Information processing device and computer-readable recording medium

Abstract

This information processing device comprises an input unit that receives a processing parameter for generation processing, a calculation unit that executes generation processing that generates a plurality of pieces of generated data on the basis of the processing parameter and outputs evaluation data that includes at least an overall score and a partial score for each piece of generated data, an extraction unit that extracts a feature quantity for each piece of generated data on the basis of the evaluation data for the piece of generated data, a determination unit that outputs a determination result for each piece of generated data, the determination result being obtained by determining, on the basis of at least the feature quantity, whether the piece of generated data is to be accepted, a merging unit that outputs data for merging results obtained by merging the pieces of generated data that the determination unit has determined are to be accepted, and an output unit that outputs the merging results, including the results of the generation processing.

Description

Information Processing Apparatus and Computer-Readable Recording Medium

[0001] The present disclosure relates to an information processing apparatus and a computer-readable recording medium.

[0002] In recent years, in various industrial fields such as the manufacturing industry, attempts at DX (Digital Transformation) using information processing technologies in generative AI (Artificial Intelligence) and large language models (LLMs: Large Language Models) have been made (for example, Patent Document 1, etc.). There is also a movement to apply LLMs and the like to business operations at manufacturing sites such as factories.

[0003] Information processing technologies using these technologies have higher inference accuracy than conventional machine learning and can flexibly analyze non-formal and ambiguous input sentences including conversational tone. Also, the inference process is complex and it is difficult to explain the results based on internal specifications. In particular, the processing of generative AI has a high random number dependency, requires a large amount of calculation time and memory usage, and often uses external resources such as cloud environments. Many of the internal specifications of the models, applications, and infrastructure used in external resources are not publicly available, and the explainability of processing results is low. Also, compared to conventional machine learning, the amount of data required for learning is large, and the man-hours required for adjusting models for individual use cases tend to increase.

[0004] Japanese Patent Application Laid-Open No. 2020-135457

[0005] When applying large language models to business operations in industrial fields, there are many industry- or site-specific terms and noises (omissions, writing habits, etc.), and the accuracy may be insufficient even if known general methods are applied. Also, in many cases, for each individual business use case, it is necessary to focus on accurately recognizing specific types of inputs that are frequently used with high precision. In the drive control of industrial machines, significant losses can occur due to misoperations or misconfigurations. It is necessary to ensure both accuracy and explainability. When applying a model to a specific application, it is often not possible to directly apply the correct answer appropriateness judgment criteria for general-purpose models. Especially in the case of LLMs, if one word makes an incorrect transition, the meaning of the generated sentence may be greatly impaired. In such a situation, there is room for improvement in improving the performance of generative AI.

[0006] The information processing device disclosed herein solves the above problem by narrowing down the group of generated results using a group of generated scores, which is one of the attributes obtained as a result of a generation process related to predetermined information, and removing noise from the group of candidate system output sentences.

[0007] Furthermore, one aspect of the present disclosure is an information processing device comprising: an input unit for receiving processing parameters for generation processing; a calculation unit that performs generation processing to generate a plurality of generated data based on the processing parameters and outputs evaluation data including at least an overall score and a partial score for each of the generated data; an extraction unit that extracts feature quantities for each of the generated data based on the evaluation data relating to the generated data; a determination unit that outputs a determination result for each of the generated data, determining whether or not to accept it based at least on the feature quantities; a merger unit that outputs merged result data obtained by merging the generated data determined to be accepted by the determination unit; and an output unit that outputs the merged result including the result of the generation processing.

[0008] This is a schematic hardware configuration diagram of an information processing device according to one embodiment of the present disclosure. This is a block diagram showing the schematic functions of an information processing device according to the first embodiment. This is a schematic diagram showing an example of a translated text generated by the calculation unit. This is a schematic diagram showing an example of a score for each token calculated in the generation process. This is a schematic diagram showing another example of a score for each token calculated in the generation process. This is a schematic diagram illustrating feature quantities related to each translated text. This is a block diagram showing the schematic functions of an information processing device according to the second embodiment. This is a block diagram showing the schematic functions of an information processing device according to the third embodiment.

