METHOD AND SYSTEM FOR CONFIGURING PREPROCESSING FOR SUPERVISED LEARNING MODELS

Abstract

An exemplary procedure includes defining a function g with a plurality of parameters a to l; ordering the function g to a supervised learning model f, where the supervised learning model f is trained with training data TDi containing an example xi and a label yi, and i is an integer; defining a function f(g(xi)) using the supervised learning model f and the function g; computing a result Zi by computing the function f(g(xi)); computing a difference Li between the result Zi and the label yi; computing a partial differential PDi of the difference Li for the plurality of parameters a to l of the function g; and updating the plurality of parameters a to l of the function g according to the partial differential PDi.Defining a pseudo-preprocessor h using the plurality of parameters a to l of the function g, if the difference Li converges during a repetition between the computation of the result Zi and the updating of the plurality of parameters a to l of the function g; placing the pseudo-preprocessor h in front of the supervised learning model f; and training the supervised learning model f by providing the training data TDi to the pseudo-preprocessor h.

Description

[TECHNICAL FIELD]

[0001] The present disclosure relates to methods and systems for configuring preprocessing used for a supervised learning model, and more particularly to methods and systems for evaluating a plurality of parameters used in the preprocessing. [BACKGROUND TO THE STATE OF THE ART]

[0002] The object detection device described in Japanese Published Patent Application No. 2019-219804 performs preprocessing on an image captured with a camera, thus detecting an object in the image. To improve the accuracy of object detection, the object detection device adjusts a variety of parameters included in the preprocessing before performing the preprocessing.

[0003] Fig. Figure 23 shows a conventional preprocessing design for a supervised learning model. Fig. 23, there is a preprocessor PP1 designed by an engineer E(A) for a supervised learning model SLM (used for 32-bit processing) using training data TD1 to TDn (n is an integer equal to or greater than two). It is assumed that the preprocessor PP1 has a large number of parameters that are, for example, unreproducible or unknown by an engineer E(B) and other engineers, and the engineer E(B) attempts to redirect or transfer the supervised learning model SLM from 32-bit processing to 64-bit processing.

[0004] However, engineer E(B) cannot reproduce the multitude of parameters used for preprocessing PP1 because the multitude of parameters is not reproducible or unknown. Therefore, engineer E(B) sets up another preprocessing PP2 with a multitude of other, unsuitable parameters that differ from the multitude of parameters used in preprocessing PP1. Consequently, engineer E(B) cannot redirect or transfer the supervised learning model SLM from 32-bit processing to 64-bit processing. [Reference list][Patent literature]

[0005] [PTL 1]: JP 2019-219804 A [SUMMARY OF THE INVENTION]

[0006] To solve the above problem, one aspect of the present disclosure provides a method comprising: defining a function g having a plurality of parameters a to l; placing the function g in front of a supervised learning model f; wherein the supervised learning model f is trained with training data TDi including an example xi and a label yi, and i is an integer; defining a function f(g(xi)) by using the supervised learning model f and the function g; calculating a result Zi by calculating the function f(g(xi)); calculating a difference Li between the result Zi and the label yi; calculating a partial differential PDi of the difference Li for the plurality of parameters a to l of the function g; updating the plurality of parameters a to l of the function g according to the partial differential PDi;Defining a pseudo-preprocessing h using the plurality of parameters a to l of the function g when the difference Li converges during the iteration between the calculation of the result Zi and the updating of the plurality of parameters a to l of the function g; arranging the pseudo-preprocessing h before the supervised learning model f; and training the supervised learning model f by providing the training data TDi to the pseudo-preprocessing h.

[0007] Another aspect of the present disclosure provides a method comprising: obtaining a first plurality of combinations COMB_a1 to COMB_l1 each comprising a selection of values ​​for one of a plurality of parameters a to l of a first function g1 arranged in front of a first supervised learning model f1 that implements feedback to the first function g1, by providing training data TDi for the first supervised learning model f1, where i is an integer;Obtaining a second plurality of combinations COMB_a2 to COMB_l2, each having a selection of values ​​on one of a plurality of parameters a to l of a second function g2 arranged in front of a second supervised learning model f2 that implements feedback to the second function g2, by providing training data TDi to the second supervised learning model f2, wherein the operation of the first supervised learning model f1 and the operation of the second supervised learning model f2 are equivalent to each other; calculating a first plurality of distributions D_a1 to D_l1 for each of the first plurality of combinations COMB_a1 to COMB_l1; calculating a second plurality of distributions D_a2 to D_l2 for each of the second plurality of combinations COMB_a2 to COMB_l2;and evaluating which of the first function g1 and the second function g2 is more robust by comparing the first plurality of distributions D_a1 to D_l1 and the second plurality of distributions D_a2 to D_l2, respectively.;

