Mplementation device for implementing simultaneous satisfaction of multiple states and multiple conditions and multiple matching

Abstract

The present invention relates to an implementation device for implementing multiple states and multiple conditions and multiple matching, and more specifically to an algorithm and system for solving combinatorial optimization problems, and more specifically to an implementation device for implementing multiple states and multiple conditions and multiple matching that provides a hybrid optimization technique for solving matching (assignment) problems between multiple objects while simultaneously satisfying multiple states / conditions.

Description

Technology Field

[0001] The present invention relates to an implementation device for implementing multiple states and multiple conditions and multiple matching, and more specifically to an algorithm and system for solving combinatorial optimization problems, and more specifically to an implementation device for implementing multiple states and multiple conditions and multiple matching that provides a hybrid optimization technique for solving matching (assignment) problems between multiple objects while simultaneously satisfying multiple states / conditions. Background Technology

[0002] Platforms supporting startups worldwide frequently require multiple matching among various entities, such as founding teams, mentors, investors, and support programs.

[0003] For example, a startup may need to find multiple mentors and investors simultaneously, and a mentor may also mentor multiple startups.

[0004] In such a many-to-many matching environment, it is difficult to guarantee effective connections using simple 1:1 matching models or manual selection methods.

[0005] In particular, since the conditions (e.g., industry suitability, region, stage, etc.) and statuses (e.g., startup growth stage, mentor's area of ​​expertise) preferred by each entity are diverse, a complex problem arises where a match satisfying multiple conditions simultaneously must be found.

[0006] However, many current startup support platforms use limited matching algorithms that consider only one or two conditions or rely on human matching, which limits matching efficiency and satisfaction.

[0007] Furthermore, incorrect matching can lead to a waste of support resources or an imbalance of opportunities, highlighting the growing need for more sophisticated matching technology.

[0008] Gale-Shapley's stable matching algorithm, a well-known traditional matching algorithm, finds a stable solution in 1:1 matching by considering the preference rankings of the participants, but it is difficult to apply directly to multiple conditions or many-to-many matching.

[0009] Meanwhile, weighted bipartite matching algorithms (e.g., the Hungarian algorithm) can find a 1:1 matching that optimizes the sum of profits in polynomial time, but if this is generalized to satisfy multiple constraints, the problem becomes rapidly complex.

[0010] In fact, when additional constraints (e.g., fairness constraints) are included in many-to-one matching, a stable solution may generally not exist (Fair Matching under Constraints: Theory and Applications), and NP-hardness is proven even on a small scale ([220606988] The Complexity of Finding Fair Many-to-One Matchings).

[0011] This implies an explosive increase in search space and computational complexity, suggesting that there are numerous research challenges that are difficult to solve with existing algorithms alone.

[0012] For example, it has been reported that when fairness is considered in many-to-one matching, such as between students and universities or startups and investors, the problem becomes NP-complete with only three colors (groups) and a limited degree ([220606988] The Complexity of Finding Fair Many-to-One Matchings).

[0013] Due to these theoretical limitations, actual platforms remain limited to heuristics that calculate a few conditions as a simple weighted sum or matching through human intervention.

[0014] Concerns about algorithmic bias have been raised because existing matching systems have not sufficiently addressed fairness (preventing bias toward specific groups) and diversity (ensuring diversity across industries and regions) (The Digital Revolution of Startup Matchmaking: AI and Computer). This is an issue that must be resolved to ensure reliability and transparency on global platforms.

[0015] In summary, there is an urgent need for research on new algorithms that can improve matching satisfaction, fairness, and diversity by simultaneously considering multiple states and conditions in various industries—such as global startup support platforms, HR matching services, resource allocation, scheduling, and logistics optimization—where multiple objects must be matched to satisfy multiple conditions simultaneously. Prior art literature

[0016] (Prior document 1) [220606988] The Complexity of Finding Fair Many-to-One Matchings (Prior document 2) The Digital Revolution of Startup Matchmaking: AI and Computer The problem to be solved

[0017] Accordingly, the present invention has been devised to resolve the aforementioned conventional problems, and the objective of the present invention is to provide a hybrid optimization computing device that combines a quark-based data structure, a quantum-inspired evolution algorithm (QiEA), simulated annealing (SA), and a chaos-based mutation technique (CM). means of solving the problem

[0018] An embodiment of the present invention for achieving the purpose of the present invention as described above is,

[0019] A quark-based input value setting unit (100) that creates a quark-based data structure containing state information in a single data unit for each of multiple objects and sets it as the input value of the problem, and

[0020] A QiEA execution unit (200) that sets an initial solution set represented by the above quark-based data structure and performs a quantum-inspired evolution algorithm (QiEA), and

[0021] In the above QiEA execution process, a probabilistic search execution unit (300) that additionally performs a simulated annealing (SA)-based probabilistic search to prevent falling into a local optimal, and

[0022] In the above search process, a mutation technique application unit (400) that applies a Chaos Mutation (CM) technique to maintain diversity of the solution set, and

[0023] It includes a matching candidate output unit (500) that calculates one or more optimal matching candidates satisfying multiple states and multiple conditions simultaneously and outputs the matching candidates.

