CN118378076A - Information processing device, information processing method and computer readable storage medium - Google Patents

Abstract

An information processing device, an information processing method and a computer-readable storage medium are disclosed. The information processing device may include: a sample generation unit configured to generate a first simulation sample set for a first task using a sample generation network obtained based on a second training sample set for a second task; and a first model training unit configured to retrain a first model pre-trained for the first task using the second training sample set under a first predetermined condition to obtain a retrained first model for predicting an object to be predicted. The first predetermined condition includes: the difference between the prediction result of the pre-trained first model for the first simulation sample set and the prediction result of the retrained first model for the first simulation sample set is within a predetermined range. The first task is different from the second task. The object to be predicted involves the first task and/or the second task.

Description

Information processing device, information processing method and computer readable storage medium

Technical Field

The present disclosure relates to the field of information processing, and in particular to an information processing device, an information processing method, and a computer-readable storage medium.

Background technique

In the field of information processing, lifelong (continuous) learning has important applications. However, when a new training sample set is used to further train the original model, the new model may lose the performance of the original model.

Summary of the invention

A brief overview of the disclosure is given below in order to provide a basic understanding of certain aspects of the disclosure. However, it should be understood that this overview is not an exhaustive overview of the disclosure. It is not intended to identify the key or important parts of the disclosure, nor is it intended to limit the scope of the disclosure. Its purpose is simply to give certain concepts of the disclosure in a simplified form as a prelude to a more detailed description given later.

In view of the above problems, an object of the present disclosure is to provide an information processing device, an information processing method, and a computer-readable storage medium that can solve one or more disadvantages in the prior art.

According to one aspect of the present disclosure, there is provided an information processing device, comprising: a sample generation unit, configured to generate a first simulation sample set for a first task using a sample generation network obtained based on a second training sample set for a second task; and a first model training unit, configured to retrain a first model pre-trained for the first task using the second training sample set under a first predetermined condition to obtain a retrained first model for predicting an object to be predicted. The first predetermined condition includes: the difference between the prediction result of the pre-trained first model for the first simulation sample set and the prediction result of the retrained first model for the first simulation sample set is within a predetermined range. The first task is different from the second task. The object to be predicted involves the first task and/or the second task.

According to another aspect of the present disclosure, there is provided an information processing method, comprising: using a sample generation network obtained based on a second training sample set for a second task to generate a first simulation sample set for a first task; and under a first predetermined condition, using the second training sample set to retrain a first model pre-trained for the first task to obtain a retrained first model for predicting an object to be predicted. The first predetermined condition comprises: the difference between the prediction result of the pre-trained first model for the first simulation sample set and the prediction result of the retrained first model for the first simulation sample set is within a predetermined range. The first task is different from the second task. The object to be predicted involves the first task and/or the second task.

According to other aspects of the present disclosure, a computer program code and a computer program product for implementing the method according to the present disclosure are also provided, as well as a computer-readable storage medium having the computer program code for implementing the method according to the present disclosure recorded thereon.

Other aspects of the embodiments of the present disclosure are given in the following description, wherein the detailed description is used to fully disclose the preferred embodiments of the embodiments of the present disclosure without imposing limitations thereon.

BRIEF DESCRIPTION OF THE DRAWINGS

The present disclosure may be better understood by referring to the detailed description given below in conjunction with the accompanying drawings, wherein the same or similar reference numerals are used throughout the drawings to represent the same or similar components. The accompanying drawings, together with the following detailed description, are included in and form a part of the present specification to further illustrate the preferred embodiments of the present disclosure and to explain the principles and advantages of the present disclosure. Among them:

FIG. 1 is a block diagram showing a functional configuration example of an information processing apparatus according to a first embodiment of the present disclosure.

FIG. 2 is a block diagram showing a functional configuration example of an information processing device according to a second embodiment of the present disclosure.

3 is a schematic diagram showing an adjustment process performed by a sample generation network acquisition unit according to a second embodiment of the present disclosure;

FIG4 shows an example of a second training sample and an example of a simulation sample;

FIG. 5 is a block diagram showing a functional configuration example of an information processing device according to a third embodiment of the present disclosure.

