Learning device, learning method, and program

Abstract

To enable selective forgetting to be realized further effectively.SOLUTION: A learning device includes: a data input unit configured to accept input of first data that is a learning target, second data for identifying the first data, and second past data that is used as data for identifying the first data at the time of learning in the past and related to learning contents to be saved; a feature extraction unit configured to extract a feature from the first data; a synthetic data generation unit configured to generate synthetic data by synthesizing the feature and the second data; and a parameter update unit configured to, based on a feature of the synthesized data obtained by inputting the synthesized data and the second past data into a machine learning model, and the second past data, update a parameter of the machine learning model.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to a learning device, a learning method, and a program. [Background technology]

[0002] In machine learning, a method called continuous learning is being researched that can additionally learn new classes without reducing the classification accuracy for previously learned classes. When additional learning of a new class is performed, the classification accuracy for previously learned classes dramatically deteriorates. This phenomenon is called catastrophic forgetting.

[0003] On the other hand, continuing to remember all previously learned classes is not always beneficial from the standpoint of privacy protection and data leakage prevention, and so there are hopes for the realization of "selective forgetting," which can explicitly select classes to be remembered and classes to be forgotten. One method for realizing selective forgetting is disclosed in Non-Patent Document 1, for example. In the method disclosed in Non-Patent Document 1, the classes to be remembered and the classes to be forgotten are controlled by superimposing a class-specific random signal on the learning sample. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Shibata, Irie, Ikami, Mitsuzumi, "Learning with selective forgetting," Proc. International Joint Conference on Artificial Intelligence (IJCAI), 2021. Summary of the Invention [Problem to be solved by the invention]

[0005] However, in machine learning-based classification, classification is made easier by extracting features that remove information unrelated to classification from data and leave only useful information. Therefore, in the method disclosed in Non-Patent Document 1, class-specific signals may be lost in the process of extracting features. The problem to be solved by the present invention is to make it possible to realize selective forgetting more effectively. [Means for solving the problem]

[0006] One aspect of the present invention is a learning device comprising: a data input unit that inputs first data to be learned, second data for identifying the first data, and second past data related to learning content to be saved, the second past data being used during past learning as data for identifying the first data; a feature extraction unit that extracts features from the first data; a synthetic data generation unit that generates synthetic data by synthesizing the features and the second data; and a parameter update unit that updates parameters of the machine learning model based on the synthetic data obtained by inputting the synthetic data and the second past data into a machine learning model and features of the second past data. Effect of the Invention

[0007] According to the present invention, selective forgetting can be realized more effectively. [Brief description of the drawings]

[0008] [Figure 1] 1 is a block diagram showing a functional configuration of a learning device 1 in an embodiment. [Diagram 2] FIG. 1 is a diagram for explaining the object of the present invention. [Diagram 3] FIG. 2 is a diagram illustrating the function of a control unit 12 in the embodiment. [Figure 4A] FIG. 1 illustrates a standard normalization layer and a linear transformation layer. [Figure 4B] FIG. 1 illustrates a standard normalization layer and a linear transformation layer. [Diagram 5] 4 is a flowchart showing the flow of a learning process of the learning device 1 in the embodiment. [Figure 6] FIG. 13 is a diagram showing the value of Sk in each method. [Figure 7] 2 is a diagram illustrating an example of a hardware configuration of a discrimination device 2 in an embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. 1 is a block diagram showing the functional configuration of a learning device 1 according to an embodiment. The learning device 1 according to an embodiment is a device that performs so-called continuous learning, enabling continuous learning. The learning device 1 updates a predetermined machine learning model by machine learning until a predetermined termination condition is satisfied.

[0010] The predetermined machine learning model at the time when the predetermined termination condition is satisfied is the trained model. Therefore, the learning device 1 acquires the trained model by updating the predetermined machine learning model by machine learning until the predetermined termination condition is satisfied.