[0009] Embodiments of this disclosure will be described below with reference to the drawings. In the following description, components having the same or similar functions will be denoted by the same reference numerals. Duplication of these components may be omitted.

[0010] In this application, "based on XX" means "based on at least XX," and includes cases where it is based on another element in addition to XX. Furthermore, "based on XX" is not limited to cases where XX is used directly, but also includes cases where it is based on something that has been calculated or processed. "XX" is any element (for example, any information).

[0011] [First Embodiment] Figure 1 is a schematic hardware configuration diagram showing the main parts of an information processing device according to the first embodiment of the present disclosure. The information processing device 1 according to this embodiment can be implemented on a personal computer attached to a control device, or on a personal computer, cell computer, fog computer 6, cloud server 7, etc., connected to the control device via a wired / wireless network. In this embodiment, an example is shown in which the information processing device 1 is implemented on a personal computer attached to a control device 3 that controls an industrial machine 4.

[0012] The CPU 11 in the information processing device 1 according to this embodiment is a processor that controls the information processing device 1 as a whole. The CPU 11 reads the system program stored in the ROM 12 via the bus 22 and controls the entire information processing device 1 according to the system program. The RAM 13 temporarily stores temporary calculation data, display data, and various data acquired from external sources.

[0013] The non-volatile memory 14 is composed of, for example, a memory backed up by a battery (not shown) or an SSD (Solid State Drive), and its stored state is maintained even when the power to the information processing device 1 is turned off. The non-volatile memory 14 stores programs and data read from external devices 72 via the interface 15, programs and data input via the input device 71, and programs and data acquired from the control device 3 that controls the industrial machine 4 or other devices via the network 5. The programs and data stored in the non-volatile memory 14 may be expanded into the RAM 13 when executed or used. In addition, various system programs, such as known analysis programs, are pre-written in the ROM 12.

[0014] Interface 15 is an interface for connecting the CPU 11 of the information processing device 1 to an external device 72 such as a USB device. System programs, data, etc., can be read from the external device 72. Furthermore, programs and data created or edited within the information processing device 1 can be stored in external storage means via the external device 72.

[0015] Interface 20 is an interface for connecting the CPU 11 of the information processing device 1 to a wired or wireless network 5. The network 5 may communicate using technologies such as serial communication (RS-485, for example), Ethernet® communication, optical communication, wireless LAN, Wi-Fi®, Bluetooth®, etc. At least one industrial machine 4 to be controlled, at least one other industrial machine 4, a fog computer 6, a cloud server 7, etc. are connected to the network 5 and exchange data with the information processing device 1.

[0016] The display device 70 displays data obtained as a result of the execution of various data, programs, etc., loaded into memory, via the interface 17. The input device 71, consisting of a keyboard and pointing device, transmits commands, data, etc., based on operator operations to the CPU 11 via the interface 18.

[0017] The generation AI device 2 is configured as a device that receives a predetermined input and generates response data. The generation AI device 2 includes a generation model that has learned to generate predetermined data based on a predetermined input. This model can use known algorithms such as VAE (Variational Auto-Encoder), GAN (Generative Adversarial Network), Diffusion Model, or Transformer model. Using the generation model, the generation AI device 2 can perform tasks such as classifying and organizing existing text knowledge to support the work of users such as operators on the manufacturing floor, a chatbot for referencing the manual of the industrial machine 4, support for creating NC programs for the industrial machine 4 using ChatGPT, parameter setting support, operation support, and outputting some kind of inference result image. The information processing device 1 according to this disclosure may be connected to the generation AI device 2 via a network 5. The generation AI device 2 may include a generation model that has learned based on various information published on the internet, for example.

