JP2011014049A - Hereditary processor, hereditary processing method and hereditary processing program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To create a converter by a stable evolutionary computation in hereditary processing.SOLUTION: A hereditary processor is provided including: a creation part which creates a current-generation converter by hereditary processing from at least one pre-generation converter which converts input data into output data; a creation part which updates the weight data of pre-generation or before to weight data of current generation; and an adaptability degree computation part which computes current generation adaptability degree according to the result obtained by weighting and comparing the converted data obtained by converting input data for learning and target data for learning with the weight data of the current generation, and old generation adaptability degree according to the result obtained by weighting and comparing the converted data and the target data for leaning with the weight data of the pre-generation or before for each of the current generation converters.

Description

  The present invention relates to a genetic processing apparatus, a genetic processing method, and a genetic processing program.

  A method for generating a filter or a converter using evolutionary computation such as a genetic algorithm or genetic programming is known (see Non-Patent Document 1). According to the method of generating a filter or a converter using such evolutionary computation, fewer filters or converters having a complicated structure that are optimal for each case and difficult to obtain analytically are obtained. It can be designed with effort and time.

Masayoshi Maebuchi and two others, “Study on Image Filter Design Using Genetic Algorithm” [online], Computer Utilization Education Council, [March 20, 2008 search], Internet <URL: http: //www.ciec. or.jp/event/2003/papers/pdf/E00086.pdf>

  By the way, when a converter is generated by such evolutionary calculation, weight data that weights the data output by the converter in each generation is used to bring the converter closer to the target converter. However, if evolution is performed using only the weight data of the current generation, the variation of the weight data between generations may be large, and evolutionary computation may become unstable.

  Therefore, an object of the present invention is to provide a genetic processing apparatus, a genetic processing method, and a genetic processing program that can solve the above-described problems. This object is achieved by a combination of features described in the independent claims. The dependent claims define further advantageous specific examples of the present invention.

  In order to solve the above-described problem, in the first aspect of the present invention, a generation unit that generates a current generation converter by genetic processing from at least one previous generation converter that converts input data into output data. A weight generation unit that updates the previous generation weight data to the current generation weight data, and the converted data obtained by converting the learning input data and the learning target data for each current generation converter. Current generation suitability according to the result of weighting and comparing with generation weight data, and old generation conformance according to the result of weighting and comparing converted data and learning target data with weight data of previous generation A genetic processing device including a fitness calculation unit for calculating a degree, a genetic processing method related to the genetic processing device, and a genetic processing program are provided.

  The above summary of the invention does not enumerate all necessary features of the present invention. In addition, a sub-combination of these feature groups can also be an invention.

1 shows a configuration of a genetic processing apparatus 10 according to the present embodiment. An example of the converter 20 which combined the processing component 22 which concerns on this embodiment in series is shown. An example of the converter 20 which combined the processing component 22 which concerns on this embodiment with the tree structure is shown. An example of a genetic operation performed on the converter 20 in which the processing components 22 according to this embodiment are combined in series is shown. An example of crossover operation performed with respect to the converter 20 which combined the process component 22 which concerns on this embodiment with the tree structure is shown. An example of the mutation operation performed with respect to the converter 20 which combined the processing component 22 which concerns on this embodiment with the tree structure is shown. An example of a method for calculating the degree of approximation, which is an example of a parameter representing the degree of fitness of the converter 20 according to the present embodiment, is shown. An example of the operation | movement flow of the genetic processing apparatus 10 which concerns on this embodiment is shown. An example of the process flow of the genetic processing apparatus 10 in step S14 of FIG. 8 is shown. An example of the processing flow of the weight generation unit 43 in step S19 of FIG. 8 is shown. 2 shows an exemplary hardware configuration of a computer 1900 according to the present embodiment.

  Hereinafter, the present invention will be described through embodiments of the invention, but the following embodiments do not limit the invention according to the claims. In addition, not all the combinations of features described in the embodiments are essential for the solving means of the invention.

  FIG. 1 shows a configuration of a genetic processing apparatus 10 according to the present embodiment. The genetic processing apparatus 10 evolves a group of transducers including at least one transducer 20 over a plurality of generations based on evolutionary computation. Here, the genetic processing apparatus 10 evolves the converter 20 using the weight data used in at least one generation before the previous generation in addition to the weight data before the current generation. Then, the genetic processing device 10 generates and updates a converter 20 suitable for converting learning input data into learning target data for each generation. The genetic processing device 10 is realized by a computer as an example. Moreover, the converter 20 may be an image filter as an example.

  The genetic processing apparatus 10 includes a converter storage unit 34, a learning input data storage unit 36, a conversion processing unit 38, a converted data storage unit 40, a learning target data storage unit 42, and a weight generation unit 43. A weight storage unit 60, a fitness calculation unit 44, a control unit 52, a selection unit 46, an update unit 48, and a generation unit 50. The converter storage unit 34 stores a plurality of converters 20 having different configurations. In this embodiment, each converter 20 has a structure in which inputs and outputs of a plurality of processing components 22 that process learning input data and output processing results as converted data are combined.

  The learning input data storage unit 36 stores learning input data to be converted by the converter 20. As an example, the learning input data may be a sample of data that is actually given to the converter 20 in an application to which the converter 20 generated by the genetic processing device 10 is applied. When the converter 20 is an image filter, the learning input data may be an image generated or photographed in advance by the user as an example.