6 is a schematic diagram showing a training process performed by a sample generation network acquisition unit according to a third embodiment of the present disclosure;

FIG7 shows an example of a second training sample and further examples of simulated samples;

FIG8 is a diagram showing a comparison between the technology according to the present disclosure and the related art;

9 is a flowchart showing an example of the flow of an information processing method according to a fourth embodiment of the present disclosure; and

FIG. 10 is a block diagram showing an example structure of a personal computer that can be employed in the embodiments of the present disclosure.

Detailed ways

Exemplary embodiments of the present disclosure will be described below in conjunction with the accompanying drawings. For the sake of clarity and conciseness, not all features of the actual implementation are described in the specification. However, it should be understood that many implementation-specific decisions must be made in the process of developing any such actual implementation in order to achieve the developer's specific goals, such as meeting those constraints related to the system and business, and these constraints may vary from implementation to implementation. In addition, it should be understood that although the development work may be very complex and time-consuming, it is only a routine task for those skilled in the art who benefit from the content of this disclosure.

It is also necessary to explain here that in order to avoid obscuring the present disclosure due to unnecessary details, only the device structure and/or processing steps closely related to the scheme according to the present disclosure are shown in the accompanying drawings, while other details that are not closely related to the present disclosure are omitted.

The embodiments according to the present disclosure are described in detail below with reference to the accompanying drawings.

<First Embodiment>

1 is a block diagram showing a functional configuration example of an information processing device 100 according to a first embodiment of the present disclosure. As shown in FIG1 , the information processing device 100 according to the first embodiment of the present disclosure may include a sample generation unit 102 and a first model training unit 104 .

The sample generating unit 102 may be configured to generate a first simulation sample set for the first task using a sample generating network obtained based on a second training sample set for the second task.

The first model training unit 104 may be configured to retrain the first model pre-trained for the first task using the second training sample set under the first predetermined condition to obtain a retrained first model for predicting the object to be predicted. For example, the first predetermined condition may include: the difference between the prediction result of the pre-trained first model for the first simulation sample set and the prediction result of the retrained first model for the first simulation sample set is within a predetermined range. The predetermined range may be obtained through a limited number of experiments or set based on experience.

For example, the second training sample set may be a training image set, a training voice set, etc. Correspondingly, the first simulation sample set generated based on the second training sample set may be an image set, a voice set, etc., and the object to be predicted may be an image, a voice, etc. For example, an image may include a facial image, an image of an animal, an image of a plant, an image of an inanimate object, etc.

The first task may be different from the second task. The object to be predicted may involve the first task and/or the second task. For example, the first task may include predicting objects of a first type set, and the second task may include predicting objects of a second type set. Accordingly, the object to be predicted may include objects of the first type set and/or objects of the second type set. For example, the first type set may include one or more types, and the second type set may include one or more types that are different from the individual types in the first type set. For example, in the case of digital character recognition, the objects of the first type set may include the numbers "0" and "1", and the objects of the second type set may include the numbers "2" and "3", but are not limited to this.

In this article, "a sample set for a specific task" may refer to a sample set used to train a predetermined model for a specific task. For example, in the case where the specific task is image recognition of a certain type of object, the sample set for the specific task may be a set of multiple images containing objects of that type. In addition, a "simulated sample set" may refer to a sample set obtained by a predetermined algorithm rather than directly by sample collection. In addition, in the text, "prediction" may include recognition, classification, etc.

In the related art, when a pre-trained model is further trained using a new training sample set, the resulting retrained model may lose the performance of the pre-trained model. In addition, since the original sample set used to obtain the pre-trained model contains a large number of samples, the cost of obtaining a large number of original samples is high, and a large number of original samples cannot be used at a low cost to constrain the prediction performance of the retrained model. In some cases, the original samples are even unknowable, so the original samples cannot be used to constrain the prediction performance of the retrained model.

