations are performed by referring to logical relationships such as graph data of the matching results. The matching unit 25 outputs a truth / false determination result for the proposition of the matching target sentence as a result of the logical calculations. A false truth / false determination result indicates that there is no logical relationship, but the result may be output separately from the determination result of S407 that there is no logical connection. The determination result of S407 that there is no logical connection cannot strictly be determined to be true or false, but indicates a result that is highly likely to be false due to the lack of logical connection, or that is presumed to be false.

[0077] The collation unit 25 generates an output based on the calculation result and transmits it to the user terminal 4 (S409). The output based on the collation result includes information indicating whether or not the target sentences are logically connected. If a logical relationship exists, the output based on the collation result may also include a sentence that forms a logical connection with the result of determining whether or not the sentence is true when the sentence is used as a proposition based on the information stored in the knowledge database NB. This allows the user to effectively utilize their accumulated knowledge by, for example, referring to the sentences that serve as evidence.

[0078] <Quantum Circuit Generation> In this embodiment, the quantum circuit constituting the quantum computer 3 includes a controlled unitary gate. Fig. 11 shows an example of a configuration diagram of a quantum circuit QC including a controlled unitary gate. The quantum circuit QC includes a quantum phase estimation circuit QC1 and a quantum superposition state preparation circuit QC2. The quantum circuit QC is composed of quantum gates such as a Hadamard gate, measurements, and further controlled unitary gates G1 to G3.

[0079] In this embodiment, the quantum phase estimation circuit QC1 is used for the calculation of quantum phase estimation. In particular, the controlled unitary gates G1 to G3 are quantum gates used for the calculation of quantum phase estimation, and affect the speed of quantum calculation, etc.

[0080] 11 is simplified, but an actual quantum circuit would be long. Therefore, it is preferable to make the quantum circuit as short as possible and to improve its efficiency by arranging quantum gates efficiently.

[0081] The circuit generation unit 26 generates a quantum circuit that processes quantum phase estimation. The circuit generation unit 26 generates the quantum circuit by generating a gene tree based on a genetic algorithm.

[0082] 11 shows a process flowchart for generating a quantum circuit. The circuit generation unit 26 generates a quantum circuit using a genetic algorithm in accordance with the process flowchart shown in FIG.

[0083] First, the circuit generation unit 26 randomly generates parent quantum circuit data (S501). The circuit generation unit 26 evaluates the fitness of the quantum circuit data, which is a gene tree generated by an evaluation function (S502). The circuit generation unit 26 selects parent quantum circuit data evaluated as having high fitness (S503). The circuit generation unit 26 randomly selects two quantum circuit data generated as a gene tree at a certain rate and partially exchanges the nodes to generate quantum circuit data (S504). The circuit generation unit 26 randomly selects one quantum circuit data generated as a gene tree at a certain rate and replaces one node with a randomly selected node to generate quantum circuit data (S505). If the circuit generation unit 26 determines that the termination condition is not satisfied (NO in S506), it returns to S502 and repeats the process. If the circuit generation unit 26 determines that the termination condition is satisfied (YES in S506), it completes the process. A quantum phase estimation circuit QC1 is designed based on the generated quantum circuit data.

[0084] The constraints of the genetic algorithm in this embodiment will be explained. The gene tree is a multi-way tree, and is composed of child nodes branching from a parent node. A node is a one-qubit gate or the nth qubit. A terminal node is a qubit. A qubit gate that handles multiple qubits is a controlled unitary gate. When a controlled unitary gate is a parent node, the child nodes are multiple qubits. When a parent node is a quantum gate, the quantum gate processes the qubits of the child node. Nodes other than the terminal node are unitary gates. A unitary gate is a gate capable of performing quantum phase estimation calculations in this embodiment. A controlled unitary gate refers to a unitary gate that handles multiple input signals.

[0085] The evaluation function in this embodiment has the following characteristics. When a superposition of possible quantum states for the basis vectors of Pauli Strings is given as input, the evaluation function evaluates the difference between the measurement result that should be obtained and the measurement result that is actually obtained, and evaluates the gene tree as having a higher fitness as the difference approaches 0. While the L1 norm, L2 norm, etc. can be used to calculate the difference, the L2 norm is particularly preferable. The evaluation function evaluates the gene tree as having a higher fitness as the number of nodes decreases, in order to converge the genetic algorithm to a circuit as short as possible. The evaluation function evaluates the gene tree as having a higher fitness as the distance between the quantum bit to be controlled and the quantum gate controlling that quantum bit decreases, in order to converge the genetic algorithm to a configuration that promotes interaction between nearby quantum bits as much as possible. The evaluation function evaluates the gene tree as having a lower fitness when it has multiple quantum bits as child nodes, or when it has the same quantum bit as a child node. The evaluation function evaluates the gene tree as having a higher fitness as the number of high-cost T gates decreases.