[0011] For ease of explanation, hereinafter, performing machine learning is also referred to as learning. Updating a machine learning model (hereinafter referred to as a "machine learning model") by machine learning means suitably adjusting the parameter values ​​in the machine learning model. In the following explanation, learning to be A means that the parameter values ​​in the machine learning model are adjusted to satisfy A. A represents a condition. In addition, hereinafter, "for learning" means that it is used to update the machine learning model. Note that a machine learning model is a set including one or more processes whose execution conditions and order are predetermined.

[0012] The machine learning model is represented by, for example, a neural network. A neural network is a circuit such as an electronic circuit, an electric circuit, an optical circuit, or an integrated circuit that represents a machine learning model. The parameters of the neural network are suitably adjusted based on the loss, and the parameters of the network are the parameters of the machine learning model that is represented. The parameters of the network are the parameters of the circuits that constitute the network.

[0013] The machine learning model updated by the learning device 1 is a machine learning model that identifies input data. The trained model obtained by the learning device 1 updating parameters is, for example, a model that performs data classification. Learning input data, learning reference data, and learning pseudo data are input to the machine learning model.

[0014] The learning input data is data to be learned. For example, the learning input data may be two-dimensional or more data such as an image, one-dimensional time series data such as voice, sound, or vibration, or symbolic data such as natural language. In this way, the learning input data may be any data that is generally applicable to machine learning. When one-dimensional time series data such as voice, sound, or vibration is used as the learning input data, it is sufficient to convert it to two dimensions by performing a short-time Fourier transform or the like to obtain a spectrogram.

[0015] The learning reference data is so-called correct answer data in machine learning. The learning reference data may be information equivalent to a class label such as an attribute, a class name, or a person ID that has been quantified, or may be data similar to the learning input data, i.e., image or voice data. The learning reference data may be any objective variable that is generally applied to machine learning. More specifically, when a label is used as the objective variable, the objective is identification or authentication, and when data similar to the learning input data is used as the variable, the objective is regression.

[0016] Hereinafter, data including at least one pair of learning input data and one pair of learning reference data will be referred to as learning data. That is, the learning data is data including at least a pair of learning input data and learning reference data, and is an example of so-called learning data. The learning data is one aspect of the first data.

[0017] The learning pseudo data is data different from the learning data. For example, the learning pseudo data is a randomly generated control code called a mnemonic code. The learning pseudo data is, for example, data of a random noise pattern. The learning pseudo data is, for example, data of a pattern in which a part of the learning input data is degraded or data of a pattern represented by the average value of the learning input data is convolved to the same size as the feature extracted from the learning input data. It may be. The pattern represented by the average value of the learning input data may be a pattern in which each pixel is represented by the average value of each pixel in a plurality of learning input data, or a pattern in which the average value of each pixel of one learning input data is assigned to each pixel. The learning pseudo data may be data in which a sample whose safety is guaranteed from the viewpoint of privacy, etc., among data of past tasks is convolved to the same size as the feature extracted from the learning input data, or may be a synthetic image generated by a generative model obtained by adversarial learning, etc. The learning pseudo data is data for identifying a class to which the learning data belongs. The learning pseudo data is one aspect of the second data.

[0018] In this specification, an example is described in which image data is mainly handled as data input to the learning device 1, but the present invention is not limited thereto. In this specification, a case in which deep learning is used is described.

[0019] In the following description, for the sake of clarity, the set of input data will be expressed as in equation (1).

[0020]

number

[0021] k is the task ID, K is the total number of tasks, D k is the data used in the kth task.

[0022] For example, each data set D k is the training input data x k and the reference data for learning y k Using the above, we express it as in the following formula (2). i k is the input data for learning. y i k is the reference data for training. i is the index of the class label.