[0024] In addition, embodiments of the present invention include,

[0025] In an implementation device for implementing the simultaneous satisfaction of multiple states and multiple conditions and multiple matching, in the pairing or assignment between multiple objects, the combination of states and constraints that each object can have are simultaneously satisfied. Effects of the invention

[0026] The implementation device for implementing simultaneous satisfaction of multiple states and multiple conditions and multiple matching according to the present invention provides the following effects.

[0027] First, by representing the multiple state information of each object as a single "quark" unit, it produces the effect of facilitating dynamic, simultaneous, and probabilistic state transitions.

[0028] Second, Quantum Inspired Algorithms (QiEA) enable a wide search space by simulating quantum superposition, observation, and evolutionary operations.

[0029] Third, Simulated Annealing (SA) is effective in preventing falling into a local optimum through probabilistic search based on an energy function (or objective function).

[0030] Fourth, chaos-based mutation techniques (CM) are effective in maintaining the diversity of the solution set and improving search performance.

[0031] In summary, the present invention enables the derivation of a globally optimal (or near-optimal) matching result while simultaneously satisfying multiple constraints by performing a hybrid optimization algorithm. Brief explanation of the drawing

[0032] FIG. 1 is a configuration diagram of an implementation device for implementing simultaneous satisfaction of multiple states and multiple conditions and multiple matching according to an embodiment of the present invention. FIG. 2 is a configuration diagram of a quark-based input value setting unit (100) of an implementation device for implementing simultaneous satisfaction of multiple states and multiple conditions and multiple matching according to an embodiment of the present invention. FIG. 3 is a configuration diagram of a QiEA execution unit (200) of an implementation device for implementing simultaneous satisfaction of multiple states and multiple conditions and multiple matching according to an embodiment of the present invention. FIG. 4 is a configuration diagram of a probabilistic search execution unit (300) of an implementation device for implementing simultaneous satisfaction of multiple states and multiple conditions and multiple matching according to an embodiment of the present invention. FIG. 5 is a configuration diagram of a variation technique application unit (400) of an implementation device for implementing simultaneous satisfaction of multiple states and multiple conditions and multiple matching according to an embodiment of the present invention. FIG. 6 is a configuration diagram of a matching candidate output unit (500) of an implementation device for implementing simultaneous satisfaction of multiple states and multiple conditions and multiple matching according to an embodiment of the present invention. Specific details for implementing the invention

[0033] The present invention is capable of various modifications and may have various embodiments. Specific embodiments are illustrated in the drawings and described in detail in the detailed description.

[0034] The effects and features of the present invention, and the methods for achieving them, will become clear by referring to the embodiments described in detail below together with the drawings.

[0035] However, the present invention is not limited to the embodiments disclosed below and can be implemented in various forms.

[0036] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. When describing with reference to the drawings, identical or corresponding components are given the same reference numerals, and redundant descriptions thereof will be omitted.

[0037] In the following embodiments, terms such as first, second, etc. are used not in a limiting sense, but for the purpose of distinguishing one component from another component.

[0038] In the following embodiments, singular expressions include plural expressions unless the context clearly indicates otherwise.

[0039] In the following embodiments, terms such as "include" or "have" mean that the features or components described in the specification are present, and do not preclude the possibility that one or more other features or components may be added.

[0040] If an embodiment can be implemented differently, a specific process sequence may be performed differently from the order described.

[0041] For example, two processes described in succession may be performed substantially simultaneously, or they may proceed in the reverse order of description.

[0042] FIG. 1 is a configuration diagram of an implementation device for implementing simultaneous satisfaction of multiple states and multiple conditions and multiple matching according to an embodiment of the present invention.

[0043] Referring to FIG. 1, an implementation device (600) for implementing simultaneous satisfaction of multiple states and multiple conditions and multiple matching is,

[0044] A quark-based input value setting unit (100) that creates a quark-based data structure containing state information in a single data unit for each of multiple objects and sets it as the input value of the problem, and

[0045] A QiEA execution unit (200) that sets an initial solution set represented by the above quark-based data structure and performs a quantum-inspired evolution algorithm (QiEA), and

[0046] In the above QiEA execution process, a probabilistic search execution unit (300) that additionally performs a simulated annealing (SA)-based probabilistic search to prevent falling into a local optimal, and

[0047] In the above search process, a mutation technique application unit (400) that applies a Chaos Mutation (CM) technique to maintain diversity of the solution set, and

[0048] It includes a matching candidate output unit (500) that calculates one or more optimal matching candidates satisfying multiple states and multiple conditions simultaneously and outputs the matching candidates.

[0049] Specifically, the quark-based input value setting unit (100) is responsible for the function of creating a quark-based data structure that includes state information in a single data unit for each of the multiple objects and setting it as the input value of the problem.