[0008] Yet another aspect of the present disclosure provides a method comprising: obtaining a plurality of combinations COMB(1) to COMB(n), each having a plurality of values ​​for a plurality of parameters a to l of a function g arranged in front of a supervised learning model f that implements feedback to the first function g, by providing training data TDi to the supervised learning model f, where n is an integer equal to or greater than two; arranging the plurality of combinations COMB(1) to COMB(n) in a space having a plurality of dimensions, the number of which is equivalent to a number of the plurality of parameters a to l of the function g; obtaining at least one sampled point SP in the space; selecting at least one combination from the plurality of combinations COMB(1) to COMB(n) that is closest to the at least one sampled point SP;and assigning the plurality of values ​​contained in the selected combination from the plurality of combinations COMB(1) to COMB(n) to the plurality of parameters a to l of the function g.;

[0009] Yet another aspect of the present disclosure provides an apparatus comprising: a first definer that defines a function g having a plurality of parameters a to l; a first arranger that arranges the function g in front of a supervised learning model f; wherein the supervised learning model f is trained with training data TDi containing an example xi and a label yi, and i is an integer; a second definer that defines a function f(g(xi)) by using the supervised learning model f and the function g; a first calculator that calculates a result Zi by calculating the function f(g(xi)); a second calculator that calculates a difference Li between the result Zi and the label yi; a third calculator that calculates a partial differential PDi of the difference Li for the plurality of parameters a to l of the function g;an updater that updates the plurality of parameters a to l of the function g according to the partial differential PDi; a third definer that defines a pseudo preprocessing h using the plurality of parameters a to l of the function g when the difference Li converges during the iteration between the calculation of the result Zi and the updating of the plurality of parameters a to l of the function g; a second arranger that arranges the pseudo preprocessing h before the supervised learning model f; and a trainer that trains the supervised learning model f by providing the training data TDi for the pseudo preprocessing h.

[0010] Yet another aspect of the present disclosure provides an apparatus comprising: a first acquirer that acquires a first plurality of combinations COMB_a1 to COMB_l1 each comprising a selection of values ​​for one of a plurality of parameters a to l of a first function g1 disposed in front of a first supervised learning model f1 that implements feedback to the first function g1 by providing training data TDi to the first supervised learning model f1, where i is an integer;a second acquirer that acquires a second plurality of combinations COMB_a2 to COMB_l2, each having a selection of values ​​on one of a plurality of parameters a to l of a second function g2 arranged in front of a second supervised learning model f2 that implements feedback to the second function g2 by providing training data TDi to the second supervised learning model f2, wherein the operation of the first supervised learning model f1 and the operation of the second supervised learning model f2 are equivalent to each other; a first calculator that calculates a first plurality of distributions D_a1 to D_l1 for the first plurality of combinations COMB_a1 to COMB_l1, respectively; a second calculator that calculates a second plurality of distributions D_a2 to D_l2 for the second plurality of combinations COMB_a2 to COMB_l2, respectively;and an evaluator that evaluates which of the first function g1 and the second function g2 is more robust by comparing the first plurality of distributions D_a1 to D_l1 and the second plurality of distributions D_a2 to D_l2, respectively.;

[0011] Yet another aspect of the present disclosure provides an apparatus comprising: a first acquirer that acquires a plurality of combinations COMB(1) to COMB(n), each having a plurality of values ​​for a plurality of parameters a to l of a function g arranged in front of a supervised learning model f that implements feedback to the first function g by providing training data TDi to the supervised learning model f, where n is an integer equal to or greater than two; an arranger that arranges the plurality of combinations COMB(1) to COMB(n) in a space having a plurality of dimensions, the number of which is equivalent to a number of the plurality of parameters a to l of the function g; a second acquirer that acquires at least one sampled point SP in the space;a selector that selects at least one of the plurality of combinations COMB(1) to COMB(n) that is closest to the at least one sampled point SP; and an allocator that assigns the plurality of values ​​included in the selected combination from the plurality of combinations COMB(1) to COMB(n) to the plurality of parameters a to l of the function g; [BRIEF DESCRIPTION OF THE DRAWINGS] Fig. 1 is a block diagram showing a configuration of a controlled system CS controlled by a controller CD according to a first embodiment. Fig. 2 shows a configuration of a function g according to the first embodiment. Fig. 3 shows configurations of training data TD1 to TDn. Fig. 4 is a block diagram showing a configuration of the control device CD according to the first embodiment from a functional point of view. Fig. 5 is a block diagram showing a configuration of the control device CD according to the first embodiment from a hardware point of view. Fig. 6 is a block diagram showing a configuration of the control device CD according to the first embodiment from a software point of view. Fig. 7 is a flowchart showing an operation (part 1) of the control device CD according to the first embodiment. Fig. 8 is a flowchart showing an operation (part 2) of the control device CD according to the first embodiment. Fig. 9 shows a configuration of the control device CD according to the first embodiment when i ranges to 80. Fig. 10 is a block diagram showing configurations of a first controlled system CS1 and a second controlled system CS2 controlled by a controller CD according to the second embodiment. Fig. 11 is a block diagram showing a configuration of the control device CD according to the second embodiment from a functional point of view. Fig. 12 is a flowchart showing an operation of the control device CD according to the second embodiment. Fig. 13 shows a plurality of combinations COMB_a1 to COMB_l1 having a selection of values ​​for a plurality of parameters a to l of the first function g1. Fig. 14 shows a plurality of combinations COMB_a2 to COMB_l2 having a selection of values ​​for a plurality of parameters a to l of the second function g2. Fig. 15 shows a plurality of distributions D_a1 of the combination COMB_a1 to D_l1 of the combination COMB_l1 and a plurality of distributions D_a2 of the combination COMB_a2 to D_l2 of the combination COMB_l2. Fig. 16 is a block diagram showing a configuration of a controlled system CS controlled by a controller CD according to a third embodiment. Fig. 17 is a block diagram showing a configuration of the control device CD according to the third embodiment from a functional point of view. Fig. 18 is a flowchart showing an operation of the control device CD according to the third embodiment. Fig. Figure 19 shows a variety of combinations COM(1) to COM(n) with a selection of values ​​related to a variety of parameters a to l of the function g. Fig. 20 shows the plurality of combinations COMB(1) to COMB(n) and a sampled point SP arranged in space. Fig. Figure 21 shows the plurality of combinations COMB(1) to COMB(n) and an unexpected sampled point SP. Fig. 22 shows the plurality of combinations COMB(1) to COMB(n) and several sampled points SP1, SP2 and SP3. Fig. Figure 23 shows a conventional preprocessing design for a supervised learning model. [DESCRIPTION OF THE EMBODIMENTS]