[0023]

number

[0024] As mentioned above, the purpose of the present invention is to learn to selectively forget unnecessary knowledge (classes in this case) from past tasks while preserving other knowledge (classes in this case). In the following explanation, we will call the classes we want to preserve (hereinafter referred to as the "preserved set") C P k In this case, the class to be selectively forgotten (hereafter called the "deletion set") is C P k The complement of , i.e., -C P k (- is placed above C).

[0025] The network in the learning device 1 is represented by f, and the parameters are formally represented by θ. That is, the network is θ This is expressed as:

[0026] The objective of the present invention can be expressed as follows using the above notation: PIf the input learning data x belongs to the correct class y, the input learning data x is mapped to the correct class y. On the other hand, if the correct class y belongs to the preserved class C P If the input data x does not belong to , then f is used to map the input data x to a class other than the correct class y. θ Learn.

[0027] In achieving the above object, the learning device 1 may set the following constraints.

[0028] (constraint conditions) When learning the kth task, the past training input data {D1,…,D k-1 Do not use}.

[0029] Formally, these objectives can be expressed as shown in Fig. 2. Fig. 2 is a diagram for explaining the objectives of the present invention. In the following explanation, a case where the above-mentioned constraints are applied will be described, but the present invention is also applicable in cases where the above-mentioned constraints are not applied.

[0030] (Specific configuration of learning device 1) The learning device 1 includes an input unit 11, a control unit 12, a communication unit 13, a storage unit 14, and an output unit 15.

[0031] The input section 11 is made up of a data input section 111 , a pseudo data input section 112 and a pseudo data re-input section 113 .

[0032] The data input unit 111 inputs the data set D shown in the above formula (2) in the k-th task. k That is, in the k-th task, the data input unit 111 inputs the learning input data x i k and the reference data for learning y i k Enter the set with .

[0033] The pseudo data input unit 112 inputs the learning pseudo data. In the following, the set of learning pseudo data is denoted by {ξ k,c} where ξk,c is the learning pseudo-data of the c-th class in the k-th task. The learning pseudo-data input to the pseudo-data input unit 112 is a form of the second data.

[0034] The pseudo-data re-input unit 113 re-inputs the learning pseudo-data input by the pseudo-data input unit 112. More specifically, the pseudo-data re-input unit 113 inputs the learning pseudo-data of {ξ k´,c}(where k´ < k) in the k-th task. That is, the pseudo-data re-input unit 113 inputs the learning pseudo-data used during past learning as data for identifying the learning data. The learning pseudo-data input to the pseudo-data re-input unit 113 is a form of the second past data.

[0035] The control unit 12 controls the entire learning device 1. The control unit 12 is configured using a processor such as a CPU (Central Processing Unit) and a memory. The control unit 12 realizes the functions of the feature extraction unit 121, the extended data generation unit 122, the identification unit 123, the loss acquisition unit 124, the parameter update unit 125, and the end determination unit 126 by executing a program.

[0036] Figure 3 is a diagram for explaining the functions of the control unit 12 in the embodiment. The feature extraction unit 121 extracts the feature s i k from the learning input data x i k、c input by the data input unit 111. The feature extraction unit 121 extracts the feature s i k by performing convolution processing and standard normalization processing on the learning input data x i k,c . s i k,c is the feature of the c-th class in the k-th task.

[0037] The extended data generation unit 122 generates extended data based on the feature s i k,cand the pseudo data ξ inputted by the pseudo data input unit 112. k,c By synthesizing the data, the extended data (synthetic data) is obtained. i k,c (~ is placed above x). The extended data generator 122 is one aspect of the composite data generator.

[0038] The extended data generator 122 generates extended data, for example, according to equation (3).

number

[0039] where ~o i k,c,hwd (~ is placed above x) is the hwd component of the extended data (the component with vertical position h, horizontal position w, and channel number d among the components output in the convolution process). s i k,c,hwd Features i k,c The hwd component of ξ k,c,hwd is, k,c The hwd component of γ d is the d-channel component of the linear transformation weights. d is the d-channel component of the bias. λ is a uniform random number in [0,1].