[0050] For example, the quark-based input value setting unit (100) is responsible for converting the state information of an object into a single data unit (quark) in a matching problem that considers multiple states and multiple conditions simultaneously, and performs the process of effectively modeling the state information of each of multiple objects and structuring it so that it can be utilized in subsequent optimization and search algorithms.

[0051] At this time, in order to perform the above function, as shown in FIG. 2, the quark-based input value setting unit (100) is,

[0052] A data collection module (110) for collecting state information of objects participating in matching and converting the collected state information into a normalized form to maintain a consistent data format,

[0053] A quark-based data structure generation module (120) that converts multiple state information for each object into a single vector (Quark Vector) to be used for search and optimization,

[0054] A state representation module (130) for creating a data structure including probabilistic weights (amplitudes) and transition possibilities between states (phases) so that they can be utilized in a quantum-inspired method (QiEA) when representing the attributes of an object,

[0055] It is characterized by including a data conversion module (140) that converts data so that it can be utilized for quantum state superposition and evolution operations in the QiEA execution unit (200).

[0056] Specifically, the data collection module (110) collects state information of objects participating in the matching, and converts the collected state information into a normalized form to maintain a consistent data format.

[0057] For example, in the case of investor-startup matching, information on investors' investment preferences and startups' business models is collected, while in the case of student-university matching, information on students' academic achievement and university admission criteria is collected.

[0058] Afterwards, the state information is converted into a normalized form to maintain a consistent data format. For example, continuous variables (age, income, etc.) are subjected to Min-Max normalization, and categorical variables (preferred industry, technology stack, etc.) are subjected to One-Hot Encoding.

[0059] The above quark-based data structure generation module (120) converts multiple state information for each object into a single vector (Quark Vector) so that it can be used for search and optimization.

[0060] At this time, the above Quark is a unit data structure containing multiple states, which allows an object to have multiple states, and converts the multiple state information for each object into a single vector (Quark Vector) so that it can be utilized for search and optimization.

[0061] In addition, key components include state, amplitude, and phase; the state is a multidimensional vector representing the attributes of an object, amplitude is a probabilistic weight of a specific state, and phase is a parameter that determines the probability of state change.

[0062] The above state representation module (130) is intended to generate a data structure including probabilistic weights (amplitudes) and transition possibilities between states (phases) so that they can be utilized in a quantum-inspired method (QiEA) when representing the attributes of an object.

[0063] For example, in the case of the startup-investor matching problem, the startup's attributes can be expressed as [AI field, Series A stage, revenue of 1 billion, high growth potential] and the investor's attributes as [preference for AI investment, experience in Series A or higher investment, minimum investment of 500 million], and each can be converted into a quark as follows.

[0064] Q_Startup = [1, 0, 1, 0.8] # (1: AI field, 0: Series B or higher, 1: Growth potential, 0.8: Stability index)

[0065] Q_Investor = [1, 1, 0, 0.9] # (1: Prefer AI investment, 1: Series A or higher, 0: Not prefer early stages, 0.9: Investment propensity)

[0066] At this time, the quark-based similarity can be calculated by comparing the above vectors through an inner product.

[0067] The above data conversion module (140) converts data so that it can be utilized for quantum state superposition and evolution operations in the QiEA execution unit (200).

[0068] That is, the quark vector is subsequently converted so that it can be used for quantum state superposition and evolution operations in the QiEA execution unit (200), and the converted data is used to find the optimal solution in probabilistic search (SA-based) and mutation techniques (CM-based).

[0069] In summary, the quark-based input value setting unit (100) converts object state information into a quark vector and probabilistically represents multiple states so that they can be utilized for search and optimization operations.

[0070] In addition, it enables the solution of multiple matching problems including the possibility of transitions between specific states, and enables the derivation of optimal matching solutions by combining with quantum-inspired evolution (QiEA) and search techniques (SA, CM).

[0071] In the case of the quarks described so far, they are fundamental particles in physics, meaning subatomic particles that cannot be broken down into smaller units like electrons, and there are six types of quarks: Up, Down, Charm, Strange, Top, and Bottom.

[0072] The above quarks combine through strong interaction to form baryons such as protons and neutrons.

[0073] At this time, the quark in the present invention is utilized as a method to integrate and represent various states of an object into a single data unit.

[0074] In other words, it defines a data unit (Quark) that includes multiple states and multiple conditions simultaneously, thereby representing the state of an object more efficiently and utilizing it for matching.

[0075] The above QiEA execution unit (200) sets an initial solution set represented by a quark-based data structure and performs a Quantum-inspired Evolutionary Algorithm (QiEA).

[0076] The above QiEA is an evolutionary algorithm that searches for an optimal solution by mimicking concepts of quantum computing (superposition, interference, observation, etc.), and is intended to solve complex matching problems involving multiple states and multiple conditions.

[0077] At this time, in order to perform the above function, as shown in FIG. 3, the QiEA execution unit (200) is,

[0078] An initial solution set generation module (210) for starting a search by setting an initial population,

[0079] A quantum state update module (220) for evolving entities through quantum bit (Q-bit) based computation,

[0080] A fitness evaluation module (230) for evaluating the fitness of each year and selecting a superior year,

[0081] A crossover and mutation operation module (240) for applying a quantum rotation gate (Q-Gate) and performing mutation operations,

[0082] It is characterized by including a search process iteration module (250) for continuously improving the optimal solution during a specified number of iterations.