[0012] In order to describe the present disclosure in more detail, embodiments of the present disclosure will be described below with reference to the accompanying drawings. First EmbodimentConfiguration of the First Embodiment

[0013] A control device according to a first embodiment of this disclosure will now be described with reference to the Fig. 1 to 9 described.

[0014] Fig. 1 is a block diagram showing a configuration of a controlled system CS controlled by a controller CD according to the first embodiment. The controller CD controls, for example, a configuration and operation of the controlled system SC.

[0015] As in Fig. 1, the controlled system CS comprises a supervised learning model f and a function g. The configuration of the supervised learning model f has already been completely designed by an engineer E(A) (shown in Fig. 23), and another engineer E(B) wants to train the operation of the supervised learning model f using training data TD 1 to TDn (n is an integer equal to or greater than two). The function g is any function that has a plurality of parameters a to l.

[0016] Fig. Figure 2 shows a configuration of the function g according to the first embodiment. As shown in Fig. 2, the function g, i.e. g(x), is represented using W, x and z. As in Fig. 2, W is a matrix, e.g., for converting and normalizing (e.g., a 3-row by 3-row vector), x is an input image (e.g., a 3-row by 1-row vector), and z is a vector (e.g., a 3-row by 1-row vector). As shown in Fig. 2, W consists of a plurality of parameters a to i and z consists of a plurality of parameters i to I, which leads to the conclusion that the function g, i.e. g(x), consists of all these parameters a to l.

[0017] Fig. 3 shows configurations of training data TD1 to TDn. As in Fig. For example, as shown in Figure 3, the training data TD1 includes an image x1 as a training example and a label y1. Here, the label y1 is a correct answer indicating whether the image x1, i.e., the example x1, is correct or not in training the supervised learning model f. Similar to the training data TD1, the training data TD2 includes an image x2 and a label y2, the training data TD3 includes an image x3 and a label y3, ..., and the training data TDn includes an image xn and a label yn.

[0018] Fig. Fig. 4 is a block diagram showing a configuration of the control device CD according to the first embodiment from a functional point of view. As shown in Fig. 4, the control device CD according to the first embodiment comprises a definer 11, an arranger 12, a calculator 13, an updater 14 and a trainer 15. Their functions will be explained later with reference to the flowcharts in the Fig. 7 to 8.

[0019] Fig. 5 is a block diagram showing a hardware configuration of the control device CD according to the first embodiment. As shown in Fig. As shown in Figure 5, the control device CD comprises an input circuit IC, a processing circuit PC and an output circuit OC.

[0020] The processing circuit PC is a special piece of hardware. The processing circuit PC implements the Fig. 4 shown functions of the definer 11, the arranger 12, the calculator 13, the updater 14 and the trainer 15.

[0021] The processing circuit PC is, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or a combination thereof.

[0022] Fig. Fig. 6 is a block diagram showing a configuration of the control device CD according to the first embodiment from a software perspective. As shown in Fig. As shown in Figure 6, the control device CD comprises an input circuit IC, a processor PR, a memory circuit PC and an output circuit OC.

[0023] The processor PR is, for example, a CPU (including a central processing unit, a central processing device, a processing device, an arithmetic device, a microprocessor, a microcomputer, or a DSP (Digital Processing)). The processor PR implements the functions of the definer 11 via the Fig. 4 trainers shown 15.

[0024] The processor PR implements the above functions using software, firmware, or a combination of software and firmware. Software and firmware are described as a program stored in the memory circuit SC.