[0040] The extended data generator 122 may perform linear transformation on the extended data. i k,c and pseudo data ξ k,c Alternatively, the extended data may be generated by linearly transforming and then combining the two.

[0041] The identification unit 123 identifies the input data based on the extended data generated by the extended data generation unit 122 and the learning pseudo data re-input by the pseudo data re-input unit 113. Note that, although the case where the identification unit 123 in the embodiment estimates a certain class label has been described, the present invention is not limited to this. For example, the identification unit 123 may perform a regression process instead of performing an identification process. Furthermore, a case where the network structure in the identification unit 123 is a multi-head structure will be described, but the present invention is not limited to this.

[0042] The feature extraction unit 121, the extended data generation unit 122, and the classification unit 123 may use a general deep learning model such as ResNet disclosed in Reference 1. For example, in a deep learning model, the layer for generating extended data is determined, and the layer before the layer for generating the extended data is set as the feature extraction unit 121, the layer for generating the extended data is set as the extended data generation unit 122, and the layer after the layer for generating the extended data is set as the classification unit 123, thereby realizing the feature extraction unit 121, the extended data generation unit 122, and the classification unit 123. (Reference 1: K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition”, In Proc. CVPR, 2016.)

[0043] As a method for the classification by the classification unit 123, for example, a method such as a support vector machine may be used.

[0044] The layer consisting of the feature extraction unit 121, the extended data generation unit 122, and the classification unit 123 may include a layer that performs batch normalization (BN) processing (hereinafter referred to as a "batch normalization layer"). The batch normalization layer includes a layer that performs standard normalization processing (hereinafter referred to as a "standard normalization layer") and a layer that performs linear transformation (hereinafter referred to as a "linear transformation layer"). When the layer consisting of the feature extraction unit 121, the extended data generation unit 122, and the classification unit 123 includes a batch normalization layer, the layer that generates extended data is a layer that is provided between the standard normalization layer and the linear transformation layer or after the linear transformation layer.

[0045] FIG. 4A is a diagram showing a standard normalization layer and a linear transformation layer. In the layers shown in FIG. 4A, a layer that generates extended data is provided between the standard normalization layer and the linear transformation layer. In the layers shown in FIG. 4A, features are extracted from learning input data by a convolution layer (Conv layer) and a standard normalization layer (Standard Normalization). Then, the features and learning pseudo data are combined to generate extended data. Then, the extended data is linearly transformed by a linear transformation layer (Scale and Shift), and then the output may be determined by ReLU. FIG. 4B is a diagram showing a standard normalization layer and a linear transformation layer. In the layers shown in FIG. 4B, a layer that generates extended data is provided after the linear transformation layer. In the layers shown in FIG. 4B, features are extracted from learning input data by a convolution layer (Conv layer), a standard normalization layer (Standard Normalization), and a linear transformation layer (Scale and Shift). In addition, the learning pseudo data is linearly transformed by the linear transformation layer. Then, the features and the linearly transformed learning pseudo data are combined to generate extended data. Then, the output may be determined by ReLU. The process in the layer shown in FIG. 4B corresponds to the process in which the extended data generating unit 122 generates extended data by linearly transforming the features and pseudo data, respectively, and then combining them.

[0046] The loss calculation unit 124 calculates losses using the learning data input by the data input unit 111, the learning pseudo data input by the pseudo data re-input unit 113, and the results of identification by the identification unit 123. For example, the loss calculation unit 124 may include regularization for restricting the range of parameters to be updated by the parameter update unit 125.

[0047] More specifically, the operation of the loss calculator 124 will be described. The loss calculator 124 calculates the loss based on the formula (4). The total loss function L for learning shown in the formula (4) is the classification loss L C , Pseudo data loss L M , selective forgetting loss L F , regularization term L R It may consist of four terms:

[0048]

number

[0049] Classification loss L C and pseudo data loss L M The term is for learning a new task. Selective forgetting loss L F and the regularization term L R The items below are for maintaining the previous tasks. Each of these items will be explained in detail below.