  The conversion processing unit 38 sequentially acquires the plurality of converters 20 stored in the converter storage unit 34. The conversion processing unit 38 converts the learning input data stored in the learning input data storage unit 36 by each of the acquired converters 20 to generate converted data.

  The converted data storage unit 40 stores the converted data generated by the conversion processing unit 38. As an example, the converted data storage unit 40 stores the generated converted data in association with the converted converter 20.

  The learning target data storage unit 42 stores learning target data that is a target of converted data generated by converting the learning input data by the converter 20. The learning target data is, for example, a sample of data that the converter 20 that has input the learning input data should actually output in an application to which the converter 20 generated by the genetic processing device 10 is applied. It's okay. When the converter 20 is an image filter, the learning target data may be an image generated or prepared in advance by a user, for example.

  The weight generation unit 43 updates the weight data before the previous generation in the comparison between the converted data and the learning target data to the current generation weight data. For example, the weight generation unit 43 generates weight data in which data is arranged in the same manner as the converted data or learning target data.

  Here, the weight generation unit 43 sets the weight for the data having a larger difference for each data of at least one set of the converted data and the learning target data than the weight for the data having the smaller difference. Increased weight data may be generated. For example, the weight generation unit 43 adds a predetermined value to the weight corresponding to data having a difference greater than the threshold in the previous generation weight data, and sets a value smaller than the predetermined value to the weight corresponding to data having the difference equal to or less than the threshold. Addition to generate weight data for the current generation.

  Instead, the weight generation unit 43 adds a predetermined value to the weight corresponding to data having a difference greater than the threshold in the previous generation weight data, and subtracts the predetermined value from the weight corresponding to data having the difference equal to or less than the threshold. Thus, current generation weight data may be generated.

  The weight storage unit 60 stores weight data for a plurality of generations updated by the weight generation unit 43 in each generation. Here, as an example, when the weight generation unit 43 updates the previous generation weight data to the current generation weight data, the weight storage unit 60 adds and stores the current generation weight data as new weight data. To do.

  The fitness calculation unit 44 weights the converted data obtained by converting the learning input data and the learning target data with the current generation weight data for each of the current generation converters 20 stored in the converter storage unit 34. Then, the current generation suitability according to the comparison result and the old generation suitability according to the comparison result obtained by weighting the converted data and the learning target data with the weight data before the previous generation are calculated. Here, the fitness is an index value indicating whether or not the learning input data is suitable for conversion into learning target data. The higher the value, the more the learning input data is converted into learning target data. It is suitable for In addition, for each current generation converter 20, the fitness calculation unit 44 weights the converted data and the learning target data with the weight data for each of the plurality of generations, and compares the plurality of generation data with a plurality of results. You may calculate the fitness by generation.

  The control unit 52 controls the number of generations of a plurality of generation suitability to be calculated by the suitability calculation unit 44 for the converter 20 of the current generation in accordance with the increment value of the current generation suitability between a plurality of generations. Hereinafter, this number of generations is referred to as the number of generations using weight data. As an example, the control unit 52 uses the maximum value of the fitness level of the converter 20 of each generation as the fitness level of each generation, and subtracts the fitness level of the previous generation from the fitness level of the current generation. It may be a value. For example, when the increment value of the current generation suitability is larger, the control unit 52 reduces the number of generations. Instead, the control unit 52 may control the number of generations using the weight data for the current generation converter 20 based on the increment value of the current generation suitability of each converter 20 and a preset reference value. Good. More specifically, when the increment value of the current generation suitability of each converter 20 is greater than or equal to the reference value, the control unit 52 uses a smaller number of generations than in the case of less than the reference value. .

  The control unit 52 may store the number of generations controlled in the previous generation. Then, the control unit 52 may control the number of generations using the weight data for the converter 20 of the current generation by increasing or decreasing the number of stored generations according to the increment value of the current generation suitability. In this case, as an example, the control unit 52 decreases the number of generations using the weight data when the average of the increment values of the current generation suitability of each converter 20 is equal to or greater than the reference value, and weights when the average value is less than the reference value. Increase the number of data generations. Note that the control unit 52 may use the increment value of the current generation suitability with respect to the average of the suitability of the plurality of generations immediately before the current generation as the increment value of the current generation suitability. An average of degree increment values or the like may be used.

  The selection unit 46 preferentially selects at least one of the converter 20 having a higher current generation suitability and the converter 20 having a higher old generation suitability among the plurality of converters 20 stored in the converter storage unit 34. The next-generation converter 20 is selected. More specifically, the selection unit 46 selects at least one converter 20 to be left by a technique that models natural selection of living things. For example, the selection unit 46 is based on each of the current generation suitability and the old generation suitability of the plurality of converters 20 stored in the converter storage unit 34, and performs at least one by genetic calculation such as elite selection and roulette selection. The converter 20 is selected.

  The update unit 48 leaves the converter 20 selected by the selection unit 46 among the plurality of converters 20 stored in the converter storage unit 34, and tricks the converter 20 not selected by the selection unit 46. . For example, the update unit 48 hesitates by deleting the converter 20 from the converter storage unit 34.

  The generation unit 50 performs genetic processing from at least one previous generation converter 20 including a plurality of processing components 22 that convert input data stored in the converter storage unit 34 into output data. Is generated. The generation unit 50 performs a genetic operation such as crossover and mutation on the at least one converter 20 remaining in the update process by the update unit 48 to generate a new converter 20.