[0025] The processor PR implements the above function by reading and executing the program from the memory circuit SC. In other words, the above program enables a computer to execute the operations and processes of the definer 11 through the trainer 15.

[0026] The memory circuit SC is, for example, a volatile or non-volatile semiconductor memory that includes a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory) and an EEPROM (Electrically Erasable Programmable Read Only Memory); a magnetic disk; a flexible disk; an optical disk; a compact disk or a DVD (Digital Versatile Disc).

[0027] Of the functions of the definer 11 to the trainer 15, some functions can be implemented by the processing circuit PC and other functions can be implemented by the processor PR.

[0028] As described above, the functions of the definer 11 to the trainer 15 in the control device CD can be realized by hardware, software, firmware or a combination thereof. Operation of the first embodiment

[0029] Fig. 7 is a flowchart showing an operation (part 1) of the control device CD according to the first embodiment.

[0030] Fig. 8 is a flowchart showing an operation (part 2) of the control device CD according to the first embodiment.

[0031] To simplify the explanation and understanding, it is assumed that an integer i can be counted up to 100 and a difference Li converges when the integer i reaches the value 80.

[0032] In step ST11, the definer 11 (shown in Fig. 4) a function g (represented in Fig. 1) with the multitude of parameters a to l (shown in Fig. 3). In addition, the definer 11 initializes the plurality of parameters a to l. After initialization by the definer 11, the arranger 12 (in Fig. 4) the function g before the supervised learning model f (in Fig. 1) or adjusts it as shown in Fig. 1 shown.

[0033] In step ST12, the definer 11 defines a function f(g(x)) using the supervised learning model f and the function g. The function f(g(x)) is formed in a nested manner, ie, the function g provides an output depending on an input image x (shown in Fig. 2), and the supervised learning model f provides an output depending on the output provided by the function g.

[0034] In step ST13, the definer 11 initializes the integer i, ie, it sets 1 to the integer i.

[0035] In step ST14, the calculator 13 calculates (in Fig. 4) a result Zi of the function f(g(xi).

[0036] In step ST15, the calculator 13 calculates a difference Li between the result Zi and the label yi (in Fig. 3), by subtracting the label yi from the result Zi.

[0037] In step ST 16, the calculator 13 calculates a partial differential PDi of the difference Li related to the plurality of parameters a to l of the function g.

[0038] In step ST 17, the updater 14 updates (in Fig. 4) the plurality of parameters a to l of the function g according to the partial differential PDi.

[0039] As in Fig. For example, as shown in Figure 7, when the integer i is 1, the plurality of parameters a to l of the function g are updated to a1 to l1, respectively. Similarly, as shown in Fig. 2, when the integer i is 2, the plurality of parameters a to l of the function g are updated to a2 to l2, respectively, and when the integer i is 3, the plurality of parameters a to l of the function g are updated to a3 to l3, respectively, and when the integer i is n, the plurality of parameters a to l of the function g are updated to an to In (not shown), respectively.

[0040] In step ST18, as a result of repeating step ST14 to step ST20, if the difference Li is determined to be converged, for example, compared with a predetermined threshold value, the flow proceeds to step ST21. On the other hand, if the difference LI is determined not to be converged, for example, compared with the predetermined value, the flow proceeds to step ST19.

[0041] Fig. 9 shows a configuration of the control device CD according to the first embodiment when i ranges to 80.

[0042] In step ST21, the definer 11 defines the pseudo preprocessing h using the plurality of parameters a to I when i reaches the value 80, that is, the plurality of parameters a80 to 180, as shown in Fig. 9. In addition, the arranger 12 replaces the function g with the pseudo-preprocessing h, more precisely, it removes the function g and arranges the pseudo-preprocessing h before the supervised learning model f or places it, as in Fig. 9 shown.

[0043] In step ST 22, the trainer 15 (shown in Fig. 4) the supervised learning model f using the pseudo-preprocessing h, specifically by providing the training data TD1 to TDn in the pseudo-preprocessing h, which is prior to the supervised learning model f, as in Fig. 9 shown.

[0044] In step 19, which follows step ST 18 which indicates NO, the integer i is only increased by one.

[0045] In step 20, the integer i is compared with 100 based on the above assumption. If the integer i is less than 100, the flow returns to step ST14; otherwise, the flow ends. Effect of the first embodiment

[0046] As described above, the controller CD according to the first embodiment defines the pseudo preprocessing h using the plurality of parameters a to l, namely the plurality of parameters a80 to 180 given when the difference Li converges, that is, the integer i reaches the value 80. Consequently, the supervised learning model f can be trained using the pseudo preprocessing h instead of the function g, even if the plurality of parameters a to l of the function g are unknown or not reproducible. Second EmbodimentConfiguration of the Second Embodiment

[0047] A control device according to a second embodiment of this disclosure will now be described with reference to the Fig. 10 to 15 described.