[0083] Specifically, the initial solution set generation module (210) generates an initial solution by utilizing a quark-based data structure, and each solution (solution) is expressed in the form of an n-bit quantum chromosome.

[0084] In addition, each entity has a quantum bit (Q-bit) state represented by probability amplitudes (amplitude values: α, β).

[0085] For example, if we express the relationship between an object (startup) and an investor having five attributes (industry, investment size, technology suitability, etc.) in a startup-investor matching problem, it is as follows.

[0086] Q_Startup = [(0.6, 0.8), (0.7, 0.7), (0.5, 0.9), (0.8, 0.6), (0.4, 0.9)]

[0087] Q_Investor = [(0.9, 0.5), (0.6, 0.8), (0.7, 0.7), (0.5, 0.9), (0.3, 0.95)]

[0088] In this case, each property is expressed as (amplitude α, β), and |α| 2 + |β| 2= 1 must be satisfied, and the initial solution set is set with random amplitude values ​​(α, β), which allows various searches to be performed.

[0089] The above quantum state update module (220) performs the function of evolving entities through quantum bit (Q-bit) based operations.

[0090] That is, the quantum state is changed by updating the amplitude values ​​(α, β) of each individual, and the update is performed using a quantum rotating gate (Q-Gate).

[0091] The above quantum rotation gate is defined as follows.

[0092]

[0093] Here, θ (rotation angle) is a value that is updated toward the optimal solution during the evolution process, and after evaluating the fitness of each individual, θ is adjusted toward the direction of the superior solution to evolve the solution.

[0094] For example, in the startup-investor matching problem, if a specific investor's preference condition changes and the startup's suitability for investment in the AI ​​field (attribute 1) becomes a more important factor, a quantum rotating gate is applied to adjust the amplitude (α, β) value of that attribute so that it can be selected with a higher probability.

[0095] The above-mentioned fitness evaluation module (230) is intended to evaluate the fitness of each solution and select the best solution.

[0096] At this time, a fitness evaluation is performed for each entity (solution), and the fitness can be defined as follows.

[0097]

[0098] Here, w i is the importance weight for each attribute, and sim(Q i ,Target i) represents the similarity with a specific matching target.

[0099] For example, the calculation of the matching score between a startup and an investor is as follows.

[0100] Fitness(Q_Startup) = 0.5 × sim(Industry) + 0.3 × sim(Investment Size) + 0.2 × sim(Technology Fit)

[0101] At this time, individuals with the top 20 to 30% fitness are selected to perform crossover and mutation in the updated set of states.

[0102] The above-mentioned crossover and mutation operation module (240) is intended to perform quantum rotation gate (Q-Gate) application and mutation operation.

[0103] At this time, the Quantum Crossover operation selects two current state sets and mixes the amplitudes (α, β).

[0104] When set 1 is (α,β) and set 2 is (α2,β2), the updated state set generation value is as follows.

[0105]

[0106] The object created in this way is passed to the next set.

[0107] In addition, Quantum Mutation randomly changes the amplitude values ​​of some entities according to random probability (mutation rate pm) and increases randomness by applying Chaos Map-based Mutation (CM).

[0108] For example, if the investment preference attribute is varied, it is as follows.

[0109] Original value: (α, β) = (0.7, 0.7)

[0110] After mutation: (α, β) = (0.6, 0.8)

[0111] In this case, the higher the mutation probability, the wider the search space becomes, and the lower the probability, the faster the convergence speed becomes.

[0112] The above search process iteration module (250) is intended to continuously improve the optimal solution during a specified number of iterations.

[0113] At this time, the search is performed until the iteration limit or solution convergence condition is satisfied, and finally, the solution with the highest fitness is delivered to the matching candidate output unit (500).

[0114] Based on the content explained so far, an example of applying QiEA to startup-investor matching is presented below.

[0115] The problem definition is to find the optimal match between 10 startups and 10 investors, and in the case of initial solution generation, the attributes of the startups and investors are converted into quark vectors (including amplitude).

[0116] Subsequently, in the case of the quantum gate update, investors with a high preference for investing in the AI ​​field amplify the amplitude of the corresponding attribute, and in the case of the fit evaluation, a matching score is calculated by comparing the startup's AI technology score with the investor's preference for AI investment.

[0117] Subsequently, in the case of crossover and mutation, the attributes of promising investor-startup matching candidates are explored by crossover and mutation, and in the case of optimal solution derivation, the investor-startup combination with the highest degree of fit is finally output.

[0118] The above probabilistic search execution unit (300) additionally performs a simulated annealing (SA)-based probabilistic search to prevent falling into a local optimal during the above QiEA execution process.