As described above, the information processing device 100 according to the first embodiment of the present disclosure uses a sample generation network obtained based on a second training sample set for a second task to generate a first simulation sample set for a first task. Therefore, in the process of retraining the pre-trained first model, the prediction performance of the pre-trained first model can be constrained by using the first simulation sample set, so as to well retain the prediction performance of the pre-trained first model, thereby obtaining a retrained first model with good prediction performance for both the first task and the second task. In addition, since the cost of obtaining the first simulation sample set is much lower than the cost of obtaining a large number of original samples, the cost can be reduced. In addition, the first simulation sample set can also be used to determine whether the retrained first model has forgotten the first task and/or the degree of forgetting the first task.

For example, when the original sample set is known, a small amount of original samples can also be obtained, and in the process of retraining the pre-trained first model, the prediction performance of the pre-trained first model is constrained by using the obtained small amount of original samples and the first simulated sample set.

As an example, the difference between the prediction result of the pre-trained first model for the first simulated sample set (hereinafter referred to as the "first prediction result") and the prediction result of the retrained first model for the first simulated sample set (hereinafter referred to as the "second prediction result") can be obtained by direct comparison. For example, in the case where the first task is object recognition, the ratio (M1/N) of the number of simulated samples M1 (M1 is a natural number greater than 0) in the first simulated sample set and the total number of samples N (N is a natural number greater than 1) in the first simulated sample set can be calculated as the difference between the first prediction result and the second prediction result: the pre-trained first model and the retrained first model recognize the simulated sample as different types of objects.

As another example, the difference between the prediction result of the pre-trained first model on the first simulation sample set and the prediction result of the retrained first model on the first simulation sample set can be represented by the output vector of the loss layer included in the pre-trained first model. In this way, the difference between the first prediction result and the second prediction result can be more accurately represented, thereby better retaining the prediction performance of the pre-trained first model for the first task.

For example, the ratio (M2/N) of the number of the following samples M2 (M2 is a natural number greater than 0) in the first simulation sample set to the total number of samples N in the first simulation sample set can be calculated as the difference between the first prediction result and the second prediction result: the difference between the output vector for the simulation sample obtained by the loss layer included in the pre-trained first model (hereinafter referred to as the "first output vector") and the output vector for the simulation sample obtained by the loss layer included in the retrained first model (hereinafter referred to as the "second output vector") is greater than or equal to the first predetermined threshold. For example, in this case, the predetermined range can be 0, but is not limited to this, and can be set according to actual needs. For example, the difference between the first output vector and the second output vector can be represented by the Euclidean distance between the first output vector and the second output vector. The first predetermined threshold can be obtained by a limited number of experiments or set based on experience.

For example, in the case where the difference between the first prediction result and the second prediction result is represented by the output vector of the loss layer included in the pre-trained first model, the pre-trained first model can be retrained by adding the loss function LD about the above difference to the loss function L corresponding to the pre-trained first model to obtain the retrained first model that satisfies the above first predetermined condition. For example, the total loss function L' to which the loss function LD about the above difference is added can be expressed by the following formula (1):

L’＝L+ω*λ*LD Formula (1)

In the above formula (1), ω is used to indicate the type of samples input to the first model during the retraining process. When the samples input to the first model are training samples in the second training sample set, ω is equal to 0, and when the samples input to the first model are simulated samples in the first simulated sample set, ω is equal to 1. λ>0 represents a weight, and the larger λ is, the closer the prediction performance of the expected retrained first model for the first task is to the prediction performance of the pre-trained first model for the first task.

For example, the loss function LD for the above difference can be expressed by the following formula (2):

In the above formula (2), A1 represents the output vector of the loss layer of the first model during the retraining process, and A2 represents the output vector of the loss layer of the pre-trained first model.

For example, the pre-trained first model may be a neural network model, such as a convolutional neural network (CNN) model, but is not limited thereto, and those skilled in the art may use a suitable model according to actual needs.

<Second Embodiment>

2 is a block diagram showing an example of a functional configuration of an information processing device 200 according to a second embodiment of the present disclosure. As shown in FIG2 , the information processing device 200 according to the second embodiment of the present disclosure may include a sample generation unit 202, a first model training unit 204, and a sample generation network acquisition unit 206. For example, the sample generation unit 202 and the first model training unit 204 may have similar functional configurations to the sample generation unit 102 and the first model training unit 104 in the first embodiment described above, and the description will be omitted or briefly described below.