[0048] Fig. 10 is a block diagram showing configurations of a first controlled system CS1 and a second controlled system CS2 controlled by a controller CD according to the second embodiment. Similar to the controller CD of the first embodiment, the controller CD of the second embodiment controls both the configurations and operations of the first controlled system CS1 and the second controlled system CS2, for example.

[0049] As in Fig. As shown in Figure 10, the first controlled system CS1 includes a first supervised learning model f1 and a first function g1, and the second controlled system CS2 includes a second supervised learning model f2 and a second function g2. The operation of the first supervised learning model f1 and the operation of the second supervised learning model f2 are equivalent to each other. In the first controlled system CS1, the first function g1 is arranged or positioned in front of the first supervised learning model f1. Similarly, in the second controlled system CS2, the second function g2 is arranged or positioned in front of the second supervised learning model f2. In the second embodiment, an engineer E(B) wishes to evaluate which of the first function g1 and the second function g2 is more robust using the training data TD1 to TDn. The evaluation of the robustness of the first function g1 and the second function g2 will be described in detail later.

[0050] Fig. 11 is a block diagram showing a configuration of the control device CD according to the second embodiment from a functional point of view. As shown in Fig. 11, the control device CD according to the second embodiment comprises a procurer 21, a calculator 22 and an evaluator 23. Their functions will be explained later with reference to the flowchart in Fig. 12 described.

[0051] Similar to the control device CD according to the first embodiment, the control device CD according to the second embodiment comprises the input circuit IC, the processing circuit PC and the output circuit OC (shown in Fig. 5). Alternatively, the control device CD according to the second embodiment may comprise the input circuit IC, the processor PR, the memory circuit SC and the output circuit OC (in Fig. 6).

[0052] Instead of the functions of the definer 11, the acquirer 12, the calculator 13, the updater 14 and the trainer 15 in the first embodiment (shown in Fig. 4) implements the processing circuit PC (shown in Fig. 5) or the processor PR (shown in Fig. 6) the functions of the procurer 21, the calculator 22 and the evaluator 23 (shown in Fig. 11). Operation of the second embodiment

[0053] Fig. 12 is a flowchart showing an operation of the control device CD according to the second embodiment.

[0054] Fig. 13 shows a plurality of combinations COMB_a1 to COMB_l1 having a selection of values ​​relating to a plurality of parameters a to l of the first function g1.

[0055] In step ST31, the procurer procures 21 (shown in Fig. 11) a variety of combinations COMB_a1 to COMB_l1, which are Fig. 13. Each of the plurality of combinations COMB_a1 to COMB_l1 comprises a selection of values ​​that are suitable for one of the plurality of parameters a to l of the first function g1 (in Fig. 10) are relevant.

[0056] As in Fig. As shown in Figure 13, the combination COMB_a1, for example, comprises a selection of values ​​a1(1), a1(2), a1(3),..., and a1(n) for the parameter a of the first function g1. The selection of values ​​a1(1), a1(2), a1(3),,,, and a1(n) are outputs returned or fed back by the first supervised learning model f1 depending on the training data TD1, TD2, TD3,,,, and TD(n).

[0057] For example, the value a1(1) is an output returned by the first supervised learning model f1 subject to the training data TD1, the value a1(2) is an output returned by the first supervised learning model f1 subject to the training data TD2,,,, and the value a1(n) is an output returned by the first supervised learning model f1 subject to the training data TDn.

[0058] Fig. 14 shows a plurality of combinations COMB_a2 to COMB_l2 having a selection of values ​​for a plurality of parameters a to l of the second function g2.

[0059] In step ST32, the acquirer 21 acquires a plurality of combinations COMB_a2 to COMB_l2. Each of the plurality of combinations COMB_a2 to COMB_l2 includes a selection of values ​​for one of the plurality of parameters a to l of the second function g2.

[0060] As in Fig. 14, the combination COMB_a2, for example, comprises a selection of values ​​a2(1), a2(2), a2(3),,,,, and a2(n) for the parameter a of the second function g2. The selection of values ​​a2(1), a2(2), a2(3),,,, and a2(n) are outputs returned or fed back by the second supervised learning model f2 depending on the training data TD1, TD2, TD3,,,, and TD(n).

[0061] For example, the value a2(1) is an output returned by the second supervised learning model f2 as a function of the training data TD1, the value a2(2) is an output returned by the second supervised learning model f2 as a function of the training data TD2,,,, and the value a2(n) is an output returned by the second supervised learning model f2 as a function of the training data TDn.

[0062] Fig. 15 shows a plurality of distributions D_a1 of the combination COMB_a1 to D_l1 of the combination COMB_l1 and a plurality of distributions D_a2 of the combination COMB_a2 to D_l2 of the combination COMB_l2.

