Data processing device, data processing system, and data processing method

Abstract

To provide a data processing device, a data processing system, and a data processing method capable of highly accurately processing data.SOLUTION: According to an embodiment, a data processing device includes a processing unit. The processing unit can execute a first evaluation index derivation operation which derives a first evaluation index from a plurality of first regression labels and a plurality of first composite regression labels. The plurality of first regression labels is derived from a plurality of first machine learning models. The plurality of first machine learning models is derived from a plurality of first sample data. The plurality of first composite regression labels is derived from a plurality of first composite machine learning models. The plurality of first composite machine learning models is derived from the plurality of first sample data and first acquisition data by first transfer learning. The plurality of first sample data is derived from first other data or first conversion other data obtainable by converting the first other data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] Embodiments of the present invention relate to a data processing device, a data processing system, and a data processing method. [Background technology]

[0002] For example, data from various electronic devices such as magnetic recording and playback devices is processed. For example, machine learning is performed through data processing. High-precision data processing is desired. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Daume'III, H., "Frustratingly Easy Domain Adaptation," Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics, pp. 256-263, June 2007. [Overview of the project] [Problems that the invention aims to solve]

[0004] Embodiments of the present invention provide a data processing device, a data processing system, and a data processing method capable of high-precision data processing. [Means for solving the problem]

[0005] According to embodiments of the present invention, the data processing device includes an acquisition unit and a processing unit. The acquisition unit is capable of acquiring first acquired data and first other data. The processing unit is capable of performing a first evaluation index derivation operation that derives a first evaluation index from a plurality of first regression labels and a plurality of first composite regression labels. The plurality of first regression labels are derived from a plurality of first machine learning models. The plurality of first machine learning models are derived from a plurality of first sample data. The plurality of first composite regression labels are derived from a plurality of first composite machine learning models. The plurality of first composite machine learning models are derived from the plurality of first sample data and the first acquired data by first transfer learning. The plurality of first sample data are derived from the first other data, or first transformed other data obtained by transforming the first other data. [Brief explanation of the drawing]

[0006] [Figure 1] Figure 1 is a schematic diagram illustrating the operation of a data processing device according to the first embodiment. [Figure 2] Figure 2 is a schematic diagram illustrating a data processing device according to the first embodiment. [Figure 3] Figures 3(a) to 3(c) are schematic diagrams illustrating the operation of the data processing device according to the first embodiment. [Figure 4] Figure 4 is a schematic diagram illustrating the operation of the data processing device according to the first embodiment. [Figure 5] Figure 5 is a schematic diagram illustrating the operation of the data processing device according to the first embodiment. [Figure 6] Figure 6 is a schematic diagram illustrating the operation of the data processing device according to the first embodiment. [Figure 7] Figure 7 is a schematic diagram illustrating the operation of the data processing device according to the first embodiment. [Figure 8] Figure 8 is a schematic diagram illustrating the operation of the data processing device according to the first embodiment. [Figure 9] Figure 9 is a schematic diagram illustrating the operation of the data processing device according to the first embodiment. [Figure 10]Figure 10 is a schematic diagram illustrating the operation of a data processing device according to the first embodiment. [Figure 11] Figures 11(a) and 11(b) are schematic diagrams illustrating the operation of the data processing device according to the first embodiment. [Figure 12] Figure 12 is a schematic diagram illustrating the operation of a data processing device according to the first embodiment. [Figure 13] Figure 13 is a schematic diagram illustrating the operation of a data processing device according to the first embodiment. [Figure 14] Figure 14 is a schematic diagram illustrating the operation of a data processing device according to the first embodiment. [Figure 15] Figure 15 is a schematic diagram illustrating the operation of a data processing device according to the first embodiment. [Figure 16] Figure 16 is a schematic diagram illustrating a data processing device according to an embodiment. [Modes for carrying out the invention]

[0007] Embodiments of the present invention will be described below with reference to the drawings. The drawings are schematic or conceptual. In this specification and in each drawing, elements similar to those described above in previous drawings are denoted by the same reference numerals, and detailed explanations are omitted where appropriate.

[0008] (First Embodiment) Figure 1 is a schematic diagram illustrating the operation of a data processing device according to the first embodiment. Figure 2 is a schematic diagram illustrating a data processing device according to the first embodiment. As shown in FIG. 2, the data processing apparatus 110 according to the embodiment includes a processing unit 71. The processing unit 71 can acquire various types of data 10D. For example, the data processing apparatus 110 may include an acquisition unit 72. The acquisition unit 72 may acquire various types of data 10D, and the data 10D acquired by the acquisition unit 72 may be supplied to the processing unit 71. The data processing apparatus 110 may include a storage unit 73. The data 10D acquired by the acquisition unit 72 may be stored in the storage unit 73. The processing unit 71 may acquire the data 10D stored in the storage unit 73 from the storage unit 73.

[0009] The acquisition unit 72 is, for example, an interface. The acquisition unit 72 may be, for example, an interface for input and output. The processing unit 71 may output information I1 regarding the processed result. The information I1 may be output via the acquisition unit 72 (interface). The processing unit 71 may be communicable with a server 74. The communication may include at least one of providing information and acquiring information. The communication may be based on any method of wired or wireless.

[0010] The data 10D may include, for example, acquired data (such as first acquired data 11 and second acquired data 12, etc.). The data 10D may include other data (such as first other data 51 and second other data 52, etc.).

[0011] As will be described later, the processing unit 71 can generate various types of data (such as first generated data 21, second generated data, first machine learning model 31, and second machine learning model 32, etc.). Each of the generated various types of data may be stored in the storage unit 73.

[0012] For example, the memory unit 73 may include a first memory area 73a and a second memory area 73b. For example, the first acquired data 11 and the first other data 51 may be stored in the first memory area 73a. The processing unit 71 may acquire the first acquired data 11 and the first other data 51 from the first memory area 73a. For example, the second acquired data 12 and the second other data 52 may be stored in the second memory area 73b. The processing unit 71 may acquire the second acquired data 12 and the second other data 52 from the second memory area 73b.

[0013] Figure 1 illustrates the first evaluation index derivation operation PP1 performed by the processing unit 71. In the first evaluation index derivation operation PP1, the processing unit 71 derives the first evaluation index 31P from a plurality of first regression labels 61b and a plurality of first composite regression labels 61Sb.