  Such a genetic processing device 10 includes a conversion process by the conversion processing unit 38, a fitness calculation process by the fitness calculation unit 44, a converter 20 selection process by the selection unit 46, an update process and a generation unit by the update unit 48. The process of generating a new converter 20 by 50 is repeated a plurality of times (for example, a plurality of generations). Thereby, the genetic processing apparatus 10 can generate the converter 20 suitable for converting the learning input data into the learning target data.

  FIG. 2 shows an example of the converter 20 in which the processing components 22 according to this embodiment are combined in series. FIG. 3 shows an example of the converter 20 in which the processing component 22 according to this embodiment is combined in a tree structure.

  The converter 20 may have a configuration in which a plurality of processing components 22 are combined in series as shown in FIG. Further, the converter 20 may have a configuration in which a plurality of processing components 22 are combined in a tree structure as shown in FIG. Further, the converter 20 may have a configuration having a plurality of output terminals with respect to one input terminal. Further, the converter 20 may have a configuration having a plurality of input ends and a plurality of output ends.

  The converter 20 performs arithmetic processing on the received input data and outputs output data. As an example, the converter 20 is executed by a program that performs operations on input data. Further, the converter 20 may be an arithmetic expression representing the content of the operation to be performed on the input data.

  The converter 20 having a configuration in which the processing components 22 are combined in a tree structure receives input data from the processing components 22 at the end of the tree structure, and outputs output data from the highest processing component 22 in the tree structure. Further, in such a converter 20, the same input data is given to each of the plurality of terminal processing components 22. Alternatively, such a converter 20 may be provided with different input data for each of the plurality of terminal processing components 22.

  Each converter 20 may be an image filter in which at least one filter component that converts an input image into an output image is combined. In this case, each processing component 22 receives the image data output from the processing component 22 arranged in the preceding stage, performs an operation on the received image data, and gives it to the processing component 22 arranged in the subsequent stage. For example, the generation unit 50 generates a current-generation image filter by genetic processing from at least one previous-generation image filter that converts an input image into an output image. In this case, the genetic processing apparatus 10 evolves an image filter group including a plurality of image filters over a plurality of generations based on evolutionary calculation.

  Moreover, the converter 20 may be the structure which combined the processing component 22 which is hardware as an example. In this case, the converter storage part 34 memorize | stores the structure data which show the data transfer relationship between each process components as a data structure showing a converter. The converter 20 may have a configuration in which processing components 22 that are programs for performing operations on data are combined. Further, the converter 20 may have a configuration in which processing components 22 that are arithmetic expressions representing arithmetic contents to be applied to data are combined.

  Further, the converter 20 may convert, for example, a one-dimensional data string, a two-dimensional data group, a three-dimensional data group, or a further multidimensional data group. The one-dimensional data string is, for example, time series data or an array data string. The two-dimensional data group is, for example, image data in which a plurality of pixel data is arranged in a two-dimensional space. The three-dimensional data group is, for example, volume data in which data values representing color or density are arranged at each lattice point in the three-dimensional space. Further, the converter 20 may output data having a dimension different from the input data.

  Each of the plurality of processing components 22 receives input data output from the processing component 22 arranged in the previous stage, performs an operation on the received input data, and gives the processed data to the processing component 22 arranged in the subsequent stage. Each of the plurality of processing components 22 may be a program module, an arithmetic expression, or the like, and performs binarization operation, histogram operation, smoothing operation, edge detection operation, morphology operation, and / or frequency with respect to received input data. Unary operations such as space operations (for example, low-pass filtering operations and high-pass filtering operations) may be performed.

  Each of the plurality of processing components 22 performs an average operation, a difference operation, and / or a fuzzy operation (for example, an OR operation, an AND operation, an algebraic sum, an algebraic product, a limit sum, a limit product, Binomial operations such as intense sum and intense product) may be performed.

  FIG. 4 shows an example of a genetic operation performed on the converter 20 in which the processing components 22 according to this embodiment are combined in series. FIG. 5 shows an example of a crossover operation performed on the converter 20 in which the processing components 22 according to the present embodiment are combined in a tree structure. FIG. 6 shows an example of a mutation operation performed on the converter 20 in which the processing component 22 according to the present embodiment is combined in a tree structure.

  For example, the generation unit 50 performs a crossover operation, which is an example of a genetic operation, on two or more converters 20 to generate two or more new converters 20. As an example, as illustrated in FIGS. 4 and 5, the generation unit 50 converts a part group 24 </ b> A of at least one converter 20 </ b> A that has already been generated to another converter 20 </ b> B that has already been generated. New transducers 20E and 20F are generated by replacing at least a part group 24B. The component group 24 is a member obtained by combining one or a plurality of processing components 22.

  Further, as an example, the generation unit 50 performs a mutation operation, which is an example of a genetic operation, on a single converter 20 to generate a new single converter 20. As an example, as illustrated in FIG. 4 and FIG. 6, the generation unit 50 replaces a part group 24 </ b> C of a part of the already generated one converter 20 </ b> C with another part group 24 </ b> G selected at random. Thus, a new converter 20G is generated.

  The generation unit 50 may leave the current generation converter 20 as the next generation converter 20 as it is. As an example, as illustrated in FIG. 4, the generation unit 50 generates a next-generation converter 20H that includes the configuration of the processing component 22 of the converter 20D as it is.