The sample generation network acquisition unit 206 may be configured to train the initial generative adversarial network based on the second training sample set for the second task to obtain a trained generative adversarial network. For example, when a random vector is used as input, the corresponding discriminator cannot distinguish between the simulated samples output by the trained generative adversarial network and the training samples in the second training sample set.

After obtaining the trained generative adversarial network, the sample generation network acquisition unit 206 may adjust the trained generative adversarial network to meet a second predetermined condition, thereby obtaining an adjusted generative adversarial network as the sample generation network.

The second predetermined condition may include: in the case of using a random vector as the input of the adjusted generative adversarial network, the prediction result of the pre-trained first model on the second simulated sample set as the output of the adjusted generative adversarial network involves the prediction result for the first task. For example, in the case where the first task includes predicting objects of a first type set, the prediction result of the pre-trained first model on the second simulated sample set involving the prediction result for the first task can be expressed as: the pre-trained first model predicts each object in each simulated sample in the second simulated sample set as an object of the first type set. For example, in the case where the first task includes recognizing the numbers "0" and "1", the prediction result of the pre-trained first model on the second simulated sample set involving the prediction result for the first task can be expressed as: the pre-trained first model predicts each object in each simulated sample in the second simulated sample set as either "0" or "1". Preferably, the number of objects predicted to be "0" is close to the number of objects predicted to be "1".

As an example, the expected prediction result of the pre-trained first model for each simulated sample in the second simulated sample set can be set. In this case, the second predetermined condition can also include: for each simulated sample in the second simulated sample set, the prediction result of the pre-trained first model for the simulated sample is the same as the expected prediction result. For example, in the case where the first task includes identifying the numbers "0" and "1", the expected prediction results of multiple simulated samples in the second simulated sample set can be set to the number "0" in advance, and the expected prediction results of other simulated samples can be set to the number "1".

As another example, the output vector of the desired loss layer may be set for each simulated sample in the second simulated sample set. In this case, the second predetermined condition may also include: for each simulated sample in the second simulated sample set, when the simulated sample is used as the input of the pre-trained first model, the difference between the output vector of the loss layer and the output vector of the desired loss layer is less than or equal to a second predetermined threshold. The second predetermined threshold may be obtained through a limited number of experiments or set based on experience.

As shown in Figure 3, a random vector can be input into a trained generative adversarial network to obtain a simulated sample, the obtained simulated sample can be input into a pre-trained first model to obtain a prediction result of the pre-trained first model for the simulated sample, and the trained generative adversarial network can be adjusted based on the prediction result.

The sample generation unit 202 can generate a first simulation sample set for the first task using the sample generation network obtained by the sample generation network acquisition unit 206. The first model training unit 204 can use the first simulation sample set generated by the sample generation unit 202 to constrain the prediction performance of the pre-trained first model during the retraining of the pre-trained first model, so as to better retain the prediction performance of the pre-trained first model. In addition, since there is no need to obtain a large number of original samples for training the pre-trained first model, the cost can be reduced.

FIG4 shows an example of a second training sample and an example of a simulated sample obtained by the sample generation unit 202 when the first task includes recognizing the numbers "0" and "1" and the second task includes recognizing the numbers "2" and "3". As can be seen from FIG4, the simulated sample includes the numbers "0" and "1". It can be seen that the simulated sample obtained by the sample generation unit 202 can well simulate the training sample for the first task.

As an example, the prediction result of the pre-trained first model for each simulated sample in the second simulated sample is obtained in the following manner, and the set of prediction results of each simulated sample in the obtained second simulated sample is used as the prediction result of the pre-trained first model for the second simulated sample set: noise (for example, Gaussian noise) is added to the simulated sample, and the prediction result of the pre-trained first model for the simulated sample with added noise is obtained as the prediction result of the pre-trained first model for the simulated sample. Using the sample generation network obtained in this way, a simulated sample set that better simulates the training sample for the first task can be obtained, thereby better retaining the prediction performance for the first task.