[0063] In step ST33, the calculator 22 (shown in Fig. 11) a variety of distributions D_a1 (shown in Fig. 15) of the combination COMB_a1 (shown in Fig. 13) to D_l1 (shown in Fig. 15) of the combination COMB_l1 (shown in Fig. 13). More specifically, the calculator 22 calculates, for example, the distribution D_a1 of the combination COMB_l1, which includes the selection of values ​​a1(1), a1(2), a1(3),,,, and a1(n) (shown in Fig. 13) with respect to the parameter a of the first function g1 (shown in Fig. 10). As in Fig. As shown in Figure 15, the distribution D_a1 of the combination COMB_a1 has an extent or a dispersion that is inherent to the distribution D_a1.

[0064] Similar to the calculation of the distribution D_a1, the calculator 22 calculates the other distributions D_b1 to D_l1, each of the distributions D_b1 to D_l1 having its own extent or spread.

[0065] In step ST34, the calculator 22 calculates a plurality of distributions D_a2 (shown in Fig. 15) of the combination COMB_a2 (shown in Fig. 14) to D_l2 (shown in Fig. 15) of the combination COMB_l2 (shown in Fig. 14). More specifically, the calculator 22 calculates, for example, the distribution D_a2 of the combination COMB_a2, which includes the selection of values ​​a2(1), a2(2), a2(3),,,, and a2(n) (shown in Fig. 14) with respect to the parameter a of the second function g2 (shown in Fig. 10). As in Fig. As shown in Figure 15, the distribution D_a2 of the combination COMB_a2 has an extension or a dispersion that is inherent to the distribution D_a2.

[0066] Similar to the calculation of the distribution D_a2, the calculator 22 calculates the other distributions D_b2 to D_l2, each of the distributions D_b2 to D_l2 having its own extent or spread.

[0067] In step ST35, the evaluator 23 (shown in Fig. 11), which corresponds to the first function g1 (shown in Fig. 10) and the second function g2 (shown in Fig. 10) is more robust by taking into account the extents or dispersions of the multitude of distributions D_a1 to D_l1 (shown in Fig. 15) and the extents or the dispersions of the multitude of distributions D_a2 to D_l2 (shown in Fig. 15) can be compared.

[0068] The evaluator 23 compares, for example, the extent or the dispersion of the distribution D_a1 and the extent or the dispersion of the distribution D_a2. Since the distribution D_a2 differs from the distribution D_a1, as in Fig. 15, expands or scatters more, the evaluator 23 evaluates the second function g2 as more robust than the first function g1 from the point of view of the distributions D_a1 and D_a2.

[0069] Similar to the extents or the spreads of the distributions D_a1 and D_a2, the evaluator 23 compares the extents or the spreads of the distributions D_b1 and D_b2, D_c1 and D_c2, ,,,, and D_l1 and D_l2.

[0070] The evaluator 23 evaluates which of the first function g1 and the second function g2 is more robust, taking into account all the results of the comparison of the distributions D_a1 and D_a2, the distributions D_b1 and D_b2, the distributions D_c1 and D_c3,,, and the distributions D_l1 and D_l2. Effect of the second embodiment

[0071] As described above, the control device CD according to the second embodiment compares the extents or the dispersions of the distributions D_a1 to D_l1 with the extents or the dispersions of the distributions D_a2 to D_l2, respectively, whereby it can be evaluated which of the first function g1 and the second function g2 is more robust. Third EmbodimentConfiguration of the Third Embodiment

[0072] A control device according to a third embodiment of this disclosure will now be described with reference to the Fig. 16 to 22 described.

[0073] Fig. 16 is a block diagram showing a configuration of a controlled system CS controlled by a controller CD according to the third embodiment. Similar to the controller CD of the first embodiment, the controller CD of the third embodiment controls both a configuration and an operation of the first controlled system CS.

[0074] As in Fig. As shown in Figure 16, the controlled system CS includes a supervised learning model f and a function g, where the function g is arranged or located in front of the supervised learning model f. In the third embodiment, an engineer E(B) wishes to assign a selection of values ​​to a plurality of parameters a to l of the function g using training data TD1 to TDn. The above-described assignment will be described in detail later.

[0075] Fig. 17 is a block diagram showing a configuration of the control device CD according to the third embodiment from a functional point of view. As shown in Fig. 17, the control device CD according to the third embodiment comprises a procurer 31, an arranger 32, a selector 33 and an allocator 34. Their functions will be explained later with reference to the flowchart in Fig. 18 described.

[0076] Similar to the control device CD according to the first embodiment, the control device CD comprises the input circuit IC, the processing circuit PC and the output circuit OC (shown in Fig. 5). Alternatively, the control device CD according to the third embodiment may comprise the input circuit IC, the processor PR, the memory circuit SC and the output circuit OC (in Fig. 6).

[0077] Instead of the functions of the definer 11, the acquirer 12, the calculator 13, the updater 14 and the trainer 15 in the first embodiment (shown in Fig. 4) implements the processing circuit PC (shown in Fig. 5) or the processor PR (shown in Fig. 6) the functions of the procurer 31, the arranger 32, the selector 33 and the organizer 34 (shown in Fig. 17). Operation of the third embodiment

[0078] Fig. 18 is a flowchart showing an operation of the control device CD according to the third embodiment.