  The selection unit 46 preferentially selects at least one of the converter 20 having a higher current generation suitability and the converter 20 having a higher old generation suitability from the one or more converters 20 generated by the generation unit 50. Choose to survive the fittest. The update unit 48 stores the selected converter 20 in the converter storage unit 34 and deletes the unselected converter 20 from the converter storage unit 34.

  FIG. 7 shows an example of a method for calculating the degree of approximation, which is an example of a parameter representing the degree of fitness of the converter 20 according to this embodiment. In the present embodiment, the fitness level calculation unit 44 calculates an approximation level indicating how much the converted data generated by converting the target data by the converter 20 approximates the learning target data. To do. That is, this degree of approximation is an index indicating that the converted data 20 is more appropriate because the converted data is more approximate to the learning target data as the value is larger. .

  As an example, the fitness calculation unit 44 compares the value of each pixel of the output image with the value of the corresponding pixel of the target image. More specifically, the fitness calculation unit 44 calculates, for example, a difference or a ratio between the value of each pixel of the output image and the value of the corresponding pixel of the target image. Then, as an example, the matching level calculation unit 44 may average or sum all the comparison results of a plurality of pixels, and may use the average or total comparison result as the approximation degree of the output image. The goodness-of-fit calculation unit 44 obtains the degree of approximation for each of the plurality of converted data converted by the plurality of converters 20 as described above.

  In this process, the fitness calculation unit 44 may obtain the degree of approximation using a weight image that represents the weight of the pixel position by the value of each pixel (for example, the luminance value). Specifically, the fitness calculation unit 44 weights the difference between the value of each pixel in the output image and the value of the corresponding pixel in the target image by multiplying the weight value of the corresponding pixel in the weighted image, for example. Thus, the degree of approximation may be calculated.

  For example, when the weight data is represented by a weight image, the fitness level calculation unit 44 calculates an approximation level by performing an operation represented by the following formula (1). In this case, as an example, the output image, the target image, and the weight image have the same size.

  In Expression (1), f represents the degree of approximation. Ymax represents the maximum luminance value.

  X is a variable indicating the pixel position in the horizontal direction of the image. y is a variable indicating the pixel position in the vertical direction of the image. xmax is a constant indicating the number of pixels in the horizontal direction of the image. tymax is a constant indicating the number of pixels in the vertical direction of the image.

  Iweight (x, y) represents the luminance value of the pixel at the (x, y) position in the weighted image. Itarget (x, y) represents the luminance value of the pixel at the (x, y) position in the target image. Ifilter (x, y) represents the luminance value of the pixel at the (x, y) position in the output image.

  That is, as shown in the numerator part in the curly braces of Equation (1), the fitness calculation unit 44 calculates the luminance of the weighted image with respect to the absolute value of the difference between the luminance value of the target image and the luminance value of the output image. A weighted difference value multiplied by the value is calculated for every pixel in the screen, and a total weighted difference value is calculated by adding the calculated weighted difference values for all pixels. Further, as shown in the denominator part in the curly braces of Expression (1), the fitness level calculation unit 44 calculates a total weight obtained by adding the luminance values of the weighted image for all pixels.

  Furthermore, the fitness calculation unit 44 normalizes the division value obtained by dividing the total weighted difference value by the total weight (in the curly braces of the expression (1)) by the reciprocal (1 / Ymax) of the maximum value of the luminance value. The value (the second term in equation (1)) is calculated. Then, the fitness level calculation unit 44 calculates a value obtained by subtracting the normalized value from 1 as the approximation level f. In this way, the fitness level calculation unit 44 can calculate the approximation level by weighting the difference between the output image and the target image for each region.

  FIG. 8 shows an example of an operation flow of the genetic processing apparatus 10 according to the present embodiment. The genetic processing device 10 repeatedly executes each process of step S13 to step S16 a plurality of times (for example, a plurality of generations) (S12, S18). Note that the genetic processing device 10 generates a randomly generated converter 20 or an already generated converter 20 as an initial converter 20 for generating the converter 20 that converts given learning input data into learning target data. The converter 20 is stored in the converter storage unit 34 in advance.

  First, the generation unit 50 acquires at least one converter 20 stored in the converter storage unit 34. Then, the generation unit 50 performs genetic operations such as a crossover operation and a mutation operation on the acquired at least one converter 20 to replace at least some of the processing components 22 with other processing components 22. At least one new converter 20 is generated (S17).

  In this case, the generation unit 50 performs a genetic operation so that each processing component 22 is introduced into the new converter 20 with a preset introduction probability. For example, when performing the crossover operation, the generation unit 50 selects one or a plurality of processing components 22 out of a plurality of processing components 22 that can be incorporated into the new converter 20 with a preset introduction probability. The part group 24 in the parent converter 20 is replaced with the part group 24 including the selected one or more processing parts 22. Then, the generation unit 50 stores the remaining converter 20 and the new converter 20 in the converter storage unit 34 as a plurality of converters 20 included in the next-generation converter group.

  Then, the conversion processing unit 38 generates converted data obtained by converting the input data for learning by the converter 20 for each of the plurality of converters 20 included in the current generation converter group in each generation ( S13). The conversion processing unit 38 stores the generated converted data in the converted data storage unit 40 in association with the converter 20.

  The weight generation unit 43 generates new weight data for the current generation from the weight data before the previous generation in order to generate appropriate weight data for each generation (S19). Details of the weight data generation processing will be described later.