[0050] Classification loss L C is introduced with the aim of accurately classifying new tasks. Specifically, the classification loss L C can be given, for example, as equation (5).

[0051]

number

[0052] In equation (5), N k is the number of training samples for the kth task, and l(f(x i k ),y ik ) is the learning input data x i k The data finally output from the discrimination unit 123 and the learning reference data y i k (e.g., class label y). For example, the loss function can be softmax cross entropy (CE) or additive margin softmax (AMS) loss.

[0053] Simulated Data Loss L M is the training input data x i k Classification loss L C In addition to the loss L, another loss using augmented data with the training pseudo data may be used to link each training pseudo data to the corresponding class. M is expressed as the following equation (6).

[0054]

number

[0055] The loss function is the classification loss L C The same one may be used, or a different one may be used.

[0056] Selective forgetting loss L F is the preserved set C P k Then, only the classes belonging to the deletion set -C P k (- is placed above C) is introduced for the purpose of forgetting classes. Specifically, this is achieved by learning only with pseudo data for training that correspond to classes in the retention set, and discarding other pseudo data for training (e.g., pseudo data for training that correspond to classes in the deletion set). More specifically, x i p The pseudo data for training to generate ξ i pThen, the loss can be given by, for example, the following equation (7).

[0057]

number

[0058] In equation (7), N p is the number of training samples in the pth task, γ SF is the weight for adjusting the strength of this loss. This loss function is i k Do not use any of the above. And, the deletion set-C P k By not using the training pseudo data for classes in (- is placed above C), these classes are catastrophically forgotten. This causes the training input data x i k It is possible to realize selective forgetting of previous tasks without using any

[0059] Regularization term L R can be achieved by using common regularization techniques (e.g., refs. 2-4) that are introduced to prevent catastrophic forgetting.

[0060] (Reference 2: James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, et al, “Overcoming catastrophic forgetting in neural networks”, PNAS, 114(13):3521-3526, 2017.)

[0061] (Reference 3: Rahaf Aljundi, Francesca Babiloni, Mohamed Elhoseiny, Marcus Rohrbach, and Tinne Tuytelaars. “Memory Aware Synapses: Learning what (not) to forget”, ECCV, pages 139-154, 2018.) (Reference 4: Li, Hoiem, “Learning without forgetting,” IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), Vol.40, No.12, pp.2935-2947, 2017.)

[0062] In this embodiment, the regularization term is the preserved set C p k Since it is necessary to prevent catastrophic forgetting only for these preserved sets, the method may be modified to provide regularization only for these preserved sets. More specifically, for example, in the case of Reference 4, the following formula (8) may be used as the regularization term.

[0063]

number

[0064] In equation (8), γ is the weight of the term, and i is the index of the class label. Here, the saved set C P k i∈C P k Change the sum as follows: y in Eq. (8) ´ o (i) and^y ´ o (i) (^ above y) is the correction between the recorded probability and the current probability. In the case of Reference 2, the following equation (9) can be used as the regularization term.

[0065]

number

[0066] In equation (9), γ is the weight of the regularization term, Γ p,q is the q-th parameter on the diagonal ^θ p,q (^ is placed on θ) is the element of the Fisher matrix of the pth previous task. Note that the saved set C P k The Fisher matrix may be modified so that the regularization term is evaluated only for the class corresponding to the regularization term. As the regularization term, a combination of these multiple regularizations may be used by linear addition or the like.

[0067] The parameter update unit 125 updates the machine learning model so that the loss acquired by the loss acquisition unit 124 is reduced. The update of the machine learning model based on the loss is specifically a process of updating the parameter values ​​of a neural network representing the machine learning model according to a predetermined rule based on the loss. More specifically, the update of the parameter values ​​of the neural network is, for example, an update of the parameter values ​​corresponding to the feature extraction unit 121, the extended data generation unit 122, and the identification unit 123. When updating the parameters, the parameter update unit 125 may update the parameters using an existing method such as a stochastic gradient method.