[0018] Figure 2 is a schematic block diagram showing the functions of the information processing device 1 according to the first embodiment of this disclosure. Each function of the information processing device 1 according to this embodiment is realized by the CPU 11 of the information processing device 1 shown in Figure 1 executing a system program and controlling the operation of each part of the information processing device 1.

[0019] The information processing device 1 of this embodiment includes an input unit 100, a calculation unit 110, an extraction unit 120, a determination unit 130, a merging unit 140, and an output unit 190. Furthermore, the RAM 13 to non-volatile memory 14 of the information processing device 1 is provided with a generation model storage unit 200, which is an area in which generation models used to generate data in a predetermined format are stored in advance.

[0020] The input unit 100 receives input of processing parameters related to information processing from the user. The processing parameters received by the input unit 100 are parameters related to the data generation process performed in the calculation unit 110. For example, if this generation process is the generation of predetermined text data using a model, the processing parameters may specify the model to be used for generation, or they may specify the type and number of data to be generated, the content of the generation process, or the data on which the generated data is based (e.g., source text, prompts, etc.). They may also be parameters such as Top-P, Top-K, and temperature that can be specified for the model used for generation. The model used for generation may be a model stored in the generation model storage unit 200, or it may be possible to specify an external model that performs generation processing, such as the generation AI device 2.

[0021] The input unit 100 may display an input screen for receiving processing parameters to the display device 70. In this case, the user can input processing parameters by operating the input device 71 while viewing the input screen. The input unit 100 receives processing parameters from the input. The input unit 100 may also receive input via an audio input device (not shown). Alternatively, it may receive processing parameters from the control device 3, fog computer 6, cloud server 7, etc., via the network 5 according to instructions from the user. The input unit 100 outputs the received processing parameters to the calculation unit 110.

[0022] The calculation unit 110 executes a generation process to generate multiple generated data based on the processing parameters received by the input unit 100. The calculation unit 110 may also read a generation model specified by the processing parameters from among the generation models stored in the generation model storage unit 200 and execute the generation process. Alternatively, it may request the generation AI device 2, etc., to execute the generation process specified by the processing parameters. This generation process may be, for example, generating an answer to a question, generating a summary, generating a translated text, generating an interpretive text, generating a corrected text, generating HTML source code, generating program code, generating an image, converting an image, generating music, converting music, etc. The calculation unit 110 executes the generation process so that multiple generated data are generated.

[0023] The calculation unit 110 further calculates evaluation data, including an overall score and a partial score, for each of the generated data. This evaluation data may use predetermined values ​​calculated by the model when generating the generated data, or it may be calculated separately by the calculation unit 110 based on predetermined values ​​calculated by the model. For example, if the generation process is translation generation, the overall score may be the confidence level of the entire generated translation. The partial score may be a score for each token, such as a word, phrase, or clause. This score is calculated based on the transition probability from the previous token. If the generation process is image generation using a GAN or the like, the overall score may use FID (Frechet Inception Distance) or Inception Score. The partial score may be calculated based on a probability distribution related to the partial image calculated in the intermediate layer. If the generation process is music generation, the overall score may use the degree of match to the song features given as a topic. Furthermore, partial scores can be calculated based on the transition probability between notes, chords, or phrases, the affinity between parts, etc. Note that while a single value is calculated for each generated data for the overall score, multiple partial scores are calculated for each generated data. A partial score is a score that evaluates a portion of the generated data from a predetermined perspective. The calculation unit 110 outputs the multiple generated data and the evaluation data related to each generated data to the extraction unit 120.

[0024] The extraction unit 120 extracts features for each generated data generated by the calculation unit 110. The features extracted by the extraction unit 120 are extracted from the evaluation data related to each generated data. For example, the extraction unit 120 may extract statistical values ​​such as the mean, median, maximum, and minimum values ​​of the partial scores included in the evaluation data related to each generated data as features. Alternatively, the number of partial scores that are below (or above) a predetermined threshold may be extracted as features. Thus, it is desirable to calculate features mainly based on the partial scores included in the evaluation data, but values ​​calculated by a predetermined formula relating to the overall score and partial scores may also be used as features. The extraction unit 120 outputs the extracted features together with the generated data and evaluation data to the determination unit 130.