  Subsequently, the fitness level calculation unit 44 calculates the fitness levels of the old generation and the current generation for each of the one or more converters 20 included in the converter group (S14). In this case, the fitness level calculation unit 44 calculates, as an example, an approximation level or a similarity level between the converted data and the learning target data as the fitness level. In addition to the degree of approximation, the degree-of-fit calculation unit 44 may calculate the degree of suitability that has been evaluated according to the small number of processing components, the low processing load, the degree of parallelism, and the like. .

  Subsequently, the selection unit 46 preferentially selects at least one of the converter 20 having the higher current generation suitability or the converter having the higher old generation suitability from among the plurality of converters 20 included in the converter group. (S15). As an example, the selection unit 46 selects at least one converter 20 whose current generation suitability or old generation suitability is higher than a reference value. Here, the selection part 46 is good also as a reference | standard that it is more than the threshold value of the preset fitness.

  In addition, the selection unit 46 may select a converter 20 in a range in which the current generation suitability or the old generation suitability is predetermined from the top among the plurality of converters 20 included in the converter group. Further, the selection unit 46 performs random conversion under the condition that the converter 20 having a higher fitness is selected with a higher probability among the current generation fitness and the previous generation fitness. The vessel 20 may be selected. As an example, the selection unit 46 may select the highest converter 20 from the current generation suitability in the last generation.

  In addition, the selection unit 46 sets the probability of selecting the converter 20 having the current generation fitness level equal to or higher than the reference value as the next generation converter 20, and selects the converter 20 having the old generation fitness level equal to or higher than the reference value as the next generation converter. It may be higher than the probability of selecting as 20. For example, the selection unit 46 may set the reference value of the fitness level for each generation, and may set the reference value of the old generation fitness level higher than the reference value of the current generation fitness level. In addition, the selection unit 46 may increase the reference value of the fitness level every time the generation becomes older. Thereby, the selection unit 46 can preferentially select the converter 20 having a higher current generation suitability with a higher probability.

  Subsequently, the update unit 48 updates the converter group stored in the converter storage unit 34 (S16). More specifically, the update unit 48 causes the converter 20 selected in step S15 among the plurality of converters 20 included in the current generation converter group to remain in the next generation, and replaces the other converters 20 with each other. The converter group is updated by hesitating. For example, the update unit 48 hesitates by deleting the converter 20 that was not selected in step S15 from the converter storage unit 34.

  The genetic processing apparatus 10 repeatedly executes the above processing for a plurality of generations (for example, tens of generations or hundreds of generations or more), and performs processing up to the last generation (for example, the Nth generation, where N is a natural number of 2 or more). After the execution, the flow is exited (S18). In this way, the genetic processing device 10 can generate the converter 20 suitable for converting the learning input data into the learning target data using the evolutionary calculation.

  FIG. 9 shows an example of the processing flow of the genetic processing apparatus 10 in step S14 of FIG. The genetic processing device 10 executes the following processing in step S14 shown in FIG.

  First, the fitness calculation unit 44 acquires the number of weight data use generations obtained in the previous generation from the control unit 52 (S21). Then, the fitness level calculation unit 44 executes the following steps S24 to S28 for each of the plurality of output data generated by each of the plurality of converters 20 of the current generation (S23, S29).

  In step S24, the goodness-of-fit calculation unit 44 compares the converted data and the learning target data with weights corresponding to the current generation weight data, and compares the converted data and the learning target data with the current generation suitability. Is calculated.

  Subsequently, the controller 52 controls the number of generations using the weight data based on the increment value of the current generation fitness calculated in step S24 and the number of generations acquired in step S21 (S25). ). For example, when the increment value of the current generation suitability of each converter 20 is equal to or larger than a preset reference value, the control unit 52 decreases the number of generations acquired in step S21. For example, when the increment value of the current generation suitability of each converter 20 is less than the preset reference value, the control unit 52 increases the number of generations acquired in step S21.

  Then, the control unit 52 updates and stores the weight data use generation number. Here, the control unit 52 may store the generation suitability in addition to the number of generations.

  Then, the fitness level calculation unit 44 executes the following step S27 for each previous generation for the number of generations calculated in step S25 (S26, S28).

  In step S27, the fitness level calculation unit 44 compares the converted data and the learning target data by weighting according to the weight data, and calculates the fitness level of the converted data and the learning target data. Here, as an example, the fitness calculation unit 44 performs weighting using weight data of the previous generation to be processed. When the genetic processing apparatus 10 finishes the process for all generations to be calculated, the process proceeds to step S29 (S28).

  Furthermore, when the genetic processing apparatus 10 finishes the processes of steps S26 to S28 for all of the plurality of output data, the genetic processing apparatus 10 ends the flow (S29). Thereby, the genetic processing apparatus 10 can select the converter 20 to be evolved in each generation by using one or a plurality of old generation fitnesses in addition to the current generation.

  FIG. 10 shows an example of the processing flow of the weight generation unit 43 in step S19 of FIG. The weight generation unit 43 performs the following processing of steps S131 to S133 in the weight data generation processing in step S19 shown in FIG. First, the weight generation unit 43 extracts at least one of the converted data output by the converter 20 for weight data generation (S131).

  As an example, the weight generation unit 43 may extract the converted data output by some of the converters 20. For example, the weight generation unit 43 may extract the converted data output from the converter 20 having the highest degree of approximation calculated in the previous generation among the plurality of converters 20. For example, the weight generation unit 43 may extract the converted data output by the converter 20 whose degree of approximation calculated in the previous generation is equal to or greater than a predetermined value.