[0014] Multiple first regression labels 61b are derived from multiple first machine learning models 31. Multiple first machine learning models 31 are derived from multiple first sample data 41. Multiple first composite regression labels 61Sb are derived from multiple first composite machine learning models 31S. Multiple first composite machine learning models 31S are derived from multiple first sample data 41 and first acquired data 11 by first transfer learning 31L. Multiple first sample data 41 are derived from first other data 51, or first transformed other data 51C obtained by transforming the first other data 51. An example of the derivation (transformation) of the first transformed other data 51C will be described later.

[0015] The following describes an example of how the first evaluation index 31P is derived. As already explained, the acquisition unit 72 can acquire the first acquisition data 11 and the first other data 51. These data are supplied to the processing unit 71.

[0016] In one example, the first acquired data 11 is small-scale data relating to the target device. The first other data 51 is large-scale data relating to at least one of the target device and devices similar to it. For example, the first acquired data 11 relates to evaluation data relating to a small-scale experiment. For example, the first other data 51 relates to evaluation data relating to mass-produced products.

[0017] In one example, the device in question is a magnetic memory Record This is a playback device. The first acquired data 11 concerns data related to a prototype of the magnetic recording playback device. The first other data 51 concerns data related to a mass-produced magnetic recording playback device.

[0018] The first acquired data 11 is, for example, target data. The first other data 51 is, for example, source data. For example, source data contains a sufficient amount of information (e.g., a large amount of information). For example, target data does not contain a sufficient amount of information. Target data contains, for example, a small amount of information. In transfer learning, by applying knowledge obtained from source data containing a sufficient amount of information, a regression model that works with high accuracy on target data that does not contain a sufficient amount of information is generated.

[0019] The first acquired data 11 includes, for example, a first feature matrix 11a with N1 rows and D1 columns, and a first acquired label 11b with N1 rows. The first other data 51 includes a first other feature matrix 51a with Np rows and D1 columns, and a first other label 51b with Np rows. "N1" is an integer greater than or equal to 2. "Np" is an integer greater than or equal to 2. "D1" is an integer greater than or equal to 1. "N1" is less than "Np".

[0020] One of the multiple first sample data 41 contains a first sample feature matrix 41a with Ns rows and D1 column, and a first sample label 41b with Ns rows, where "Ns" is an integer greater than or equal to 2, and "Ns" is less than "Np".

[0021] For example, a portion of the first other data 51 is extracted to obtain one of multiple first sample data 41. For example, 1,000 data points are extracted from 8,000 first other data 51 to become one of multiple first sample data 41. For example, 1,000 data points are extracted from 8,000 first other data 51 to become another one of multiple first sample data 41. The number of multiple first sample data 41 is 200. A portion of another one of the multiple first sample data 41 may be the same as a portion of one of the multiple first sample data 41.

[0022] The derivation (i.e., extraction) of multiple first sample data 41 may be performed, for example, by bootstrap sampling. The processing unit 71 can derive multiple first sample data 41 from the first other data 51 or the first transformed other data 51C by bootstrap sampling.

[0023] The processing unit 71 derives multiple first synthetic machine learning models 31S from multiple first sample data 41 and first acquired data 11 by first transfer learning 31L. In one example, multiple first synthetic data 31SD may be used in the first transfer learning 31L. The first transfer learning 31L may include arbitrary transfer learning. An example of the first transfer learning 31L will be described later.

[0024] The processing unit 71 derives multiple first synthetic regression labels 61Sb from multiple first synthetic machine learning models 31S.

[0025] Meanwhile, the processing unit 71 derives multiple first machine learning models 31 from multiple first sample data 41. The processing unit 71 derives multiple first regression labels 61b from the multiple first machine learning models 31.

[0026] From the multiple first regression labels 61b and multiple first composite regression labels 61Sb derived as described above, the processing unit 71 derives the first evaluation index 31P.

[0027] The first evaluation metric 31P is based on the difference between multiple primary regression labels 61b and multiple primary composite regression labels 61Sb.

[0028] As described above, in this embodiment, multiple first regression labels 61b and multiple first composite regression labels 61Sb are derived from the first acquired data 11 (small-scale data) and the first other data 51 (large-scale data). Based on the multiple first regression labels 61b and multiple first composite regression labels 61Sb, a first evaluation index 31P is derived. The first evaluation index 31P corresponds to the "likelihood" of the derived multiple first composite regression labels 61Sb.

[0029] For example, a composite regression label for the first acquired data 11 (small data) is derived through transfer learning based on the first external data 51 (large data). In this case, for example, if the characteristics of the first external data 51 (large data) match the characteristics of the first acquired data 11 to a high degree, the value of the first evaluation index 31P (e.g., difference) will be small. For example, if the characteristics of the first external data 51 (large data) match the characteristics of the first acquired data 11 to a low degree, the value of the first evaluation index 31P (e.g., difference) will be large.

[0030] In this embodiment, the "likelihood" of the synthetic regression labels derived by transfer learning can be determined. High-precision "likelihood" enables high-precision data processing. According to this embodiment, a data processing device capable of high-precision data processing can be provided.

[0031] In this embodiment, for example, multiple first sample data 41 are derived from the first other data 51 (large-scale data). By using the multiple first sample data 41, multiple first regression labels 61b and multiple first composite regression labels 61Sb can be derived. There is variability in the differences between these values. The first evaluation index 31P is used as an evaluation index that corresponds to the variability in the differences. This allows us to understand the "likelihood" of the result.

[0032] Figures 3(a) to 3(c) are schematic diagrams illustrating the operation of the data processing device according to the first embodiment. Figures 3(a) to 3(c) show the target device being magnetically recorded. Record This corresponds to the case of a playback device. Below, as an example of a quantity corresponding to the first feature matrix 11a, magnetic memory Record The erase width in the playback device is adopted. The quantity corresponding to the first feature matrix 11a is magnetic memory Record Other quantities related to the playback device may also be used. Below, the data for quantities corresponding to the first feature matrix 11a (e.g., erasure width) will be explained in a simplified manner using a model.

[0033] The horizontal axis in Figure 3(a) represents the elimination width EW. The elimination width EW corresponds, for example, to the first feature matrix 11a. Let "S" be one of the multiple first regression labels 61b. Let "t" be one of the multiple first composite regression labels 61Sb. s The vertical axis in Figure 3(a) represents the difference (t) between one of the multiple first regression labels 61b and one of the multiple first composite regression labels 61Sb. s -S)

[0034] The erasure width EW is width EW1 to width EW k It can vary between these values. "k" is an integer greater than or equal to 1. Figure 3(b) corresponds to the characteristics when the erasure width EW is width EW1. Figure 3(c) corresponds to when the erasure width EW is width EW k This corresponds to the characteristics at that time. The horizontal axis in Figures 3(b) and 3(c) is the difference (t s -S). The vertical axis in Figures 3(b) and 3(c) is frequency Pg. Figures 3(b) and 3(c) show the difference (t s This corresponds to the histogram of -S). Figures 3(b) and 3(c) show the difference (t s This corresponds to the distribution of -S). In Figures 3(a) to 3(c), the data in the data processing device 110 is illustrated in a model manner. For example, the shape of the histogram (distribution) shown in Figures 3(b) and 3(c) is an example and may differ from the histogram of the actual data.