[0025] The determination unit 130 determines whether or not to accept each generated data based on the features extracted by the extraction unit 120. It then outputs the result of the determination to the merger unit 140. The determination unit 130 may, for example, compare the features related to each generated data with a predetermined threshold and determine to accept the data if the features are greater than (or less than) the threshold. Alternatively, the threshold may be set as the acceptance ratio among multiple generated data, and the generated data may be accepted starting with the data with the largest (or smallest) features related to the generated data to satisfy the acceptance ratio. Furthermore, the determination may be made using multiple conditions, such as comparing the overall score included in the evaluation data with a first threshold and accepting generated data if the overall score is greater than the first threshold and the features of that data are greater than a second threshold. The conditions for accepting generated data should be predetermined considering the characteristics of the evaluation data and features. It is desirable that the conditions for accepting generated data include at least a condition related to the features. The determination unit 130 outputs the determination result of the acceptance conditions together with the generated data and evaluation data to the merger unit 140.

[0026] The merger unit 140 outputs merged data, which is the result of merging the generated data that has been selected for acceptance based on the judgment results related to the acceptance conditions for each feature. For example, if the generated data is text data representing a sentence, the merger unit 140 may merge multiple generated data that have been determined to be accepted into list-format data or tabular data. Also, if the generated data is image data or music data, the merger unit 140 may merge multiple generated data that have been determined to be accepted into list-format data or a single combined data. Furthermore, the merger unit 140 may perform the merge by extracting the common parts of the generated data that have been selected for acceptance. The processing related to the merger by the merger unit 140 may be a process of creating a union of multiple generated data as described above, or a process of calculating the common parts. The merger unit 140 outputs the merged generated data to the output unit 190.

[0027] The output unit 190 outputs the merged result, including the result of the generation process. The output unit 190 may also output the merged result, including the result of the generation process, to the display device 70. Alternatively, it may output to other applications specified by the user. Examples of other applications include applications related to machine learning. In this case, learning can be performed based on the merged result, including the result of the generation process. The output unit 190 may also transmit the merged result, including the result of the generation process, to other computers such as the fog computer 6 or the cloud server 7 via the network 5.

[0028] The operation of the information processing device 1, which has the above configuration, will be explained using Figures 3 to 6. In this example, we consider the case where data suitable for training a chatbot on variations of questions entered by an operator during past interactions with the operator is generated. A translation generation model that translates from Japanese to Japanese with the same meaning is used as the generation model. The question sentence used as a processing parameter for generation is the sentence "There is a strange noise coming from the machine and motor." The temperature, which is also a processing parameter for generation, is set to produce a moderate amount of randomness. For simplicity of explanation, the number of translation sentences to be generated will be 10.

[0029] Figure 3 is a schematic diagram showing an example of a translated sentence generated by the calculation unit 110. Each line illustrated in Figure 3 is a translated sentence obtained as a result of processing the sentence "There is a strange noise coming from the machine and motor" using specified processing parameters. Many of these translated sentences have the same meaning as the original sentence, but some have a different meaning from the original.

[0030] Figure 4 is a schematic diagram showing an example of the score for each token calculated during the generation process. In the example in Figure 4, the score for each token is shown for the translated sentence "There is a strange noise coming from the machine and motor," which is No. 2 in Figure 3, with each word treated as a token. Each score is calculated based on the probability of transitioning from the previous word to the next word, and the higher the value, the higher the probability that it has the same meaning as the original sentence. In the example in Figure 4, there are no tokens with extremely low scores. Therefore, this indicates that the translated sentence is likely to have a meaning close to the original sentence.