  Subsequently, the weight generation unit 43 calculates a difference for each data of at least one data extracted for generating weight data in step S131 (S132). The weight generation unit 43 calculates a difference between data at corresponding positions in the converted data and the learning target data. Here, in the case of image data, for example, the weight generation unit 43 may calculate a difference for each pixel between the extracted at least one converted data and the target data.

  Further, when a plurality of sets of data are extracted in step S131, for example, the weight generation unit 43 sums or averages the difference between the plurality of sets of data for each data (for example, for each pixel as a region). Instead of this, the weight generation unit 43 may select, for each data, one largest difference among the differences between the plurality of sets of data.

  Subsequently, the weight generation unit 43 generates weight data for the current generation based on the difference for each data calculated in step S132 (S133). More specifically, the weight generation unit 43 updates the weight data stored in the weight storage unit 60 so that the weight of data with a larger difference is greater than the weight of data with a smaller difference. The weight storage unit 60 may store weight data generated in advance by the user in the first generation. Further, the weight storage unit 60 may store weight data in which the weight for each data is the same.

  In addition, the weight generation unit 43 calculates the difference for each data between the learning target data and each of the at least one converted data selected in descending order of approximation in each of at least one generation before the previous generation. Weight data for the current generation that is monotonically non-decreasing with respect to the total may be generated. Such weight generation unit 43 can generate appropriate weight data stably for a long period of time. For example, the weight generation unit 43 generates weight data based on an arithmetic expression represented by the following expression (2).

W _{N, M in} Equation (2) represents the value of the Mth (M is an integer greater than or equal to 1) weight data included in the Nth generation (N is an integer greater than or equal to 1) weight data, and W _{N + 1, M} Represents the value of the Mth weight data included in the (N + 1) th generation weight data. F (X) in Equation (2) represents a monotone non-decreasing function with X as a variable. That is, F (X) represents a value that does not decrease at least when X changes in the increasing direction. F (X) may be a monotonically increasing function. Note that F (X) and X are values in a range that can be taken by each data included in the data. Α in Expression (2) represents a reference value serving as a boundary for increasing or decreasing the weight according to F (D _{K, M} ). Note that α in Expression (2) may be a value that changes from generation to generation.

Further, D _{K, M} in the formula (2) is represented by the following formula (3). That is, _{DK, M} in the equation (2) is included in the Mth value (P _{K, M} ) included in the selected converted data in the Kth generation (K is an integer) and the learning target data. The Mth value (T _M ) and the absolute value of the difference are represented.

  As shown in such an equation (2), the weight generation unit 43 performs, for each generation, for each data between each of the at least one converted data selected in descending order of approximation and the learning target data. The difference may be calculated. Subsequently, the weight generation unit 43 may calculate the total of the calculated differences for each data, and divide each of the totals for each data by the number of generations of the previous generation. The weight generation unit 43 may generate current generation weight data including a value obtained by level-converting each division result for each data using a predetermined monotonous non-decreasing function.

  Note that the monotonic non-decreasing function F (X) in the equation (2) may be a function as shown in the following equation (4) as an example. In the function represented by the following equation (4), the change amount of the weight data is large in the range where the difference value is small, and the change amount of the weight data is small in the range where the difference value is large. By using such a function, the weight generation unit 43 can change the weight more sensitively in a region where the difference value is small.

  The monotonic non-decreasing function F (X) in Expression (2) is, for example, a function in which the change amount of the weight data is small in the range where the difference value is small and the change amount of the weight data is large in the range where the difference value is large. May be. By using such a function, the weight generation unit 43 can change the weight more sensitively in a region where the difference value is large.

For example, the weight generation unit 43 may generate weight data based on an arithmetic expression represented by the following expression (5). In the equation (5), W _K represents a weight given to the difference of the Kth generation. Parameters other than W _{K in} equation (5) are the same as in equation (2).

  As shown in Expression (5), the weight generation unit 43 calculates a difference for each data between each of the at least one converted data selected in descending order of approximation and the learning target data in each generation. You can do it. Subsequently, the weight generation unit 43 may calculate, for each data, a sum obtained by multiplying the calculated difference by a weight set in advance for each generation, and divide each of the totals for each data by the number of generations of the current generation. . The weight generation unit 43 may generate next-generation weight data including a value obtained by level-converting each division result for each data using a predetermined monotonous non-decreasing function.

  According to such a genetic processing apparatus 10 according to the present embodiment, it is possible to calculate the degree of fitness between the converted data and the learning target data by appropriately weighting each data with the weight data. Thereby, according to the genetic processing apparatus 10, the converted data approximated to the learning target data can be appropriately selected from the plurality of converted data.

  According to the genetic processing apparatus 10 shown above, by selecting the converters 20 using a plurality of generations of weight data, the influence of fluctuations in the weight data is reduced and the converter 20 is more appropriately used. It can evolve.

  FIG. 11 shows an example of a hardware configuration of a computer 1900 according to this embodiment. A computer 1900 according to this embodiment is connected to a CPU peripheral unit including a CPU 2000, a RAM 2020, a graphic controller 2075, and a display device 2080 that are connected to each other by a host controller 2082, and to the host controller 2082 by an input / output controller 2084. An input / output unit having a communication interface 2030, a hard disk drive 2040, and a DVD drive 2060, and a legacy input / output unit having a ROM 2010, a flexible disk drive 2050, and an input / output chip 2070 connected to the input / output controller 2084 Is provided.