[0035] As shown in Figures 3(b) and 3(c), the difference (t sThe histogram (distribution) of (-S) changes according to the erasure width EW. As shown in FIGS. 3(b) and 3(c), in the histogram (distribution), the half-value width (for example, the half-value half-width) is used as the index HW. The first evaluation index 31P may be, for example, the average of the index HW.

[0036] Let the "i"-th index HW be "HW i ". The "HW" which is the first evaluation index 31P is represented by the following first formula. [Number]

[0037] In the first formula, "n" is the number of types of erasure width EW. As shown in the first formula, the first evaluation index 31P corresponds to the average of the half-value width (index HW) of the histogram of the difference (ts - S) between one of the plurality of first regression labels 61b and one of the plurality of first composite regression labels 61Sb. Thus, the first evaluation index 31P corresponds to the average regarding the distribution of the difference between one of the plurality of first regression labels 61b and one of the plurality of first composite regression labels 61Sb.

[0038] For example, when the index HW is small, the "likelihood" is high. When the index HW is small, the "likelihood" is low. By using such a first evaluation index 31P, the "likelihood" of the derived plurality of first composite regression labels 61Sb can be known.

[0039] Hereinafter, an example of the first transfer learning 31L will be described. Hereinafter, an example of the first transfer learning 31L based on one of the plurality of first sample data 41 will be described.

[0040] FIG. 4 is a schematic diagram illustrating the operation of the data processing apparatus according to the first embodiment. As shown in FIG. 4, as already described, one of the plurality of first sample data 41 includes a first sample feature matrix 41a of Ns rows and D1 columns and a first sample label 41b of Ns rows.

[0041] In the first operation OP1, the processing unit 71 can generate one of several first synthetic machine learning models 31S based on the first acquired data 11 and the first generated data 21 based on one of several first sample data 41. The first generated data 21 is, for example, one example of several first synthetic data 31SD (see Figure 1).

[0042] As shown in Figure 4, the first generated data 21 includes a first generated matrix 21a and a first generated label 21b. The first generated matrix 21a is a matrix with (Ns+N1) rows and (3×D1) columns. The first generated label 21b is data with (Ns+N1) rows. The number of columns in the first generated label 21b is, for example, 1.

[0043] As shown in Figure 4, the first generator matrix 21a includes the first matrix data M1, the second matrix data M2, and the third matrix data M3.

[0044] The components of the first matrix data M1 include the row-direction combination of the first sample feature matrix 41a and the first feature matrix 11a.

[0045] The components of the second matrix data M2 include the matrix Mxa1 (i.e., the 0 matrix) of the 0 component in row Ns and column D1, and the first feature matrix 11a, with a row-direction concatenation between them.

[0046] The components of the third matrix data M3 include the row-direction combination of the first sample feature matrix 41a and the matrix Mxb1 (i.e., the 0 matrix) which has 0 components in row N1 and column D1.

[0047] The components of the first generated label 21b include a row-direction combination of the first sample label 41b and the first acquired label 11b.

[0048] For example, (Ns+N1) / D1 may be 250 or greater. Let (Ns+N1) / D1 be the first ratio R1. The first ratio R1 may be 500 or greater.

[0049] In this embodiment, such first generated data 21 is generated. Based on the first generated data 21, a first synthetic machine learning model 31S is generated. The first synthetic machine learning model 31S enables high-precision data processing.

[0050] Thus, transfer learning is performed in this embodiment. In the transfer learning according to this embodiment, for example, first generated data 21 is derived by combining first acquired data 11 (target data) with one of a plurality of first sample data 41. A machine learning model based on such first generated data 21 is used. This results in higher accuracy than in the first reference example in which a machine learning model based only on the target data is used.

[0051] Figures 5 to 7 are schematic diagrams illustrating the operation of the data processing device according to the first embodiment. Figure 5 illustrates one of several first sample data 41. Figure 6 illustrates the first acquired data 11. Figure 7 illustrates the first generated data 21.

[0052] As shown in Figure 5, one of the multiple first sample data 41 contains the first sample label 41b in Ns rows. The first sample label 41b includes, for example, ys_1;ys_2;…ys_Ns. One of the multiple first sample data 41 contains the first sample feature matrix 41a in Ns rows and D1 column. The first row of the first sample feature matrix 41a includes, for example, xs_1,1;xs_1,2;…xs_1,D1. The second row of the first sample feature matrix 41a includes, for example, xs_2,1;xs_2,2;…xs_2,D1. The Nsth row of the first sample feature matrix 41a includes, for example, xs_Ns,1;xs_Ns,2;…xp_Ns,D1.

[0053] As shown in Figure 6, the first acquired data 11 includes the first acquired labels 11b in N1 rows. The first acquired labels 11b include, for example, y1_1;y1_2;…y1_N1. The first acquired data 11 includes the first feature matrix 11a in N1 rows and D1 columns. The first row of the first feature matrix 11a includes, for example, x1_1,1;x1_1,2;…x1_1,D1. The second row of the first feature matrix 11a includes, for example, x1_2,1;x1_2,2;…x1_2,D1. The N1st row of the first feature matrix 11a includes, for example, x1_N1,1;x1_N1,2;…x1_N1,D1.

[0054] As shown in Figure 7, the first generated data 21 contains the first generated label 21b for (Ns+N1) rows. The first generated label 21b contains, for example, ys_1;ys_2;…ys_Ns;y1_1;y1_2;…y1_N1.

[0055] The first generated data 21 includes the first generated matrix 21a, which has (Ns+N1) rows and (3×D1) columns.

[0056] The first row of the first generator matrix 21a contains, for example, xs_1,1;xs_1,2;…xs_1,D1;D1 zeros;xs_1,1;xs_1,2;…xs_1,D1. The second row of the first generator matrix 21a contains, for example, xs_2,1;xs_2,2;…xs_2,D1;D1 zeros;xs_2,1;xs_2,2;…xs_2,D1. The Ns row of the first generator matrix 21a contains, for example, xs_Ns,1;xs_Ns,2;…xs_Ns,D1;D1 zeros;xs_Ns,1;xs_Ns,2;…xs_Ns,D1.