[0031] Figure 5 is a schematic diagram illustrating another example of the score per token calculated during the generation process. In the example in Figure 4, the score per token is shown for the translated sentence "There is a strange noise coming from the machine and Mozart," which is No. 6 in Figure 3, when each word is treated as a token. In the example in Figure 5, the score per token is extremely low for the word "Mozart." Therefore, it is highly likely that this translated sentence has a different meaning from the original text.

[0032] The calculation unit 110 calculates a partial score for each token when this word is treated as a token. The calculation unit 110 includes this value in the evaluation data and outputs it to the extraction unit 120.

[0033] In this example, the extraction unit 120 extracts the minimum value of this partial score from the evaluation data as a feature. This is because, in translated texts, if there is an extremely low value among the scores for each token, the meaning of the entire text often becomes distorted. While this is not the only way to calculate features, at least in this example, using the minimum value of the score for each token as a feature makes it possible to narrow down to appropriate translated texts.

[0034] Figure 6 is a schematic diagram illustrating the feature quantities associated with each translation. As illustrated in Figure 6, the extracted feature quantities for translations No. 6, No. 7, and No. 10 are extremely low.

[0035] The determination unit 130 compares the feature quantities calculated in this way with a predetermined threshold and selects translated sentences whose feature quantities are greater than the threshold for acceptance. For example, consider the case where the threshold is set to 0.2. In this case, the determination unit 130 compares the feature quantities of each translated sentence exemplified in Figure 6 with the threshold and determines that translated sentences No. 1 to 5, No. 8, and No. 9 are to be accepted. The merging unit 140 then merges these translated sentences that have been determined to be accepted into a list format. The output unit 190 outputs the list of merged translated sentences to the chatbot's learning unit for use in training.

[0036] The information processing device 1 according to this embodiment, having the above configuration, can, for example, when building a chatbot that responds to questions from operators based on individual business data originating from a manufacturing site in a factory, automatically generate data to a certain extent for use in training the chatbot, and then easily narrow down the data from the generated data to data suitable for training. It is also expected that this can be used for classifying and organizing knowledge from accumulated data, and for generating data used for training anomaly detection devices.

[0037] [Second Embodiment] The following describes an information processing device according to a second embodiment of the present disclosure. The information processing device 1 according to this embodiment has the same hardware configuration as the information processing device 1 according to the first embodiment.

[0038] Figure 7 is a schematic block diagram showing the functions of the information processing device 1 according to the second embodiment of this disclosure. Each function of the information processing device 1 according to this embodiment is realized by the CPU 11 of the information processing device 1 shown in Figure 1 executing a system program and controlling the operation of each part of the information processing device 1.

[0039] The information processing device 1 according to this embodiment includes an input unit 100, a calculation unit 110, an extraction unit 120, a determination unit 130, a merging unit 140, and an output unit 190, as well as a learning unit 150 and an inference unit 160. Furthermore, the RAM 13 to non-volatile memory 14 of the information processing device 1 is provided with a generation model storage unit 200, which is an area where a generation model used to generate data in a predetermined format is pre-stored, and a learning model storage unit 210, which is an area for storing a model that has learned the extracted features.

[0040] The input unit 100, calculation unit 110, extraction unit 120, merging unit 140, and output unit 190 of the information processing device 1 according to this embodiment are the same as the functions of the first embodiment.

[0041] The learning unit 150 performs a learning process based on the features extracted by the extraction unit 120 for multiple generated data sets, and generates a learning model. The learning unit 150 may, for example, perform unsupervised learning using the features extracted by the extraction unit 120 for each generated data set, and create a model related to unsupervised learning. In this case, the learning unit 150 should generate a model that has learned the distribution of features related to each generated data set. Alternatively, the learning unit 150 may, for example, perform supervised learning using the correct answer data provided by the user for each generated data set, and create a model related to supervised learning. In this case, the learning unit 150 learns whether or not to receive generated data related to the input features. The learning unit 150 stores the created model in the learning model storage unit 210. The learning model can be, for example, a cluster set, a neural network, a decision tree, a support vector machine, etc.