  The host controller 2082 connects the RAM 2020 to the CPU 2000 and the graphic controller 2075 that access the RAM 2020 at a high transfer rate. The CPU 2000 operates based on programs stored in the ROM 2010 and the RAM 2020 and controls each unit. The graphic controller 2075 acquires image data generated by the CPU 2000 or the like on a frame buffer provided in the RAM 2020 and displays it on the display device 2080. Instead of this, the graphic controller 2075 may include a frame buffer for storing image data generated by the CPU 2000 or the like.

  The input / output controller 2084 connects the host controller 2082 to the communication interface 2030, the hard disk drive 2040, and the DVD drive 2060, which are relatively high-speed input / output devices. The communication interface 2030 communicates with other devices via a network. The hard disk drive 2040 stores programs and data used by the CPU 2000 in the computer 1900. The DVD drive 2060 reads a program or data from the DVD 2095 and provides it to the hard disk drive 2040 via the RAM 2020.

  The input / output controller 2084 is connected to the ROM 2010, the flexible disk drive 2050, and the relatively low-speed input / output device of the input / output chip 2070. The ROM 2010 stores a boot program that the computer 1900 executes at startup and / or a program that depends on the hardware of the computer 1900. The flexible disk drive 2050 reads a program or data from the flexible disk 2090 and provides it to the hard disk drive 2040 via the RAM 2020. The input / output chip 2070 connects the flexible disk drive 2050 to the input / output controller 2084 and inputs / outputs various input / output devices via, for example, a parallel port, a serial port, a keyboard port, a mouse port, and the like. Connect to controller 2084.

  A program provided to the hard disk drive 2040 via the RAM 2020 is stored in a recording medium such as the flexible disk 2090, the DVD 2095, or an IC card and provided by the user. The program is read from the recording medium, installed in the hard disk drive 2040 in the computer 1900 via the RAM 2020, and executed by the CPU 2000.

  A program installed in the computer 1900 and causing the computer 1900 to function as the genetic processing apparatus 10 includes a conversion processing module, a weight generation module, a weight storage module, a fitness calculation module, a control module, a selection module, and an update. A module and a generation module. These programs or modules work on the CPU 2000 and the like to make the computer 1900, the conversion processing unit 38, the weight generation unit 43, the weight storage unit 60, the fitness calculation unit 44, the control unit 52, and the selection unit 46. And the update unit 48 and the generation unit 50.

  The information processing described in these programs is read into the computer 1900, whereby the conversion processing unit 38, which is a specific means in which the software and the various hardware resources described above cooperate, the weight generation unit 43, , Functioning as a weight storage unit 60, a fitness calculation unit 44, a control unit 52, a selection unit 46, an update unit 48, and a generation unit 50. And the specific genetic processing apparatus 10 according to the intended purpose is constructed | assembled by implement | achieving the calculation or processing of the information according to the intended purpose of the computer 1900 in this embodiment by these specific means.

  As an example, when communication is performed between the computer 1900 and an external device or the like, the CPU 2000 executes a communication program loaded on the RAM 2020 and executes a communication interface based on the processing content described in the communication program. A communication process is instructed to 2030. Under the control of the CPU 2000, the communication interface 2030 reads transmission data stored in a transmission buffer area or the like provided on a storage device such as the RAM 2020, the hard disk drive 2040, the flexible disk 2090, or the DVD 2095, and transmits it to the network. Alternatively, the reception data received from the network is written into a reception buffer area or the like provided on the storage device. As described above, the communication interface 2030 may transfer transmission / reception data to / from the storage device by a DMA (direct memory access) method. Instead, the CPU 2000 transfers the storage device or the communication interface 2030 as a transfer source. The transmission / reception data may be transferred by reading the data from the data and writing the data to the transfer destination 2030 or the storage device.

  In addition, the CPU 2000 reads all or necessary portions from files or databases stored in an external storage device such as the hard disk drive 2040, the DVD drive 2060 (DVD 2095), and the flexible disk drive 2050 (flexible disk 2090). The data is read into the RAM 2020 by DMA transfer or the like, and various processes are performed on the data on the RAM 2020. Then, CPU 2000 writes the processed data back to the external storage device by DMA transfer or the like. In such processing, since the RAM 2020 can be regarded as temporarily holding the contents of the external storage device, in the present embodiment, the RAM 2020 and the external storage device are collectively referred to as a memory, a storage unit, or a storage device. Various types of information such as various programs, data, tables, and databases in the present embodiment are stored on such a storage device and are subjected to information processing. Note that the CPU 2000 can also hold a part of the RAM 2020 in the cache memory and perform reading and writing on the cache memory. Even in such a form, the cache memory bears a part of the function of the RAM 2020. Therefore, in the present embodiment, the cache memory is also included in the RAM 2020, the memory, and / or the storage device unless otherwise indicated. To do.

  In addition, the CPU 2000 performs various operations, such as various operations, information processing, condition determination, information search / replacement, etc., described in the present embodiment, specified for the data read from the RAM 2020 by the instruction sequence of the program. Is written back to the RAM 2020. For example, when performing the condition determination, the CPU 2000 determines whether or not the various variables shown in the present embodiment satisfy the conditions such as large, small, above, below, equal, etc., compared to other variables or constants. If the condition is satisfied (or not satisfied), the program branches to a different instruction sequence or calls a subroutine.