[0057] The (Ns+1)th row of the first generator matrix 21a contains, for example, x1_1,1;x1_1,2;…x1_1,D1;x1_1,1;x1_1,2;…x1_1,D1;D1. The (Ns+2)th row of the first generator matrix 21a contains, for example, x1_2,1;x1_2,2;…x1_2,D1;x1_2,1;x1_2,2;…x1_2,D1;D1. The (Ns+N1)th row of the first generator matrix 21a contains, for example, x1_N1,1;x1_N1,2;…x1_N1,D1;x1_N1,1;x1_N1,2;…x1_N1,D1;D1.

[0058] Based on this first generated data 21, a first synthetic machine learning model 31S is generated. For example, the processing unit 71 generates the first synthetic machine learning model 31S using at least one selected from the group consisting of kernel regression, linear regression, Ridge regression, Lasso regression, Elastic Net, gradient boosting regression, random forest regression, k-nearest neighbor regression, and logistic regression.

[0059] Kernel regression may include, for example, at least one of Gaussian process regression and SVR (Support Vector Regression).

[0060] The processing unit 71 may also perform the following processes in the first operation OP1. Figure 8 is a schematic diagram illustrating the operation of the data processing device according to the first embodiment. As shown in Figure 8, in the first operation OP1, the processing unit 71 can input the first composite regression matrix 61Sa into the first composite machine learning model 31S to further derive the first composite regression label 61Sb.

[0061] The first composite regression matrix 61Sa has N1 rows (3 × D1) and 1 column. The first composite regression matrix 61Sa contains the first composite regression matrix data K1, the second composite regression matrix data K2, and the third composite regression matrix data K3. The components of the first composite regression matrix data K1 include the first feature matrix 11a. The components of the second composite regression matrix data K2 include the first feature matrix 11a. The components of the third composite regression matrix data K3 include the matrix Mxc1 (i.e., the 0 matrix) with N1 rows and 1 column and the component 0. The derived first composite regression label 61Sb has N1 rows.

[0062] The first composite regression label 61Sb obtained in this way has high accuracy. For example, in the first reference example above, a machine learning model based on the first acquired data 11 (target data) is used. In the first reference example, the accuracy of the regression label obtained using this machine learning model is low. In the embodiment, a first composite regression label 61Sb with higher accuracy than in the first reference example is obtained.

[0063] Figure 9 is a schematic diagram illustrating the operation of the data processing device according to the first embodiment. Figure 9 illustrates the first composite regression matrix 61Sa. The first row of the first composite regression matrix 61Sa contains, for example, x1_1,1;x1_1,2;…x1_1,D1;x1_1,1;x1_1,2;…x1_1,D1;D1 zeros. The second row of the first composite regression matrix 61Sa contains, for example, x1_2,1;x1_2,2;…x1_2,D1;x1_2,1;x1_2,2;…x1_2,D1;D1 zeros. synthesis The N1th row of the regression matrix 61a contains, for example, x1_N1,1;x1_N1,2;…x1_N1,D1;x1_N1,1;x1_N1,2;…x1_N1,D1;D1 zeros. Such a first synthetic regression matrix 61Sa is input to the first synthetic machine learning model 31S to obtain the first synthetic regression label 61Sb.

[0064] Figure 10 is a schematic diagram illustrating the operation of a data processing device according to the first embodiment. Figure 10 illustrates a first composite regression label 61Sb. As shown in Figure 10, the first composite regression label 61Sb includes, for example, z1_1;z1_2;…z1_N1. Such a first composite regression label 61Sb has high accuracy.

[0065] The operations described in Figures 4 to 10 may be performed on another of the multiple first sample data 41. This will derive multiple first composite regression labels 61Sb.

[0066] In this embodiment, the processing unit 71 may further derive an error index based on the first evaluation index 31P. The error index corresponds to an estimated value of the error to the "more reliable trend" when the first acquired data 11 (small-scale data) can be acquired as large-scale data.

[0067] The memory unit 73 may be capable of storing at least one of the first acquired data 11, the first other data 51, and the first evaluation index 31P.

[0068] The memory unit 73 may be capable of storing at least one of the following: multiple first regression labels 61b, multiple first composite regression labels 61Sb, multiple first machine learning models 31, multiple first sample data 41, multiple first composite machine learning models 31S, and first transformed other data 51C.

[0069] The following describes an example of the first transformation data 51C. Figures 11(a) and 11(b) are schematic diagrams illustrating the operation of the data processing device according to the first embodiment. These diagrams show the magnetic recording of the target equipment. Record This corresponds to the case of a playback device. Figure 11(a) corresponds to the first other data 51. Figure 11(b) corresponds to the first transformed other data 51C. The horizontal axis of these figures is the quantity Va corresponding to the first feature matrix 11a. The quantity Va may also be the erasure width EW. The vertical axis of these figures is the frequency Pg.

[0070] As shown in Figure 11(a), the distribution of quantity Va in the first other data 51 is curved. By transforming this first other data 51, the first transformed other data shown in Figure 11(b) is obtained. In the first transformed other data 51C, the frequency Pg is homogenized within the range of the target quantity Va. For example, in the first other data 51 exemplified in Figure 11(a), some of the data relating to quantity Va with a high frequency Pg is rejected.

[0071] For example, the change in frequency Pg relative to the first feature matrix 11a of the first transformed other data 51C is smaller than the change in frequency Pg relative to the first feature matrix 11a of the first acquired data 11. By using the first transformed other data 51C transformed in this way, for example, multiple first synthetic machine learning models 31S with higher accuracy can be obtained.

[0072] As described above, in the first evaluation index derivation operation PP1, the processing unit 71 derives the first evaluation index 31P from the first other data 51 and the first acquired data 11. In this embodiment, the processing unit 71 may also perform the second evaluation index derivation operation PP2 described below.

[0073] Figure 12 is a schematic diagram illustrating the operation of a data processing device according to the first embodiment. As shown in Figure 12, the acquisition unit 72 can further acquire the second other data 52. For example, as explained with respect to Figure 2, the acquisition unit 72 may acquire the second other data 52. The second other data 52 acquired by the acquisition unit 72 is supplied to the processing unit 71.

[0074] As shown in Figure 12, in the second evaluation index derivation operation PP2, the processing unit 71 derives the second evaluation index 32P from a plurality of second regression labels 62b and a plurality of second composite regression labels 62Sb.