[0079] Fig. Figure 19 shows a variety of combinations COM(1) to COM(n) with a selection of values ​​related to a variety of parameters a to l of the function g.

[0080] In step ST41, the procurer procures 31 (shown in Fig. 17) a variety of combinations COMB(1) to COMB(n) (in Fig. 19). Each of the plurality of combinations COMB(1) to COMB(n) has a selection of values ​​relevant to the plurality of parameters a to l of the function g.

[0081] As in Fig. 19, the combination COMB(1), for example, has a selection of values ​​a(1), b(1), c(1),,,, and l(1) that are relevant for the plurality of parameters a to l of the function g. Here, the selection of values ​​a(1), b(1), c(1),,,, and l(1) are outputs returned or fed back by the supervised learning model f as a function of the training data TD1, the selection of values ​​a(2), b(2), c(2),,,, and l(2) are outputs returned by the supervised learning model f as a function of the training data TD2, ,,,; and the selection of values ​​a(n), b(n), c(n),,,, and l(n) are outputs returned by the supervised learning model f as a function of the training data TDn.

[0082] Fig. 20 shows the plurality of combinations COMB(1) to COMB(n) and a sampled point SP arranged in space.

[0083] In step ST42, the arranger 42 (as in Fig. 17) the multitude of combinations COMB(1) to COMB(n) in a space with 12 dimensions, the number of which corresponds to the number of the multitude of parameters a to l of the function g, as in Fig. 20 shown.

[0084] In step ST43, the acquirer 41 acquires a sampled point SP, e.g., using a known scheme or method as shown in Fig. 20 shown.

[0085] In step 44, the selector selects 33 (in Fig. 17) from the multitude of Fig. 20, select one that is positioned or located closest to the sampled point SP, ie the combination COMB(3).

[0086] In step 45, the allocator 34 (shown in Fig. 17) the selected combination COMB(3), more precisely, the selection of values ​​a(3), b(3), c(3),,,, and I(3) (shown in Fig. 19) contained in the selected combination COMB(3) to the plurality of parameters a to l of the function g. Effect of the third embodiment

[0087] As described above, according to the third embodiment, the control device CD selects one of the plurality of combinations COMB(1) to COMB(n) that is placed or located closest to the sampled point SP, and specifically selects the combination COMB(3) that enables the selection of values ​​a(3), b(3), c(3),,,, and l(3) that are possibly suitable for the plurality of parameters a to l of the function g. Modification of the third embodiment

[0088] Fig. Figure 21 shows the plurality of combinations COMB(1) to COMB(n) and an unexpected sampled point SP.

[0089] The above known scheme or method applied to the sampled point SP may produce an unexpected sampled point USP (shown in Fig. 21) which is positioned or located further or furthest away from a group of several combinations, e.g. COMB(3), COM(6), COMB(2) and COMB(1), instead of the sampled point SP (shown in Fig. 20), which is more likely to be expected in the upper group. If only the unexpected sampled point USP is available for selection in step ST44 above, the assignment in step ST45 cannot provide suitable values ​​for the plurality of parameters a to l of the function g.

[0090] Fig. 22 shows the plurality of combinations COMB(1) to COMB(n) and the sampled points SP1, SP2 and SP3.

[0091] To increase the accuracy of both selecting and matching, the procurer procures 31 (as in Fig. 17) in the modification of the third embodiment as shown in Fig. 22, a plurality of sampled points, for example, three sampled points SP1, SP2 and SP3 instead of the one sampled point SP in the third embodiment (as shown in Fig. 20). Even if the sampled point SP2 is unexpected, at least one of the other sampled points SP1 and SP3 may possibly be expected, allowing the selector 33 to select the combination COMB(3) placed or located closest to the sampled point SP1, or the combination COMB(1) placed or located closest to the sampled point SP3. In this way, the selection of the values ​​contained in the combination COMB(3) or COMB(1) that may be appropriate can be assigned to the plurality of parameters a to l of the function g, similarly to the third embodiment. QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] JP 2019-219804

[0002] JP 2019-219804 A

[0005]