[0119] Generally, QiEA finds the optimal solution through quantum state updates (Quantum Rotation Gate) and amplitude adjustment, but because there is a risk of convergence to a local optimal solution during the search process, to prevent this, the probabilistic search execution unit (300) applies a simulated annealing (SA) technique to accept less superior solutions with a specific probability during the search process, thereby strengthening the ability to search for the global optimal solution.

[0120] At this time, in order to perform the above function, as shown in FIG. 4, the probabilistic search execution unit (300) is,

[0121] An initialization module (310) for selecting one candidate solution (Qi) from the current set of quantum states and setting it as the initial solution,

[0122] It includes a new solution generation and evaluation module (320) for generating a new solution (Q_new) by applying a small mutation to the current solution (Qi) and evaluating the fitness of the new solution (Q_new).

[0123] Specifically, the initial setup module (310) is responsible for setting an initial solution (Qi) to start the search for performing a probabilistic search based on simulated annealing (SA), and determines the starting point for performing the SA search based on candidate solutions (matching combinations) generated through a set of quantum states in QiEA (Quantum Inspired Evolution Algorithm).

[0124] The above new solution generation and evaluation module (320) performs the function of generating a new solution (Q_new) by applying mutation to the current solution (Qi) and evaluating the fitness of the new solution, and as a core part of SA, it plays a role in maintaining diversity of search by accepting the new solution with a specific probability.

[0125] The above mutation technique application unit (400) applies a Chaos Map-based Mutation (CM) technique to maintain diversity of the solution set during the above search process.

[0126] Specifically, it is responsible for applying Chaos Mutation (CM) techniques to maintain diversity in the solution set and improve search performance during the execution of QiEA (Quantum-inspired Evolutionary Algorithm) and SA (Simulated Annealing).

[0127] While general mutation operations transform entities through random operations, Chaos Mutation (CM) is a technique that performs more unpredictable mutations by utilizing the chaotic characteristics occurring in nonlinear dynamical systems.

[0128] At this time, in order to perform the above function, as shown in FIG. 5, the variation technique application unit (400) is,

[0129] A chaos map generation module (410) for generating a chaos map-based random number sequence for mutation operations,

[0130] A mutation target selection module (420) for selecting attributes to apply mutation to in the current solution (Qi),

[0131] A mutation operation application module (430) for performing mutations using random numbers based on a chaos map,

[0132] A post-mutation fitness evaluation module (440) for evaluating the fitness of the mutated solution (Q_new),

[0133] It includes a probabilistic acceptance decision module (450) for performing SA-based probabilistic acceptance determination and repetition.

[0134] Specifically, in the above-mentioned Chaos Map generation module (410), the Chaos Map is a technique for generating unpredictable random numbers using the Chaos characteristics occurring in a non-linear dynamic system, and can provide a wider search space than a general random variation.

[0135] At this time, to determine mutation probabilities based on the chaos map, chaos equations such as the Logistic Map or Tent Map are used.

[0136] The Logistic Map equation is as follows.

[0137]

[0138] Here, xn is the current value (0 < xn < 1), and r is the confusion coefficient (3.57 ≤ r ≤ 4).

[0139] For example, if the initial value x0 = 0.5 and the chaos coefficient r = 3.8 are set, the following random number sequence can be generated through the chaos map.

[0140] 0.9500, 0.1824, 0.5673, 0.9335, 0.2367, ...

[0141] The random number sequence generated in this way is used to adjust the value of the attribute targeted for mutation.

[0142] The above mutation target selection module (420) is for selecting an attribute to apply a mutation to in the current solution (Qi).

[0143] In this case, the mutation target is selected with a certain probability to maintain diversity in the search space, and the probability of a specific attribute being mutated can be dynamically adjusted according to changes in the search state and fitness.

[0144] For example, when the properties of the current solution (Qi) are as follows,

[0145] Qi = [0.8, 0.3, 0.6] # (AI investment preference, Series A stage, growth potential)

[0146] If the mutation target attribute selected using the Chaos Map is Series A stage (0.3), the mutation is applied to that attribute.

[0147] The above mutation operation application module (430) is intended to perform mutations using random numbers based on a chaos map.

[0148] In this process, the existing attribute value is combined with the Chaos Map random number and converted into a new value.

[0149] For example, the variations in the Series A stage attribute of the startup-investor matching problem are as follows.

[0150] Existing property value: 0.3 (Series A step)

[0151] Random number extracted from Chaos Map: 0.1824

[0152] After applying the mutation,

[0153] 0.3 → 0.3 + (0.1824 × variation strength) = 0.355.

[0154] The new solution (Q_new) with the mutation applied in this way is evaluated for its fitness by comparing it with the existing solution (Qi).

[0155] The above mutation-post-fitness evaluation module (440) evaluates the fitness of the new solution when a new solution (Q_new) with the mutation applied is generated.

[0156] In this case, fitness can be defined differently depending on the search objective and is generally calculated as follows.