[0075] Multiple second regression labels 62b are derived from multiple second machine learning models 32. Multiple second machine learning models 32 are derived from multiple second sample data 42. Multiple second composite regression labels 62Sb are derived from multiple second composite machine learning models 32S. Multiple second composite machine learning models 32S are derived from multiple second sample data 42 and first acquired data 11 by second transfer learning 32L. In one example, multiple second composite data 32SD may be used in the second transfer learning 32L. Multiple second sample data 42 are derived from second other data 52, or second transformed other data 52C obtained by transforming the second other data 52.

[0076] In the second evaluation index derivation operation PP2, the same processing as in the first evaluation index derivation operation PP1 may be performed. The second evaluation index 32P is derived from the second other data 52, which is different from the first other data 51.

[0077] The processing unit 71 may perform a designation operation to select one of the first other data 51 and the second other data 52 based on the result of comparing the first evaluation index 31P and the second evaluation index 32P.

[0078] For example, if the first evaluation index 31P is smaller than the second evaluation index 32P, then the first other data 51 is a better fit for the first acquired data 11. In this case, the regression data obtained using the first other data 51 may be used preferentially.

[0079] For example, if the second evaluation metric 32P is smaller than the first evaluation metric 31P, then the second set of other data 52 is a better fit for the first set of acquired data 11. In this case, the regression data obtained using the second set of other data 52 may be used preferentially.

[0080] In this embodiment, the processing unit 71 may also perform regression on other acquired data using one of the specified first other data 51 and second other data 52. Regression data for the other acquired data can be obtained with higher accuracy.

[0081] Figure 13 is a schematic diagram illustrating the operation of a data processing device according to the first embodiment. As shown in Figure 13, the processing unit 71 may also perform the third evaluation index derivation operation PP3. In the third evaluation index derivation operation PP3, the processing unit 71 derives a third evaluation index 33P from a plurality of third regression labels 63b and a plurality of third composite regression labels 63Sb.

[0082] Multiple third regression labels 63b are derived from multiple third machine learning models 33. Multiple third machine learning models 33 are derived from multiple third sample data 43. Multiple third composite regression labels 63Sb are derived from multiple third composite machine learning models 33S. Multiple third composite machine learning models 33S are derived from multiple third sample data 43 and first acquired data 11 by third transfer learning 33L. In one example, multiple third composite data 33SD may be used in the third transfer learning 33L. Multiple third sample data 43 are derived from third other data 53, or third transformed other data 53C obtained by transforming the third other data 53.

[0083] For example, based on the results of comparing the first evaluation metric 31P, the second evaluation metric 32P, and the third evaluation metric 33P, one of the first other data 51, the second other data 52, and the third other data 53 may be specified. These evaluation metrics can be used to determine the optimal source data.

[0084] Figure 14 is a schematic diagram illustrating the operation of a data processing device according to the first embodiment. As shown in Figure 14, the processing unit 71 can perform regression based on specified other data 50x. The specified other data 50x is specified based on the first evaluation index 31P and the second evaluation index 32P, etc. The processing unit 71 can perform regression on the first acquired data 11 using one of the specified first other data 51 and second other data 52.

[0085] For example, a machine learning model 35x is derived by transfer learning 35xL based on specified other data 50x. Synthetic data 35xSD may be used in transfer learning 35xL. Synthetic data 35xSD is the synthetic data in the corresponding machine learning model 35x. The machine learning model 35x performs regression on the first acquired data 11 to obtain regression labels 66. The specified other data 50x is suitable for regression. The regression labels 66 may be stored in the memory unit 73.

[0086] Figure 15 is a schematic diagram illustrating the operation of a data processing device according to the first embodiment. As shown in Figure 15, there may be other acquired data 11x that are relatively similar to the first acquired data 11. In this embodiment, regression may be performed on the other acquired data 11x using one of the specified first other data 51 and second other data 52. The machine learning model 35x performs regression on the other acquired data 11x to obtain regression labels 66x. The specified other data 50x is suitable for regression. The regression labels 66x obtained in this way fit the other acquired data 11x with high accuracy. The regression labels 66x may be stored in the storage unit 73.

[0087] In this embodiment, the other acquired data 11x may be obtained from the first device 81. The processing unit 71 may also supply the results (regression labels 66x) obtained by performing regression on the other acquired data 11x to the first device 81. This supply may be performed, for example, via a server 74.

[0088] For example, if the target device is a magnetic recording and playback device, the first device 81 may be a magnetic recording and playback device used by a user. Data obtained from the first device 81 (another acquired data 11x) is subjected to regression processing. The result obtained from the regression processing (regression label 66x) is supplied to the first device 81. The regression label 66x may include, for example, the operating conditions of the first device 81. The regression label 66x may also be a shift (e.g., an anomaly) in the operating conditions of the first device 81. The characteristics of the first device 81 can be predicted with high accuracy.

[0089] For example, during the development of a target device, a small sample size may result in low regression accuracy. In this embodiment, even with a small sample size, the characteristics of the developed item can be evaluated with high accuracy.

[0090] In this embodiment, the first acquired data 11, the first other data 51, and the second other data 52 may include characteristics of the magnetic recording and playback device.

[0091] The characteristics of a magnetic recording and reproducing device include, for example, at least one selected from the group consisting of SNR (Signal-Noise Ratio), BER (Bit Error Rate), Fringe BER, EWAC (Erase Width at AC erase), MWW (Magnetic Write track Width), OW (Over Write), SOVA-BER (Soft Viterbi Algorithm-BER), VMM (Viterbi Metric Margin), RRO (Repeatable RunOut), and NRRO (Non-Repeatable RunOut).

[0092] In this embodiment, "D1" may be 1. For example, a high first ratio R1 can be obtained. High-precision processing can be performed with a smaller "D1".

[0093] In the data processing system 210 according to the embodiment (see Figure 1), multiple elements (for example, the processing unit 71 and the storage unit 73) may be provided in different locations. Information may be transmitted and received by any communication method. For example, multiple parts included in the processing unit 71 may be provided in different locations.

[0094] For example, the data processing system 210 may include one or more acquisition units 72 and one or more processing units 71 (see Figure 2). One or more acquisition units 72 are capable of acquiring first acquired data 11 and first other data 51.