Claims

[1] Method comprising: Defining a function g with a plurality of parameters a to l; Placing the function g in front of a supervised learning model f; where the supervised learning model f is trained with training data TDi containing an example xi and a label yi, and i is an integer; Define a function f(g(xi)) by using the supervised learning model f and the function g; Calculate a result Zi by calculating the function f(g(xi)); Calculate a difference Li between the result Zi and the label yi; Calculating a partial differential PDi of the difference Li for the plurality of parameters a to l of the function g; Updating the plurality of parameters a to l of the function g according to the partial differential PDi; defining a pseudo-preprocessing h using the plurality of parameters a to l of the function g when the difference Li converges during an iteration between the calculation of the result Zi and the updating of the plurality of parameters a to l of the function g; Placing the pseudo-preprocessing h before the supervised learning model f; and Train the supervised learning model f by providing the training data TDi for pseudo-preprocessing h. [2] Method comprising: Obtaining a first plurality of combinations COMB_a1 to COMB_l1, each comprising a selection of values ​​for one of a plurality of parameters a to l of a first function g1 arranged in front of a first supervised learning model f1 that implements feedback to the first function g1, by providing training data TDi for the first supervised learning model f1, where i is an integer; Obtaining a second plurality of combinations COMB_a2 to COMB_l2, each comprising a selection of values ​​for one of a plurality of parameters a to l of a second function g2 arranged in front of a second supervised learning model f2 that implements feedback to the second function g2, by providing training data TDi to the second supervised learning model f2, wherein the operation of the first supervised learning model f1 and the operation of the second supervised learning model f2 are equivalent to each other; Calculating a first plurality of distributions D_a1 to D_l1 for each of the first plurality of combinations COMB_a1 to COMB_l1; Calculating a second plurality of distributions D_a2 to D_l2 for the second plurality of combinations COMB_a2 to COMB_l2; and Evaluate which of the first function g1 and the second function g2 is more robust by comparing the first plurality of distributions D_a1 to D_l1 and the second plurality of distributions D_a2 to D_l2, respectively. [3] Method comprising: Obtaining a plurality of combinations COMB(1) to COMB(n), each having a plurality of values ​​for a plurality of parameters a to l of a function g arranged in front of a supervised learning model f that implements feedback to the first function g, by providing training data TDi to the supervised learning model f, where n is an integer equal to or greater than two; Arranging the plurality of combinations COMB(1) to COMB(n) in a space having a plurality of dimensions, the number of which is equivalent to a number of the plurality of parameters a to l of the function g; Obtaining at least one sampled point SP in space; Selecting at least one combination from the plurality of combinations COMB(1) to COMB(n) that is closest to the at least one sampled point SP; and Assigning the plurality of values ​​contained in the selected combination from the plurality of combinations COMB(1) to COMB(n) to the plurality of parameters a to l of the function g. [4] Facility comprising: a first definer that defines a function g with a plurality of parameters a to l; a first arranger that arranges the function g before a supervised learning model f; wherein the supervised learning model f is trained with training data TDi containing an example xi and a label yi, and i is an integer; a second definer that defines a function f(g(xi)) by using the supervised learning model f and the function g; a first calculator that calculates a result Zi by calculating the function f(g(xi)); a second calculator that calculates a difference Li between the result Zi and the label yi; a third calculator that calculates a partial differential PDi of the difference Li for the plurality of parameters a to l of the function g; an updater that updates the plurality of parameters a to l of the function g according to the partial differential PDi; a third definer that defines a pseudo preprocessing h using the plurality of parameters a to l of the function g when the difference Li converges during an iteration between the calculation of the result Zi and the updating of the plurality of parameters a to l of the function g; a second arranger that arranges the pseudo-preprocessing h before the supervised learning model f; and a trainer that trains the supervised learning model f by providing the training data TDi for the pseudo-preprocessing h. [5] Facility comprising: a first acquirer that acquires a first plurality of combinations COMB_a1 to COMB_l1, each comprising a selection of values ​​for one of a plurality of parameters a to l of a first function g1 arranged in front of a first supervised learning model f1 that implements feedback to the first function g1, by providing training data TDi to the first supervised learning model f1, where i is an integer; a second acquirer that acquires a second plurality of combinations COMB_a2 to COMB_l2, each comprising a selection of values ​​for one of a plurality of parameters a to l of a second function g2 arranged in front of a second supervised learning model f2 that implements feedback to the second function g2 by providing training data TDi to the second supervised learning model f2, wherein the operation of the first supervised learning model f1 and the operation of the second supervised learning model f2 are equivalent to each other; a first calculator that calculates a first plurality of distributions D_a1 to D_l1 for the first plurality of combinations COMB_a1 to COMB_l1, respectively; a second calculator which calculates a second plurality of distributions D_a2 to D_l2 for the second plurality of combinations COMB_a2 to COMB_l2; and an evaluator that evaluates which of the first function g1 and the second function g2 is more robust by comparing the first plurality of distributions D_a1 to D_l1 and the second plurality of distributions D_a2 to D_l2, respectively. [6] Facility comprising: a first acquirer that acquires a plurality of combinations COMB(1) to COMB(n) each having a plurality of values ​​for a plurality of parameters a to l of a function g arranged in front of a supervised learning model f that implements feedback to the first function g, by providing training data TDi to the supervised learning model f, where n is an integer equal to or greater than two; an arranger that arranges the plurality of combinations COMB(1) to COMB(n) in a space having a plurality of dimensions, the number of which is equivalent to a number of the plurality of parameters a to l of the function g; a second procurer that procures at least one sampled point SP in space; a selector that selects at least one of the plurality of combinations COMB(1) to COMB(n) that is closest to the at least one sampled point SP; and an allocator that allocates the plurality of values ​​contained in the selected combination from the plurality of combinations COMB(1) to COMB(n) to the plurality of parameters a to l of the function g.

Images