[0068] The end determination unit 126 determines whether a predetermined end condition is satisfied. The predetermined end condition may be set in advance. For example, the predetermined end condition may be that the loss is reduced by a threshold or more, or that the parameters are updated a predetermined number of times.

[0069] The communication unit 13 includes a communication interface for connecting the learning device 1 to an external device. The communication unit 13 communicates with the external device via wired or wireless communication. The external device is, for example, a device that transmits learning data.

[0070] The storage unit 14 is configured using a non-transitory computer-readable storage medium device such as a magnetic hard disk device or a semiconductor storage device. The storage unit 14 stores various information related to the learning device 1. The storage unit 14 stores information input via, for example, the input unit 11 or the communication unit 13. The storage unit 14 stores, for example, a machine learning model. The storage unit 14 stores, for example, various information generated by the execution of the machine learning model.

[0071] It should be noted that the learning data and the learning pseudo data do not necessarily have to be input only to the input unit 11, and do not necessarily have to be input only to the communication unit 13. The learning data and the learning pseudo data may be input from either the input unit 11 or the communication unit 13. For example, the learning data may be input to the input unit 11, and the learning pseudo data may be input to the communication unit 13. The learning data and the learning pseudo data may be stored in advance in the memory unit 14.

[0072] The output unit 15 outputs various information. The output unit 15 includes a display device such as a CRT (Cathode Ray Tube) display, a liquid crystal display, or an organic EL (Electro-Luminescence) display. The output unit 15 may be configured as an interface that connects these display devices to the learning device 1. The output unit 15 outputs information input to the input unit 11, for example. The output unit 15 may display learning data input to the input unit 11 or the communication unit 13, for example. The output unit 15 may display the execution result of a machine learning model, for example.

[0073] FIG. 5 is a flowchart showing the flow of the learning process of the learning device 1 in the embodiment. The data input unit 111 inputs learning input data for a new task (step S101). The data input unit 111 outputs the input learning input data to the control unit 12. The pseudo data input unit 112 inputs learning pseudo data corresponding to the new task (step S102). The pseudo data input unit 112 outputs the input learning pseudo data to the control unit 12. The pseudo data re-input unit 113 inputs learning pseudo data corresponding to a past task (step S103). The pseudo data re-input unit 113 outputs the input learning pseudo data to the control unit 12.

[0074] The feature extraction unit 121 acquires the learning input data output from the data input unit 111. The feature extraction unit 121 extracts features from the learning input data (step S104). The feature extraction unit 121 outputs the generated features to the extended data generation unit 122.

[0075] The extended data generating unit 122 generates extended data based on the features output from the feature extracting unit 121 and the learning pseudo data output from the pseudo data input unit 112 (step S105). The identifying unit 123 performs identification based on the extended data generated by the extended data generating unit 122 and the learning pseudo data corresponding to the past task output from the pseudo data re-input unit 113 (step S106). The loss acquiring unit 124 acquires a loss based on the identification result obtained by the identifying unit 123, the learning input data output from the data input unit 111, and the learning pseudo data corresponding to the past task output from the pseudo data re-input unit 113 (step S107). For example, the loss acquiring unit 124 acquires the loss by the above formula (4).

[0076] The parameter update unit 125 updates the parameters of the machine learning model based on the loss acquired by the loss acquisition unit 124 (step S108). The termination determination unit 126 determines whether or not the termination condition is satisfied (step S109). If the termination condition is not satisfied (step S109-NO), the learning device 1 executes the processes from step S104 onward.

[0077] On the other hand, if the termination condition is satisfied (step S109-YES), the data output section 30 outputs the classification result of the classification section 123 (step S110).