[0095] As shown in Figure 1, one or more processing units 71 can perform the first evaluation index derivation operation PP1. In the first evaluation index derivation operation PP1, the processing unit 71 derives the first evaluation index 31P from a plurality of first regression labels 61b and a plurality of first composite regression labels 61Sb. As already explained, the plurality of first regression labels 61b are derived from a plurality of first machine learning models 31. The plurality of first machine learning models 31 are derived from a plurality of first sample data 41.

[0096] Multiple first synthetic regression labels 61Sb are derived from multiple first synthetic machine learning models 31S. Multiple first synthetic machine learning models 31S are derived from multiple first sample data 41 and first acquired data 11 by first transfer learning 31L. Multiple first sample data 41 are derived from first other data 51, or first transformed other data 51C obtained by transforming the first other data 51.

[0097] According to the data processing system 210, according to the embodiment, a data processing system capable of high-precision data processing is provided.

[0098] Figure 16 is a schematic diagram illustrating a data processing device according to an embodiment. As shown in Figure 16, the data processing device 110 includes a processing unit 71, an acquisition unit 72, and a storage unit 73. The processing unit 71 is, for example, an electrical circuit. The storage unit 73 may include, for example, at least one of ROM (Read Only Memory) and RAM (Random Access Memory). Any storage device may be used as the storage unit 73.

[0099] The data processing device 110 may include a display unit 79b and an input unit 79c, etc. The display unit 79b may include various types of displays. The input unit 79c may include, for example, a device having an operating function (e.g., a keyboard, mouse, touch input panel, or voice recognition input device).

[0100] The embodiment may include a program. The program causes the computer (processing unit 71) to perform the above operations. The embodiment may also include a storage medium in which the above program is stored.

[0101] (Second Embodiment) The second embodiment relates to a data processing method. The data processing method according to the embodiment causes the processing unit 71 to perform a first evaluation index derivation operation PP1. In the first evaluation index derivation operation PP1, the processing unit 71 derives a first evaluation index 31P from a plurality of first regression labels 61b and a plurality of first composite regression labels 61Sb. The plurality of first regression labels 61b are derived from a plurality of first machine learning models 31. The plurality of first machine learning models 31 are derived from a plurality of first sample data 41.

[0102] Multiple first synthetic regression labels 61Sb are derived from multiple first synthetic machine learning models 31S. Multiple first synthetic machine learning models 31S are derived from multiple first sample data 41 and first acquired data 11 by first transfer learning 31L. Multiple first sample data 41 are derived from first other data 51, or first transformed other data 51C obtained by transforming the first other data 51. According to this embodiment, a data processing method capable of high-precision data processing can be provided.

[0103] The embodiment may include the following configuration (e.g., proposed technical details). (Composition 1) Acquisition section, Processing unit and Equipped with, The acquisition unit is capable of acquiring the first acquired data and the first other data. The processing unit is capable of performing a first evaluation index derivation operation that derives a first evaluation index from a plurality of first regression labels and a plurality of first composite regression labels. The aforementioned multiple first regression labels are derived from multiple first machine learning models. The aforementioned multiple first machine learning models are derived from multiple first sample data, The aforementioned multiple first synthetic regression labels are derived from multiple first synthetic machine learning models. The plurality of first synthetic machine learning models are derived from the plurality of first sample data and the first acquired data by first transfer learning. The plurality of first sample data are derived from the first other data, or from first transformed other data obtained by transforming the first other data, in a data processing device.

[0104] (Configuration 2) The data processing device according to configuration 1, wherein the processing unit is capable of deriving the plurality of first sample data from the first other data or the first transformed other data by bootstrap sampling.

[0105] (Composition 3) The first acquired data includes a first feature matrix with N1 rows and D1 columns, and a first acquired label with N1 rows. The aforementioned first other data includes a first other feature matrix of Np rows and D1 columns, and a first other label of Np rows, The above N1 is an integer greater than or equal to 2, The aforementioned Np is an integer greater than or equal to 2, The aforementioned D1 is an integer greater than or equal to 1, The data processing device according to configuration 1 or 2, wherein N1 is smaller than Np.

[0106] (Composition 4) One of the aforementioned plurality of first sample data includes a first sample feature matrix with Ns rows and D1 column, and a first sample label with Ns rows. The aforementioned Ns is an integer greater than or equal to 2, The data processing device according to configuration 3, wherein Ns is smaller than Np.

[0107] (Composition 5) The data processing device according to configuration 4, wherein the change in frequency of the first transformed data relative to the first feature matrix is ​​smaller than the change in frequency of the first acquired data relative to the first feature matrix.

[0108] (Composition 6) The processing unit is capable of performing a first synthetic machine learning model derivation operation to derive the plurality of first synthetic machine learning models. In the first synthetic machine learning model derivation operation, the processing unit can generate one of the plurality of first synthetic machine learning models based on the first acquired data and first generated data based on one of the plurality of first sample data. The first generated data includes a first generated matrix of (Ns+N1) rows and (3×D1) columns, and a first generated label of (Ns+N1) rows. The aforementioned first generator matrix includes the first matrix data, the second matrix data, and the third matrix data. The components of the aforementioned first matrix data include the row-direction join of the aforementioned first feature matrix and the aforementioned first sample feature matrix, The components of the second matrix data include a matrix of zero components in row Ns and column D1, and the first sample feature matrix, joined in the row direction. The components of the third matrix data include the combination in the row direction of the first sample feature matrix and the matrix of zero components in row N1 and column D1, A data processing device according to any one of configurations 4 to 5, wherein the components of the first generated label include the combination of the first sample label and the first acquired label in the row direction.

[0109] (Composition 7) The data processing device described in configuration 6, wherein (Ns+N1) / D1 is 250 or greater.

[0110] (Composition 8) The processing unit is capable of deriving one of the multiple first synthetic regression labels by inputting one of the multiple first synthetic regression matrices into one of the multiple first synthetic machine learning models during the first synthetic machine learning model derivation operation. One of the aforementioned plurality of first composite regression matrices has N1 rows (3 × D1) columns, One of the plurality of first composite regression matrices includes first composite regression matrix data, second composite regression matrix data, and third composite regression matrix data. The components of the aforementioned composite regression matrix data include the aforementioned feature matrix, The components of the aforementioned second composite regression matrix data include the aforementioned first feature matrix, The data processing apparatus according to configuration 6 or 7, wherein the components of the third composite regression matrix data include a matrix of zero components in N1 rows and D1 column.

[0111] (Composition 9) The data processing device according to any one of configurations 3 to 8, wherein the first evaluation index corresponds to the mean of the distribution of the difference between one of the plurality of first regression labels and one of the plurality of first composite regression labels.