[0157] Goodness of fit of the existing solution (Qi),

[0158] Fitness(Qi) = 0.8 * AI Investment Preference + 0.3 * Series A + 0.6 * Growth Potential = 1.41

[0159] Fitness of the mutated new solution (Q_new),

[0160] Fitness(Q_new) = 0.8 * AI Investment Preference + 0.355 * Series A Stage + 0.6 * Growth Potential = 1.465

[0161] In other words, if the fitness increases, the mutated solution (Q_new) is considered a superior solution to the existing solution (Qi).

[0162] The above probabilistic acceptance decision module (450) is for performing SA-based probabilistic acceptance determination and iteration, and although the fitness of a new solution (Q_new) may be lower than that of an existing solution (Qi), it can be accepted with a certain probability by applying SA (Simulated Annealing)-based probabilistic search.

[0163] The probabilistic acceptance decision module (450) performing this determines whether to accept a solution by applying an SA-based probabilistic acceptance function.

[0164] The above matching candidate output unit (500) calculates one or more optimal matching candidates that simultaneously satisfy multiple states and multiple conditions, and outputs the matching candidates.

[0165] At this time, in order to perform the above function, as illustrated in FIG. 6, the matching candidate output unit (500) is,

[0166] An optimal matching selection module (510) for selecting optimal matching candidates by considering multiple states and multiple conditions,

[0167] A matching candidate evaluation module (520) for determining the final candidate group by evaluating the suitability of selected candidates,

[0168] A matching candidate sorting and filtering module (530) for sorting in order of high suitability and selecting output targets according to filtering criteria,

[0169] A matching candidate output module (540) for visualizing matching candidates and providing them to the user,

[0170] It includes a user feedback reflection module (550) for re-outputting matching candidates by reflecting user feedback.

[0171] Specifically, the optimal matching selection module (510) primarily selects candidates that satisfy specific evaluation criteria from among various matching combinations (solution sets) derived during the QiEA and SA-based search process.

[0172] In this process, a Quark-based data structure is utilized to compare matching conditions between objects using multidimensional vectors, and the likelihood of matching between each object (investor-startup, student-university, etc.) is evaluated.

[0173] For example, when the candidate solutions (Qi) with a high probability of matching between a startup and an investor are as follows,

[0174] Q1 = [0.8, 0.3, 0.6] # AI startup, Series A stage, growth potential

[0175] Q2 = [0.9, 0.5, 0.7] # AI Investor, Series A or higher investment experience, Investment reliability

[0176] Q3 = [0.7, 0.4, 0.5] # Fintech startup, Series B stage, moderate growth potential

[0177] Among these, a combination that simultaneously satisfies multiple states and multiple conditions is selected and passed to the next step.

[0178] The above matching candidate evaluation module (520) quantifies the similarity between matching candidates to select combinations with high fitness, and calculates a score using a fitness evaluation function.

[0179] For example, the fitness of the existing solution (Qi),

[0180] Fitness(Q1) = 0.8 * AI Fit + 0.3 * Investment Stage Fit + 0.6 * Growth Potential = 1.41

[0181] Based on the fit scores calculated in this way, a group of excellent matching candidates is selected.

[0182] The above matching candidate sorting and filtering module (530) may apply filtering that reflects user preferences and may output only matching candidates that satisfy specific conditions.

[0183] For example, in the case of the startup-investor matching problem or filtering conditions,

[0184] Minimum investment of 500 million won or more

[0185] Investment experience in the AI ​​field is essential.

[0186] Growth potential score 0.6 or higher

[0187] At this point, after sorting in order of suitability, only the final output targets satisfying the above conditions are retained.

[0188] The above matching candidate output module (540) provides the final selected matching candidate to the user and includes a visualization function.

[0189] At this time, the matching results can be presented in various formats such as tables, graphs, and recommendation scores, and on certain platforms, they may be provided in real-time through a dashboard.

[0190] The above user feedback reflection module (550) is intended to re-output matching candidates by reflecting user feedback.

[0191] In other words, if the user adjusts the matching candidates or changes specific conditions, a new recommendation list is generated by reflecting that information.

[0192] Meanwhile, the above-mentioned quark-based data structure stores multiple states that each object can have in the form of a probability distribution or amplitude, and

[0193] The above quantum-inspired evolution algorithm is characterized by considering each quark state as a logical structure similar to a quantum bit (qubit) and performing state transitions during the evolution operation.

[0194] In summary, a quark-based data structure represents the multiple states that each object can have as a single data unit and stores state information in the form of a probability distribution or amplitude.

[0195] Unlike traditional binary representation (0 or 1) methods, this is designed so that the attributes of an object can have multiple states and utilizes the concept of quantum states to maximize search efficiency.

[0196] In other words, quark-based data structures store the multiple states that each object can have in the form of probability distributions or amplitudes, thereby enabling dynamic switching of states during the search process.

[0197] In this case, the state of each object has a probabilistic weight, and is determined to a single state when a specific operation is performed.

[0198] In addition, the Quantum Inspired Evolution Algorithm (QiEA) treats quark states as a logical structure similar to qubits and performs state transitions during the evolution operation, and to this end, applies a quantum rotation gate (Q-Gate) to adjust the amplitude of the object state.