[0078] Next, the experimental results of this embodiment will be described. As the datasets, the object image recognition dataset CIFAR-100 (https: / / www.cs.toronto.edu / ~kriz / cifar.html), the bird image recognition dataset CUB-200-2011 (Wah, Branson, Welinder, Perona, Belongie, “The Caltech-UCSD Birds-200-2011 dataset,” 2011.), and the car image recognition dataset Stamford Cars (Krause, Stark, Deng, Li, “3D object representation for fine-grained categorization,” Proc. IEEE / CVF International Conference on Computer Vision Workshop, 2013.) were used. CIFAR-100 is a dataset consisting of 60,000 object images in 100 classes, CUB-200-2011 is a dataset consisting of approximately 12,000 bird images in 200 classes, and Stanford Cars is a dataset consisting of approximately 16,000 car images in approximately 200 classes.

[0079] ResNet-18 was used as the classification model. Vanilla, LwF, LwF*, EWC, EWC*, EWC*+LwF*, MAS, MAS+LwF*, and LSF were used as comparison methods for the method of this embodiment. M , LSF E The experiment was carried out using a technique called Vanilla is a classification loss L C LwF is a method disclosed in Reference 4. LwF* is a method that improves LwF and is disclosed in Non-Patent Document 1.

[0080] EWC is a method disclosed in Reference 2. EWC* is an improved version of EWC, and is disclosed in Non-Patent Document 1. EWC*+LwF* is a combination of EWC* and LwF*, and is disclosed in Non-Patent Document 1.

[0081] MAS is a method disclosed in Reference 3. MAS+LwF* is a method combining MAS and LwF*, and is disclosed in Non-Patent Document 1. LSF M is a method that combines EWC* and LwF* with the method proposed in Non-Patent Document 1. E is a method that combines the method proposed in Non-Patent Document 1 with MAS and LwF*.

[0082] As an evaluation index for each method, we used the LSFM (Learning with Selective Forgetting Measure) S described in Non-Patent Document 1. S k The calculation method is disclosed in Non-Patent Document 1.

[0083] Figure 6 shows the S k The value of S in this embodiment (StandMix) is larger than the value of S in any of the methods. Therefore, it can be said that this embodiment can control memory and forgetting more effectively than the compared methods.

[0084] 7 is a diagram showing an example of a hardware configuration of the classification device 2 in an embodiment. The classification device 2 performs classification processing using a trained model obtained by the learning device 1. The classification device 2 includes a control unit 21 including a processor 26 such as a CPU and a memory 27 connected by a bus, and executes a program. The classification device 2 functions as a device including the control unit 21, an input unit 22, a communication unit 23, a storage unit 24, and an output unit 25 by executing the program.

[0085] More specifically, the processor 26 reads out the program stored in the storage unit 24, and stores the read out program in the memory 27. The processor 26 executes the program stored in the memory 27, whereby the identification device 2 functions as a device including the control unit 21, the input unit 22, the communication unit 23, the storage unit 24, and the output unit 25.

[0086] The control unit 21 controls the operation of various functional units included in the classification device 2. The control unit 21 classifies input data, for example, by using a trained model obtained by the learning device 1. Specifically, the control unit 21 identifies the data by inputting the data input to the classification device 2 into the trained model. For example, the control unit 21 performs class classification of the data.

[0087] The input unit 22 includes input devices such as a mouse, a keyboard, and a touch panel. The input unit 22 may be configured as an interface that connects these input devices to the device itself. The input unit 22 accepts input of various information for the device itself. The input unit 22 accepts, for example, an input instructing the start of processing. The input unit 22 accepts, for example, an input of data to be processed.

[0088] The communication unit 23 includes a communication interface for connecting the device to an external device. The communication unit 23 communicates with the external device via a wired or wireless connection. The external device with which the communication unit 23 communicates is, for example, an output destination of the processed data. In such a case, the communication unit 23 outputs the processed data to the external device by communicating with the external device.