[0112] (Composition 10) The data processing device according to any one of configurations 3 to 8, wherein the first evaluation index corresponds to the mean of the half-width of the histogram of the difference between one of the plurality of first regression labels and one of the plurality of first composite regression labels.

[0113] (Composition 11) The processing unit is capable of further deriving an error index based on the first evaluation index, as described in any one of configurations 1 to 10.

[0114] (Composition 12) Equipped with additional memory, The data processing device according to any one of configurations 1 to 11, wherein the storage unit is capable of storing at least one of the first acquired data, the first other data, and the first evaluation index.

[0115] (Composition 13) The data processing device according to configuration 12, wherein the storage unit is capable of storing at least one of the plurality of first regression labels, the plurality of first composite regression labels, the plurality of first machine learning models, the plurality of first sample data, the plurality of first composite machine learning models, and the first transformed other data.

[0116] (Composition 14) The aforementioned acquisition unit can further acquire second other data, The processing unit is capable of performing a second evaluation index derivation operation, which derives a second evaluation index from a plurality of second regression labels and a plurality of second composite regression labels. The aforementioned multiple second regression labels are derived from multiple second machine learning models. The aforementioned multiple second machine learning models are derived from multiple second sample data, The aforementioned multiple second synthetic regression labels are derived from multiple second synthetic machine learning models. The aforementioned plurality of second synthetic machine learning models are derived from the plurality of second sample data and the first acquired data by second transfer learning. The plurality of second sample data are derived from the second other data, or second transformed other data obtained by transforming the second other data, according to one of configurations 1 to 13.

[0117] (Composition 15) The data processing device according to configuration 14, wherein the processing unit can perform a designation operation to designate one of the first other data and the second other data based on the result of comparing the first evaluation index and the second evaluation index.

[0118] (Composition 16) The data processing device according to configuration 15, wherein the processing unit can perform regression on other acquired data using one of the designated first other data and second other data.

[0119] (Composition 17) The aforementioned separate acquired data is obtained from the first device. The data processing device according to configuration 16, wherein the processing unit is capable of supplying the results obtained by performing the regression on the other acquired data to the first device.

[0120] (Composition 18) One or more acquisition units, One or more processing units, Equipped with, The one or more acquisition units are capable of acquiring first acquired data and first other data. The one or more processing units described above are capable of performing a first evaluation index derivation operation that derives a first evaluation index from a plurality of first regression labels and a plurality of first composite regression labels. The aforementioned multiple first regression labels are derived from multiple first machine learning models. The aforementioned multiple first machine learning models are derived from multiple first sample data, The aforementioned multiple first synthetic regression labels are derived from multiple first synthetic machine learning models. The plurality of first synthetic machine learning models are derived from the plurality of first sample data and the first acquired data by first transfer learning. A data processing system in which the plurality of first sample data are derived from the first other data, or from first transformed other data obtained by transforming the first other data.

[0121] (Composition 20) The processing unit is instructed to perform the first evaluation index derivation operation. In the first evaluation index derivation operation, the processing unit shall The first evaluation metric is derived from multiple first regression labels and multiple first composite regression labels. The aforementioned multiple first regression labels are derived from multiple first machine learning models. The aforementioned multiple first machine learning models are derived from multiple first sample data, The aforementioned multiple first synthetic regression labels are derived from multiple first synthetic machine learning models. The aforementioned plurality of first synthetic machine learning models are derived from the plurality of first sample data and first acquired data by first transfer learning. A data processing method in which the plurality of first sample data are derived from first other data, or first transformed other data obtained by transforming the first other data.

[0122] According to the embodiment, a data processing device, a data processing system, and a data processing method capable of high-precision data processing can be provided.

[0123] Embodiments of the present invention have been described above with reference to examples. However, the present invention is not limited to these examples. For example, the specific configuration of each element, such as a processing unit, an acquisition unit, and a storage unit, included in a data processing device, data processing system, and data processing method, is included within the scope of the present invention as long as a person skilled in the art can appropriately select from what is known to implement the present invention and obtain similar effects.

[0124] Combinations of two or more elements from each example, to the extent technically feasible, are also included within the scope of the present invention, insofar as they encompass the gist of the invention.

[0125] All data processing devices, data processing systems, and data processing methods that a person skilled in the art can design and implement based on the data processing devices, data processing systems, and data processing methods described above as embodiments of the present invention also fall within the scope of the present invention, insofar as they encompass the gist of the present invention.

[0126] Within the scope of the concept of this invention, a person skilled in the art would be able to conceive of various modifications and alterations, and it is understood that such modifications and alterations also fall within the scope of this invention.

[0127] While several embodiments of the present invention have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be carried out in a variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims of the invention and its equivalents. [Explanation of symbols]

[0128] 10D…Data, 11, 12…First and second acquired data, 11a…First feature matrix, 11b…First acquired label, 11x…Acquired data, 21…First generated data, 21a…First generated matrix, 21b…First generated label, 31~33…First to third machine learning models, 31L~33L…First to third transfer learning, 31P~33P…First to third evaluation metrics, 31S~33S…First to third synthetic machine learning models, 31SD~33SD…First to third synthetic data, 35x…Machine learning model, 35xL…Transfer learning, 35xSD…Synthetic data, 41~43…First to third sample data, 41a…First sample feature matrix, 41b…First sample label, 50x…Other data, 51~53…1st~3rd other data, 51C~53C…1st~3rd transformed other data, 51a…1st other feature matrix, 51b…1st other label, 61Sa…1st composite regression matrix, 61Sb~63Sb…1st~3rd composite regression labels, 61a…1st regression matrix, 61b~63b…1st~3rd regression labels, 66, 66x…Regression labels, 71…Processing unit, 72…Acquisition unit, 73…Storage unit, 73a, 73b…1st and 2nd storage areas, 74…Server, 79b…Display unit, 79c…Input unit, 81…1st device, 110…Data processing unit, 210…Data processing system, EW…Erasure width, HW…Index, I1…Information, K1~K3…1st~3rd composite regression matrix data, M1~M3…1st to 3rd matrix data, Mxa1, Mxb1, Mxc1…matrices, OP1…1st action, PP1~PP3…1st to 3rd evaluation index derivation actions, Pg…frequency

Claims

1. Acquisition section, Processing unit and Equipped with, The acquisition unit is capable of acquiring first acquired data and first other data. The processing unit is capable of performing a first evaluation index derivation operation that derives a first evaluation index from a plurality of first regression labels and a plurality of first composite regression labels. The aforementioned multiple first regression labels are derived from multiple first machine learning models. The aforementioned multiple first machine learning models are derived from multiple first sample data, The aforementioned plurality of first synthetic regression labels are derived from a plurality of first synthetic machine learning models. The plurality of first synthetic machine learning models are derived from the plurality of first sample data and the first acquired data by first transfer learning. The plurality of first sample data are derived from the first other data, or from the first transformed other data obtained by transforming the first other data. The first acquired data includes a first feature matrix with N1 rows and D1 columns, and a first acquired label with N1 rows. The aforementioned first other data includes a first other feature matrix with Np rows and D1 columns, and a first other label with Np rows. The aforementioned N1 is an integer of 2 or more, The aforementioned Np is an integer of 2 or more, The above D1 is an integer greater than or equal to 1, The N1 is a data processing device that is smaller than the Np.