[0042] The learning unit 150 may perform sequential learning during the operation of the information processing apparatus 1. For example, when using a model of unsupervised learning, the learning unit 150 may sequentially use the feature amounts related to the generated data obtained during the operation of the information processing apparatus 1 for sequential learning. Also, when using a model of supervised learning, the learning unit 150 may perform learning at the timing when the user inputs correct data for the obtained generated data and its feature amounts. In the latter case, for example, a certain amount of generated data and feature amounts may be accumulated, and a mechanism may be provided such that the correct data is input by the user in a timely manner.

[0043] The inference unit 160 executes inference processing using the learning model created by the learning unit 150. For example, when the learning model generated by the learning unit 150 is a model of unsupervised learning, the inference unit 160 may infer that it is a reception target when it is a feature amount having a value exceeding a predetermined ratio among the models. Also, for example, when the learning model generated by the learning unit 150 is a model of supervised learning, the inference unit 160 may infer that it is a reception target when the output when input to the model is a value corresponding to reception. The inference unit 160 outputs the inferred result together with the generated data to the determination unit 130. Note that when the learning by the learning unit 150 is not sufficient, the inference unit 160 may output the generated data and its feature amounts to the determination unit 130 without performing inference processing.

[0044] The determination unit 130 determines whether to receive each generated data based on the result inferred by the inference unit 160. The determination unit 130 may be configured to determine that it receives the generated data inferred to be a reception target by the inference unit 160. Note that when the learning by the learning unit 150 is not sufficient and the inference processing is not performed by the inference unit 160, the determination unit 130 may determine whether to receive each generated data based on a predetermined criterion as in the first embodiment.

[0045] When the information processing apparatus 1 according to the present embodiment can prepare generated data and its correct labels in advance before actual operation, it creates a model related to machine learning using these data as learning data and uses this to perform a more accurate narrowing down of the generated data by determining whether to receive it.

[0046] [Third Embodiment] Hereinafter, an information processing apparatus according to the third embodiment of the present disclosure will be described. The information processing apparatus 1 according to the present embodiment has the same hardware configuration as the information processing apparatus 1 according to the first embodiment.

[0047] FIG. 8 shows schematically in block diagram the functions provided in the information processing apparatus 1 according to the second embodiment of the present disclosure. Each function provided in the information processing apparatus 1 according to the present embodiment is realized by the CPU 11 provided in the information processing apparatus 1 shown in FIG. 1 executing a system program and controlling the operations of each part of the information processing apparatus 1.

[0048] The information processing apparatus 1 according to the present embodiment further includes a selection unit 115, a cut-out unit 170, and an auxiliary unit 180 in addition to the input unit 100, calculation unit 110, extraction unit 120, determination unit 130, merging unit 140, learning unit 150, inference unit 160, and output unit 190. Also, in the RAM 13 to non-volatile memory 14 of the information processing apparatus 1, there are provided a generation model storage unit 200 which is an area where a generation model used for generating data in a predetermined format is stored in advance, and a learning model storage unit 210 which is an area for storing a model that has learned extraction features.

[0049] The input unit 100, calculation unit 110, extraction unit 120, determination unit 130, merging unit 140, learning unit 150, inference unit 160, and output unit 190 provided in the information processing apparatus 1 according to the present embodiment are the same as the respective functions according to the second embodiment.

[0050] The selection unit 115 selects from multiple generated data based on the evaluation data calculated by the calculation unit 110. The selection unit 115 may, for example, select a predetermined number of generated data from the top of the overall score, based on the overall score included in the evaluation data calculated for each generated data. Conversely, it may exclude a predetermined number of generated data from the bottom of the overall score and select the remaining generated data. It may also perform selection considering partial scores in addition to the overall score. The selection unit 115 outputs the selected generated data to the extraction unit 120. The selection unit 115 selects generated data, for example, to reduce the processing load performed by the extraction unit 120 and beyond, and for example, to exclude data that is clearly not for use.