[0199] The goal is to repeatedly perform this state transition process to optimize the search space and derive a solution that satisfies multiple states and multiple conditions.

[0200] Meanwhile, the above-mentioned simulated annealing (SA)-based search is characterized by performing a probabilistic candidate solution search in a direction that reduces the target function value defined by the energy function, while inducing escape from the local optimum by dynamically adjusting the acceptance probability according to the temperature parameter.

[0201] In other words, Simulated Annealing (SA)-based search is a technique that performs probabilistic search to prevent getting stuck in a local optimum in the search space and to converge to a global optimum.

[0202] SA is an algorithm inspired by the 'annealing' process in physics, where metals are heated and then slowly cooled to form an optimal crystal structure. It enables the exploration of various solutions by allowing less superior solutions to be accepted based on specific probabilities during the search process.

[0203] In the SA-based search process, the search direction is set to decrease the Objective Function Value defined by the energy function, and the acceptance probability when moving from a specific state to the next state is dynamically adjusted according to the temperature parameter (T).

[0204] When the temperature is high, the degrees of freedom of search are increased to allow the exploration of more states, and as the temperature gradually decreases, the convergence process is strengthened by selecting superior solutions.

[0205] When the search begins, an initial temperature (T0) is set, and based on this, the probabilistic acceptance during the search process is determined.

[0206] New candidate year (Q new If ) is generated, the existing solution (Q old The fitness is calculated by evaluating the difference from the existing solution, and if the fitness is higher than the existing solution, it is accepted unconditionally, but even if the fitness is lower, it is accepted with a certain probability by applying a probabilistic acceptance function.

[0207] New Year (Q new Fitness(Q new )) is the existing solution (Q oldIf it is lower than ), the stochastic acceptance formula is defined as follows.

[0208]

[0209] Here, P represents the probability of accepting a new solution, and ΔE = Fitness(Q new ) - Fitness(Q old )(goodness of fit difference), where T is the current temperature, and k is the Boltzmann constant (generally set to 1).

[0210] For example, when the temperature is high, the P value is large, so the search proceeds freely and various candidate solutions can be explored.

[0211] However, as the search progresses, the temperature gradually decreases (e.g., Tew = Told × α, 0 < α < 1), and the probability of accepting a solution with low fitness gradually decreases, increasing the likelihood of convergence to the optimal solution.

[0212] Through this process, SA-based search enhances the ability to find the global optimal solution by considering various candidate solutions in the early stages of the search, and improves convergence performance by favoring solutions with high fitness in the later stages.

[0213] This has the characteristic of resolving the problem of getting stuck in a local optimum and inducing the search for a better global optimum.

[0214] Meanwhile, the above-mentioned chaos map-based mutation technique is characterized by maintaining diversity across the entire search space by using a logistic map or Sinai map based on chaos theory to mutate a part of the solution set or by applying it to assign a random seed when generating initial entities.

[0215] In other words, Chaos Map-Based Mutation is a technique that maintains diversity in the search space by mutating a part of the solution set based on Chaos Theory or by assigning a random seed when generating the initial population.

[0216] While general mutation techniques perform random mutations, utilizing chaos maps allows for mutations that follow specific patterns while maintaining randomness, enabling a balanced exploration of the search space.

[0217] Chaos map-based mutation techniques play a role in diversifying search paths to prevent falling into a local optimum during the search process, and enable more effective utilization of the global search space by adjusting the randomness of the initial solution set.

[0218] In summary, the chaos map-based mutation technique is characterized by maintaining diversity across the entire search space by using a Logistic Map or Sinai Map based on chaos theory to mutate a portion of the solution set or by applying it to assign a random seed when generating the initial population.

[0219] This technique first generates a random number sequence based on non-linear dynamics using a chaos map generation module, and determines the entities or attributes to be mutated within the search space through a mutation target selection module.

[0220] A mutation operation using random numbers from a chaos map is applied to the selected mutation targets, and the effectiveness of the search process is verified by evaluating the fitness after mutation.

[0221] In addition, the mutation rate can be increased in the early stages of the search to maintain the degrees of freedom, and the mutation intensity can be reduced in the later stages to induce convergence to the optimal solution.

[0222] Through this approach, the chaos map-based mutation technique enables effective global exploration of the search space and contributes to preventing getting stuck in a local optimum.

[0223] Meanwhile, to verify optimal matching candidates, the method further includes a superior solution selection unit that determines whether each candidate solution satisfies specific constraints (e.g., time, resources, preferences, etc.) for all object state combinations, calculates the degree of satisfaction or energy value, and selects superior solutions based on this.

[0224] In other words, the Optimal Solution Selection Unit performs the role of verifying optimal matching candidates. It determines whether the combination of states of all objects for each candidate solution satisfies specific constraints (e.g., time, resources, preferences, etc.), calculates the degree of satisfaction, and is responsible for selecting the optimal solution.

[0225] This enables the final selection of the most suitable solution among the candidate solutions derived by the search algorithms (QiEA and SA). To this end, the superior solution selection unit undergoes the processes of verifying the constraints of candidate solutions, evaluating constraint satisfaction and energy values, selecting and sorting superior solutions, and confirming and outputting the optimal matching candidates.