  Further, the CPU 2000 can search for information stored in a file or database in the storage device. For example, in the case where a plurality of entries in which the attribute value of the second attribute is associated with the attribute value of the first attribute are stored in the storage device, the CPU 2000 displays the plurality of entries stored in the storage device. The entry that matches the condition in which the attribute value of the first attribute is specified is retrieved, and the attribute value of the second attribute that is stored in the entry is read, thereby associating with the first attribute that satisfies the predetermined condition The attribute value of the specified second attribute can be obtained.

  The program or module shown above may be stored in an external recording medium. As the recording medium, in addition to the flexible disk 2090 and the DVD 2095, an optical recording medium such as DVD or CD, a magneto-optical recording medium such as MO, a tape medium, a semiconductor memory such as an IC card, and the like can be used. Further, a storage device such as a hard disk or RAM provided in a server system connected to a dedicated communication network or the Internet may be used as a recording medium, and the program may be provided to the computer 1900 via the network.

  As mentioned above, although this invention was demonstrated using embodiment, the technical scope of this invention is not limited to the range as described in the said embodiment. It will be apparent to those skilled in the art that various modifications or improvements can be added to the above-described embodiment. It is apparent from the scope of the claims that the embodiments added with such changes or improvements can be included in the technical scope of the present invention.

  The order of execution of each process such as operations, procedures, steps, and stages in the apparatus, system, program, and method shown in the claims, the description, and the drawings is particularly “before” or “prior to”. It should be noted that the output can be realized in any order unless the output of the previous process is used in the subsequent process. Regarding the operation flow in the claims, the description, and the drawings, even if it is described using “first”, “next”, etc. for convenience, it means that it is essential to carry out in this order. It is not a thing.

DESCRIPTION OF SYMBOLS 10 Genetic processing apparatus, 20 Converter, 22 Processing components, 24 Parts group, 34 Converter storage part, 36 Learning input data storage part, 38 Conversion processing part, 40 Converted data storage part, 42 Learning target data storage Unit, 43 weight generation unit, 44 fitness calculation unit, 46 selection unit, 48 update unit, 50 generation unit, 52 control unit, 60 weight storage unit, 1900 computer, 2000 CPU, 2010 ROM, 2020 RAM, 2030 communication interface, 2040 hard disk drive, 2050 flexible disk drive, 2060 DVD drive, 2070 input / output chip, 2075 graphic controller, 2080 display device, 2082 host controller, 2084 input / output controller, 2090 flexible disk Click, 2095 DVD

Claims (8)

A generator that generates a current generation converter by genetic processing from at least one previous generation converter that converts input data into output data;
A weight generation unit that updates the weight data of the previous generation to the weight data of the current generation;
For each current generation converter, the current generation suitability according to the result of weighted comparison of the converted data obtained by converting the learning input data and the learning target data with the weight data of the current generation, and A fitness calculation unit that calculates the previous generation fitness according to the result of comparing the converted data and the learning target data with weight data before and after the previous generation;
A genetic processing apparatus comprising: The genetic according to claim 1, further comprising: a selection unit that preferentially selects at least one of the converter having a higher current generation fitness and the converter having a higher older generation fitness as a next-generation transducer. Processing equipment. The selection unit sets a probability of selecting a converter having the current generation fitness level equal to or higher than a reference value as the next generation converter, and sets a converter having the old generation fitness level equal to or higher than the reference value to the next generation converter. The genetic processing device according to claim 2, wherein the genetic processing device is higher than a probability of selecting as. A weight storage unit that stores a plurality of generational weight data updated by the weight generation unit in each generation;
For each of the current generation converters, the fitness calculator calculates a plurality of weights according to a result of comparing the converted data and the learning target data by weighting each of the plurality of generation weight data. The genetic processing device according to any one of claims 1 to 3, wherein the fitness for generation is calculated. A controller for controlling the number of generations of the plurality of generation suitability to be calculated for the converter of the current generation by the suitability calculating unit according to an increment value of the current generation suitability between a plurality of generations; The genetic processing apparatus according to claim 4. The generation unit generates an image filter of the current generation by genetic processing from at least one image filter of the previous generation, using the image filter that converts an input image into an output image as the converter. The genetic processing device according to 1. Generating a current generation converter by genetic processing from at least one previous generation converter for converting input data to output data;
A weight generation step of updating the weight data before the previous generation to the weight data of the current generation;
For each current generation converter, the current generation suitability according to the result of weighted comparison of the converted data obtained by converting the learning input data and the learning target data with the weight data of the current generation, and A fitness calculation step of calculating an old generation fitness according to a result of comparing the converted data and the learning target data by weighting the weight data of the previous generation or earlier; and
A genetic processing method comprising: A genetic processing program for causing a computer to function as a genetic processing device,
The computer,
A generator that generates a current generation converter by genetic processing from at least one previous generation converter that converts input data into output data;
A weight generation unit that updates the weight data of the previous generation to the weight data of the current generation;
For each current generation converter, the current generation suitability according to the result of weighted comparison of the converted data obtained by converting the learning input data and the learning target data with the weight data of the current generation, and A fitness calculation unit that calculates the previous generation fitness according to the result of comparing the converted data and the learning target data with weight data before and after the previous generation;
A genetic processing program that functions as a genetic processing device.

Images