[0051] The extraction unit 170 extracts a predetermined amount of data from the merged data obtained by the merger unit 140. The extraction unit 170 may also extract data from the merged data such that a predetermined number of generated data items are included, for example, from the top of the overall score. Alternatively, the extraction may take into account partial scores and features in addition to the overall score. The extraction unit 170 outputs the extracted data to the auxiliary unit 180.

[0052] The auxiliary unit 180 adds auxiliary information to the data extracted by the extraction unit 170 from the merged data. The auxiliary unit 180 may, for example, add information about tokens related to features as auxiliary information to each generated data. Alternatively, it may add statistics such as the minimum and maximum partial scores for each generated data, or the mean and median values ​​related to multiple features, as auxiliary information. The auxiliary unit 180 outputs the data with the added auxiliary information to the output unit 190. The output unit 190 then outputs the data with the added auxiliary information.

[0053] The information processing device 1 according to this embodiment selects the generated data to be processed, and by reducing the processing of unnecessary data, the overall processing load can be reduced. Furthermore, by extracting the necessary amount of data from the resulting data, only the data that the user can examine can be provided, and by adding auxiliary information, the effort required to examine the data can be reduced.

[0054] While embodiments of this disclosure have been described in detail above, this disclosure is not limited to the individual embodiments described above. These embodiments can be added, replaced, modified, partially deleted, etc., in any way that does not depart from the spirit of the invention or from the idea and intent of this disclosure derived from the claims and their equivalents. For example, the order of operations and processes in the embodiments described above are shown as examples only and are not limited thereto. The same applies when numerical values ​​or mathematical formulas are used in the description of the embodiments described above.

[0055] The following are annotations regarding embodiments of the present disclosure. (Annotation 1) An information processing device (1) according to one aspect of the present disclosure includes: an input unit (100) that accepts processing parameters for generation processing; a calculation unit (110) that performs generation processing to generate a plurality of generated data based on the processing parameters and outputs evaluation data including at least an overall score and a partial score for each of the generated data; an extraction unit (120) that extracts feature quantities for each of the generated data based on the evaluation data relating to the generated data; a determination unit (130) that outputs a determination result that determines whether or not to accept each of the generated data based at least on the feature quantities; a merger unit (140) that outputs merged result data obtained by merging the generated data that the determination unit (130) has determined to accept; and an output unit (190) that outputs the merged result including the result of the generation processing.

[0056] (Note 2) The feature quantities extracted by the information processing device (1) according to another aspect of the present disclosure include the minimum value of the partial score, and the determination unit (130) makes a binary determination of acceptance or rejection based on the feature quantities and a predetermined threshold value. (Note 3) The threshold value used by the information processing device (1) according to another aspect of the present disclosure indicates the acceptance rate of a plurality of the generated data.

[0057] (Note 4) The merger processing performed by the merger unit (140) of the information processing device (1) in other embodiments of this disclosure is performed by calculating the union or common part of the generated data to be received. (Note 5) The generation processing performed by the calculation unit (110) of the information processing device (1) in other embodiments of this disclosure takes text data as input and outputs text data.

[0058] (Note 6) An information processing device (1) according to another aspect of the present disclosure further comprises: a learning unit (150) that performs a learning process based on the features extracted by the extraction unit (120) and creates a learning model; and an inference unit (160) that infers whether or not to accept the generated data using the learning model based on the features relating to the generated data, wherein the determination unit (130) determines whether or not to accept the generated data based on the result of the inference by the inference unit (160).

[0059] (Note 7) An information processing device (1) according to another aspect of the present disclosure further comprises: a selection unit (115) that selects the generated data based on the evaluation data calculated by the calculation unit (110); an extraction unit (170) that extracts a predetermined amount of data from the merged result data merged by the merger unit (140); and an auxiliary unit (180) that adds auxiliary information to the data extracted by the extraction unit (170).