[0226] For example, in the case of verifying constraints on candidate solutions, constraints are set according to the characteristics of the problem, such as time, resources, and preferences, and each candidate solution must satisfy these conditions.

[0227] After determining whether the conditions are met for each candidate year in this way, only the years that satisfy the conditions are passed to the next step.

[0228] Subsequently, for the evaluation of constraint satisfaction and energy values, the satisfaction of candidate solutions satisfying the constraints is quantitatively evaluated, and if necessary, the energy function is calculated to establish a comparison criterion.

[0229] At this time, the satisfaction level is calculated by evaluating how well each constraint is satisfied and converting it into a score (range 0 to 1), and weights may be applied if necessary.

[0230] Subsequently, for the selection and sorting of superior solutions, a process is performed to compare candidate solutions with calculated satisfaction and energy values ​​to sort superior solutions and select top candidates.

[0231] In this case, the sorting method can be based on fitness order or order of lowest energy value.

[0232] In other words, after sorting in order of highest satisfaction score, the top candidates are selected and confirmed as the optimal matching solution.

[0233] Subsequently, for the confirmation and output of optimal matching candidates, the process of confirming the optimal matching solution finally selected from among the superior solutions and outputting it as a matching candidate is performed.

[0234] This process may also incorporate additional user requirements (e.g., filtering specific industry groups, adjusting investment scale, etc.).

[0235] After determining the optimal matching solution in this way, the final matching result is provided to the user.

[0236] Through the present invention, multiple state information of each object is expressed in units of a single "quark," thereby facilitating dynamic, simultaneous, and probabilistic state transitions.

[0237] The embodiments according to the present invention described above may be implemented in the form of a computer program that can be executed through various components on a computer, and such a computer program may be recorded on a computer-readable medium.

[0238] At this time, the medium may include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory.

[0239] Meanwhile, the above computer program may be specially designed and configured for the present invention, or it may be one known and available to those skilled in the art of computer software.

[0240] Examples of computer programs may include not only machine code, such as that generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter, etc.

[0241] The specific embodiments described in this invention are examples and do not limit the scope of the invention in any way.

[0242] For the sake of brevity of the specification, descriptions of conventional electronic configurations, control systems, software, and other functional aspects of said systems may be omitted.

[0243] Additionally, the connections of lines or connecting members between the components shown in the drawings are exemplary representations of functional connections and / or physical or circuit connections, and may be replaced or additionally represented as various functional connections, physical connections, or circuit connections in the actual device.

[0244] In addition, unless there is a specific mention such as "essential" or "importantly," it may not be a component absolutely necessary for the application of the present invention.

[0245] In the specification of the present invention (particularly in the claims), the use of the term "above" and similar descriptive terms may be in both singular and plural forms.

[0246] In addition, where a range is described in the present invention, it includes an invention applying individual values ​​belonging to said range (unless otherwise stated), and is equivalent to describing each individual value constituting said range in the detailed description of the invention.

[0247] Finally, regarding the steps constituting the method according to the present invention, unless the order is explicitly stated or contrary to it, said steps may be performed in a suitable order.

[0248] The present invention is not necessarily limited to the order in which the above steps are described.

[0249] The use of all examples or exemplary terms (e.g., etc.) in the present invention is merely for the purpose of describing the invention in detail, and the scope of the invention is not limited by said examples or exemplary terms unless limited by the claims.

[0250] In addition, those skilled in the art will understand that various modifications, combinations, and changes may be configured according to design conditions and factors within the scope of the patent claims or equivalents to which they are added. Explanation of the symbols

[0251] 100 : Quark-based input value setting section 200 : QiEA Execution Unit 300 : Probabilistic search execution unit 400 : Variation Technique Application Section 500 : Matching Candidate Output Section

Claims

Claim 1 An implementation device for implementing simultaneous satisfaction of multiple states and multiple conditions and multiple matching comprises: a quark-based input value setting unit (100) that generates a quark-based data structure containing state information for each of multiple objects in a single data unit and sets it as the input value of the problem; a QiEA execution unit (200) that sets an initial solution set represented by the quark-based data structure and performs a Quantum Inspired Evolution Algorithm (QiEA); a probabilistic search execution unit (300) that additionally performs a simulated annealing (SA)-based probabilistic search to prevent falling into a local optimum during the QiEA execution process; a mutation technique application unit (400) that applies a Chaos Map-based Mutation (CM) technique to maintain diversity of the solution set during the search process; and a matching candidate output unit (500) that calculates one or more optimal matching candidates satisfying multiple states and multiple conditions simultaneously and outputs the matching candidates. An implementation device for implementing satisfaction and multiple matching. Claim 2 An implementation device for implementing simultaneous satisfaction of multiple states and multiple conditions and multiple matching, wherein, in the first paragraph, the multiple states and multiple conditions simultaneously satisfy the combination of states and constraints that each object may have in pairing or assignment between multiple objects.

Images