2. One of the aforementioned plurality of first sample data includes a first sample feature matrix with Ns rows and D1 column, and a first sample label with Ns rows, The aforementioned Ns is an integer of 2 or more, The data processing apparatus according to claim 1, wherein Ns is smaller than Np.

3. The processing unit is capable of performing a first synthetic machine learning model derivation operation to derive the plurality of first synthetic machine learning models. In the first synthetic machine learning model derivation operation, the processing unit can generate one of the plurality of first synthetic machine learning models based on the first acquired data and first generated data based on one of the plurality of first sample data. The first generated data includes a first generated matrix of (Ns + N1) rows and (3 × D1) columns, and a first generated label of (Ns + N1) rows, The first generator matrix includes the first matrix data, the second matrix data, and the third matrix data. The components of the first matrix data include the row-direction combination of the first feature matrix and the first sample feature matrix, The components of the second matrix data include a matrix of zero components in row Ns and column D1, and the first sample feature matrix, joined in the row direction. The components of the third matrix data include the combination in the row direction of the first sample feature matrix and the matrix of zero components in row N1 and column D1, The data processing apparatus according to claim 2, wherein the components of the first generated label include the combination of the first sample label and the first acquired label in the row direction.

4. The processing unit is capable of deriving one of the multiple first synthetic regression labels by inputting one of the multiple first synthetic regression matrices into one of the multiple first synthetic machine learning models during the first synthetic machine learning model derivation operation. One of the aforementioned plurality of first composite regression matrices has N1 rows (3 × D1) columns, One of the plurality of first composite regression matrices includes first composite regression matrix data, second composite regression matrix data, and third composite regression matrix data. The components of the aforementioned composite regression matrix data include the aforementioned feature matrix, The components of the aforementioned second composite regression matrix data include the aforementioned first feature matrix, The data processing apparatus according to claim 3, wherein the components of the third composite regression matrix data include a matrix of zero components in N1 rows and D1 column.

5. Acquisition unit, Processing unit and Equipped with, The acquisition unit is capable of acquiring first acquired data and first other data. The processing unit is capable of performing a first evaluation index derivation operation that derives a first evaluation index from a plurality of first regression labels and a plurality of first composite regression labels. The aforementioned multiple first regression labels are derived from multiple first machine learning models. The aforementioned multiple first machine learning models are derived from multiple first sample data, The aforementioned plurality of first synthetic regression labels are derived from a plurality of first synthetic machine learning models. The plurality of first synthetic machine learning models are derived from the plurality of first sample data and the first acquired data by first transfer learning. The plurality of first sample data are derived from the first other data, or from the first transformed other data obtained by transforming the first other data. The acquisition unit is capable of acquiring second other data, The processing unit is capable of performing a second evaluation index derivation operation that derives a second evaluation index from a plurality of second regression labels and a plurality of second composite regression labels. The aforementioned multiple second regression labels are derived from multiple second machine learning models. The aforementioned multiple second machine learning models are derived from multiple second sample data, The aforementioned multiple second synthetic regression labels are derived from multiple second synthetic machine learning models. The plurality of second synthetic machine learning models are derived from the plurality of second sample data and the first acquired data by second transfer learning. The plurality of second sample data are derived from the second other data, or second transformed other data obtained by transforming the second other data, in a data processing device.

6. The data processing device according to claim 5, wherein the processing unit can perform a designation operation to designate one of the first other data and the second other data based on the result of comparing the first evaluation index and the second evaluation index.

7. The data processing device according to claim 6, wherein the processing unit can perform regression on other acquired data using one of the designated first other data and second other data.

8. One or more acquisition units, One or more processing units, Equipped with, The one or more acquisition units are capable of acquiring first acquired data and first other data. The one or more processing units described above are capable of performing a first evaluation index derivation operation that derives a first evaluation index from a plurality of first regression labels and a plurality of first composite regression labels. The aforementioned multiple first regression labels are derived from multiple first machine learning models. The aforementioned multiple first machine learning models are derived from multiple first sample data, The aforementioned plurality of first synthetic regression labels are derived from a plurality of first synthetic machine learning models. The plurality of first synthetic machine learning models are derived from the plurality of first sample data and the first acquired data by first transfer learning. The plurality of first sample data are derived from the first other data, or from the first transformed other data obtained by transforming the first other data. The first acquired data includes a first feature matrix with N1 rows and D1 columns, and a first acquired label with N1 rows. The aforementioned first other data includes a first other feature matrix with Np rows and D1 columns, and a first other label with Np rows. The aforementioned N1 is an integer of 2 or more, The aforementioned Np is an integer of 2 or more, The above D1 is an integer greater than or equal to 1, The N1 is a data processing system that is smaller than the Np.

9. The processing unit is instructed to perform the first evaluation index derivation operation. In the first evaluation index derivation operation, the processing unit, The first evaluation metric is derived from multiple first regression labels and multiple first composite regression labels. The aforementioned multiple first regression labels are derived from multiple first machine learning models. The aforementioned multiple first machine learning models are derived from multiple first sample data, The aforementioned plurality of first synthetic regression labels are derived from a plurality of first synthetic machine learning models. The plurality of first synthetic machine learning models are derived from the plurality of first sample data and the first acquired data by first transfer learning. The aforementioned plurality of first sample data are derived from first other data, or first transformed other data obtained by transforming the aforementioned first other data. The first acquired data includes a first feature matrix with N1 rows and D1 columns, and a first acquired label with N1 rows. The aforementioned first other data includes a first other feature matrix with Np rows and D1 columns, and a first other label with Np rows. The aforementioned N1 is an integer of 2 or more, The aforementioned Np is an integer of 2 or more, The above D1 is an integer greater than or equal to 1, A data processing method in which N1 is smaller than Np.

Images