[0060] (Note 8) A computer-readable recording medium according to one aspect of the present disclosure records a program that causes the computer to operate as: an input unit (100) that accepts processing parameters for generation processing; a calculation unit (110) that performs generation processing to generate a plurality of generated data based on the processing parameters and outputs evaluation data including at least an overall score and a partial score for each of the generated data; an extraction unit (120) that extracts feature quantities for each of the generated data based on the evaluation data relating to the generated data; a determination unit (130) that outputs a determination result that determines whether or not to accept each of the generated data based at least on the feature quantities; a merger unit (140) that outputs merged result data obtained by merging the generated data that the determination unit (130) has determined to accept; and an output unit (190) that outputs the merged result including the results of the generation processing.

[0061] 1 Information processing device 2 Generative AI device 3 Control device 4 Industrial machine 5 Network 6 Fog computer 7 Cloud server 11 CPU 12 ROM 13 RAM 14 Non-volatile memory 15, 17, 18, 20 Interface 22 Bus 70 Display device 71 Input device 72 External device 100 Input unit 110 Calculation unit 115 Selection unit 120 Extraction unit 130 Judgment unit 140 Merging unit 150 Learning unit 160 Inference unit 170 Extraction unit 180 Auxiliary unit 190 Output unit 200 Generative model storage unit 210 Learning model storage unit

Claims

1. An information processing device comprising: an input unit that accepts processing parameters for generation processing; a calculation unit that executes generation processing to generate multiple generated data based on the processing parameters and outputs evaluation data for each generated data that includes at least an overall score and a partial score; an extraction unit that extracts feature quantities for each generated data based on the evaluation data relating to the generated data; a determination unit that outputs a determination result that determines whether or not to accept each generated data based at least on the feature quantities; a merger unit that outputs merged result data obtained by merging the generated data that the determination unit has determined to accept; and an output unit that outputs the merged result including the results of the generation processing.

2. The information processing apparatus according to claim 1, wherein the feature quantity includes the minimum value of the partial score, and the determination unit performs a binary determination of acceptance or rejection based on the feature quantity and a predetermined threshold value.

3. The information processing apparatus according to claim 2, wherein the threshold value indicates the reception ratio of a plurality of the generated data.

4. The information processing apparatus according to claim 1, wherein the processing related to the merger by the merger unit is performed by calculating the union or common part of the generated data to be received.

5. The information processing apparatus according to claim 1, wherein the generation process performed by the calculation unit takes text data as input and outputs text data.

6. The information processing apparatus according to claim 1, further comprising: a learning unit that performs a learning process based on the features extracted by the extraction unit and creates a learning model; and an inference unit that infers whether or not to accept the generated data using the learning model based on the features relating to the generated data, wherein the determination unit determines whether or not to accept the generated data based on the result of the inference by the inference unit.

7. The information processing apparatus according to claim 1, further comprising: a selection unit for selecting the generated data based on the evaluation data calculated by the calculation unit; an extraction unit for extracting a predetermined amount of data from the merged result data merged by the merger unit; and an auxiliary unit for adding auxiliary information to the data extracted by the extraction unit.

8. A computer-readable recording medium that records a program causing a computer to operate as: an input unit that accepts processing parameters for generation processing; a calculation unit that performs generation processing to generate multiple generated data based on the processing parameters and outputs evaluation data including at least an overall score and a partial score for each of the generated data; an extraction unit that extracts feature quantities for each of the generated data based on the evaluation data relating to the generated data; a determination unit that outputs a determination result that determines whether or not to accept each of the generated data based at least on the feature quantities; a merging unit that outputs merged result data obtained by merging the generated data that the determination unit has determined to accept; and an output unit that outputs the merged result including the results of the generation processing.

Images