[0089] The communication unit 23 may communicate with, for example, the learning device 1. In this case, the communication unit 23 acquires a trained model obtained by, for example, the learning device 1.

[0090] The storage unit 24 is configured using a non-transitory computer-readable storage medium device such as a magnetic hard disk device or a semiconductor storage device. The storage unit 24 stores various information related to the identification device 2. The storage unit 24 stores, for example, a trained model acquired via the communication unit 23. The storage unit 24 stores, for example, data to be processed that is input to the input unit 22.

[0091] The output unit 25 outputs various information. The output unit 25 includes a display device such as a CRT display, a liquid crystal display, or an organic EL display. The output unit 25 may be configured as an interface that connects these display devices to the output unit 25. The output unit 25 outputs information input to the input unit 22, for example.

[0092] The learning device 1 may be implemented using a plurality of information processing devices communicably connected via a network. In this case, each functional unit of the learning device 1 may be distributed and implemented in the plurality of information processing devices.

[0093] All or part of the functions of the learning device 1 and the identification device 2 may be realized using hardware (including circuitry) such as an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array). The program may be recorded on a computer-readable recording medium. Examples of computer-readable recording media include portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, and storage devices such as hard disks built into a computer system. The program may be transmitted via a telecommunications line.

[0094] Although an embodiment of the present invention has been described in detail above with reference to the drawings, the specific configuration is not limited to this embodiment, and designs that do not deviate from the gist of the present invention are also included. [Explanation of symbols]

[0095] 1...Learning device, 2...Classification device, 11...Input unit, 12...Control unit, 13...Communication unit, 14...Memory unit, 15...Output unit, 21...Control unit, 22...Input unit, 23...Communication unit, 24...Memory unit, 25...Output unit, 26...Processor, 27...Memory, 111...Data input unit, 112...Pseudo data input unit, 113...Pseudo data re-input unit, 121...Feature extraction unit, 122...Augmented data generation unit, 123...Classification unit, 124...Loss acquisition unit, 125...Parameter update unit, 126...Termination determination unit

Claims

1. a data input unit for inputting first data to be learned, second data for identifying the first data, and second past data used in past learning as data for identifying the first data, the second past data being related to learning content to be saved; a feature extraction unit that extracts features from the first data; a synthetic data generating unit for generating synthetic data by synthesizing the feature and the second data; a parameter update unit that updates parameters of the machine learning model based on features of the synthetic data obtained by inputting the synthetic data and the second past data into a machine learning model and the second past data; A learning device comprising:

2. The feature extraction unit extracts features from the first data by convolution processing and standard normalization processing; The synthetic data generation unit generates the synthetic data by synthesizing the feature and the second data and performing a linear transformation process. The learning device according to claim 1 .

3. The feature extraction unit extracts features from the first data by convolution processing, standard normalization processing, and linear transformation processing; The synthetic data generating unit generates synthetic data by synthesizing the feature and the linearly transformed second data. The learning device according to claim 1 .

4. an identification unit that identifies the first data by using features of the composite data and the second past data; a loss acquiring unit that acquires a loss by using the first data, the second past data, and a result of the identification by the identification unit; Further comprising: A learning device according to any one of claims 1 to 3.

5. the loss acquisition unit acquires the loss based on regularization that restricts a range of parameters updated by the parameter update unit, a loss based on a classification accuracy in the synthetic data, a loss based on a classification accuracy in the first data, and a loss based on a classification accuracy in the second data. The learning device according to claim 4.

6. inputting first data to be learned, second data for identifying the first data, and second past data used in past learning as data for identifying the first data, the second past data relating to learning content to be saved; Extracting features from the first data; generating combined data by combining the features and the second data; updating parameters of the machine learning model based on features of the synthetic data obtained by inputting the synthetic data and the second past data into a machine learning model and the second past data; How to learn.

7. A program for causing a computer to operate as the learning device according to any one of claims 1 to 4.

Images