[0086] In this embodiment, the circuit generation unit 26 can generate quantum circuit data by applying a genetic algorithm to each tensor product of the Pauli operator. The circuit generation unit 26 integrates the quantum circuit data generated for each tensor product to create a quantum circuit used for quantum encoding in which the graph Laplacian is applied to a unitary matrix. If a quantum circuit for quantum encoding is generated all at once using a genetic algorithm, problems such as the algorithm not converging or generation taking a long time can occur. By using a method of integrating circuits for calculating tensor products, the above-mentioned problems can be resolved.

[0087] The circuit generation unit 26 generates quantum circuit data using a gene tree based on the constraints and evaluation function described above. When the circuit generation unit 26 generates a controlled unitary gate as a parent node, it generates child nodes related to quantum bits that serve as at least two or more input signals. At this time, two or more child nodes are generated. This makes it possible to generate an efficient quantum circuit that includes an arrangement of controlled unitary gates used in the quantum phase estimation calculation according to this embodiment.

[0088] REFERENCE SIGNS LIST 1 Information processing system 2 Information processing device 21 Text acquisition unit 22 Conversion unit 221 Vector conversion unit 222 Analysis unit 223 Graph conversion unit 23 Instruction unit 24 Storage processing unit 25 Collation unit 26 Circuit generation unit 3 Quantum computer 4 User terminal NB Knowledge base

Claims

1. An information processing system that enables knowledge database matching using a quantum computer, comprising: a text acquisition unit, a conversion unit, and an instruction unit; the text acquisition unit acquires a plurality of texts; the conversion unit acquires a graph Laplacian by converting logical relationships between propositions based on the propositions of the texts; the conversion unit comprises an analysis unit that maps text vectors obtained by vectorizing the texts as point cloud data and outputs as an analysis result the identity between text vectors that are connected by changes in the radius of each point; and a graph conversion unit that generates the graph Laplacian based on the text vectors and the analysis result; and the instruction unit outputs a quantum encoding instruction to the quantum computer for applying the graph Laplacian to a unitary matrix.

2. An information processing system that enables knowledge database matching by a quantum computer, comprising a text acquisition unit, a conversion unit, an instruction unit, and a circuit generation unit, wherein the text acquisition unit acquires a plurality of texts, the conversion unit acquires a graph Laplacian that converts logical relationships between propositions based on the propositions of the texts, the instruction unit outputs a quantum encoding instruction to the quantum computer for applying the graph Laplacian to a unitary matrix, and the circuit generation unit generates a quantum circuit that processes quantum-encoded information in response to the quantum encoding instruction based on a gene tree that is a multi-way tree, and the quantum circuit is generated based on constraints that generate a gene tree consisting of a parent node having a unitary gate and a child node that is either a quantum bit, a unitary gate for a single quantum bit, or a unitary gate having one or more control bits.

3. An information processing system according to claim 1 or claim 2, further comprising a comparison unit that acquires eigenvalues ​​from the quantum computer as quantum computation results for the quantum encoding instruction, and determines whether or not a logical relationship exists from the eigenvalues.

4. An information processing system as described in claim 3, further comprising a knowledge database for storing information necessary for the quantum encoding instruction, wherein the matching unit matches the sentence to be matched with information necessary for the quantum encoding instruction including graph data or a graph Laplacian, and outputs a quantum encoding instruction using the information necessary for the quantum encoding instruction which is the matching result.

5. An information processing system as described in claim 1, further comprising a comparison unit that acquires an eigenvalue as a quantum computation result for the quantum encoding instruction, and determines that there is no logical connection when a zero-dimensional Betti number derived from the eigenvalue is not 1.

6. The information processing system according to claim 2, wherein the circuit generation unit generates a quantum circuit based on a constraint for generating a gene tree in which nodes other than terminal nodes are unitary gates.

7. An information processing system according to claim 2 or claim 6, wherein the circuit generation unit evaluates the generated quantum circuit using an evaluation function that evaluates the closeness of the arrangement of interacting quantum bits and that favors quantum circuits with a small number of nodes.

8. An information processing method enabling knowledge database matching by a quantum computer, comprising the steps of: a computer acquiring a plurality of sentences; acquiring a graph Laplacian by transforming the logical relationship between propositions based on the propositions of the sentences; and outputting a quantum encoding instruction to the quantum computer for applying the graph Laplacian to a unitary matrix, wherein in the step of acquiring the graph Laplacian, the computer maps the sentence vectors obtained by vectorizing the sentences as point cloud data, and outputs the identity between the sentence vectors connected by changes in the radius of each point as an analysis result, and generates the graph Laplacian based on the sentence vectors and the analysis result.

9. An information processing method that enables knowledge database matching by a quantum computer, comprising the steps of: acquiring a plurality of sentences; acquiring a graph Laplacian that transforms the logical relationship between propositions based on the propositions of the sentences; outputting a quantum encoding instruction to the quantum computer for applying the graph Laplacian to a unitary matrix; and generating a quantum circuit that processes quantum-encoded information based on a gene tree that is a multi-way tree in response to the quantum encoding instruction, wherein the quantum circuit is generated based on constraints that generate a gene tree composed of a parent node having a unitary gate and a child node that is either a quantum bit, a unitary gate for a single quantum bit, or a unitary gate having one or more control bits.

Images