As another example, the prediction result of the pre-trained first model for each simulated sample in the second simulated sample is obtained in the following manner, and the set of prediction results of each simulated sample in the obtained second simulated sample is used as the prediction result of the pre-trained first model for the second simulated sample set: noise (e.g., Gaussian noise) is added to the features of the simulated sample extracted by the feature extraction layer included in the pre-trained first model, and the prediction result based on the feature with added noise is obtained as the prediction result of the pre-trained first model for the second simulated sample. The sample generation network obtained in this way can be used to obtain a simulated sample set that better simulates the training samples for the first task, thereby better retaining the prediction performance for the first task.

<Third Embodiment>

5 is a block diagram showing an example of a functional configuration of an information processing device 500 according to the third embodiment of the present disclosure. As shown in FIG5 , the information processing device 500 according to the third embodiment of the present disclosure may include a sample generation unit 502, a first model training unit 504, and a sample generation network acquisition unit 506. For example, the sample generation unit 502 and the first model training unit 504 may have similar functional configurations to the sample generation unit 102 and the first model training unit 104 in the first embodiment described above, and the description will be omitted or briefly described below.

The sample generation network acquisition unit 506 may be configured to: train the initial generation network including one or more coding layers based on the second training sample set for the second task to achieve a third predetermined condition, thereby obtaining a trained generation network including one or more coding layers as the sample generation network. The third predetermined condition may include: when the second training sample set is used as the input of the trained generation network, the prediction result of the pre-trained first model on the second simulated sample set as the output of the trained generation network involves the prediction result for the first task.

The sample generation unit 502 can use the sample generation network obtained by the sample generation network acquisition unit 506 to generate a first simulation sample set for the first task based on the second training sample set. The first model training unit 504 can use the first simulation sample set generated by the sample generation unit 502 to constrain the prediction performance of the pre-trained first model during the retraining of the pre-trained first model, so as to better retain the prediction performance of the pre-trained first model. In addition, since it is not necessary to obtain a large number of original samples for training the pre-trained first model, the cost can be reduced.

As an example, the expected prediction result of the pre-trained first model for each simulated sample in the second simulated sample set can be set. In this case, the third predetermined condition can also include: for each simulated sample in the second simulated sample set, the prediction result of the pre-trained first model for the simulated sample is the same as the expected prediction result.

As another example, the output vector of the desired loss layer may be set for each simulated sample in the second simulated sample set. In this case, the third predetermined condition may also include: for each simulated sample in the second simulated sample set, when the simulated sample is used as the input of the pre-trained first model, the difference between the output vector of the loss layer and the output vector of the desired loss layer is less than or equal to a third predetermined threshold. The third predetermined threshold may be obtained through a limited number of experiments or set based on experience.

As shown in Figure 6, during the training process of the generative network, the second training sample can be input into the generative network to obtain a simulated sample, and then the obtained simulated sample can be input into the pre-trained first model to obtain the actual prediction result of the pre-trained first model for the simulated sample (for example, the output vector of the loss layer), and gradient back propagation is performed based on the difference between the actual prediction result and the expected prediction result (for example, the expected output vector of the loss layer) to train the generative network.

As shown in Figure 6, in some examples, the initial generation network may further include one or more decoding layers. The output of the last encoding layer may be used as the input of the first decoding layer, and the output of the last decoding layer may be used as the output of the trained generation network.

For example, the encoding layer can be used to reduce the dimensionality of the input, while the decoding layer can be used to increase the dimensionality of the output.

For example, each encoding layer may be a fully connected layer including K1*K1 neurons, and each decoding layer may be a fully connected layer including K2*K2 neurons, where K1 and K2 are each a natural number greater than 1. In some examples, K1 may be equal to K2.

7 shows an example of a second training sample and an example of a simulated sample obtained by the sample generation unit 502 based on the second training sample when the first task includes recognizing the numbers "0" and "1" and the second task includes recognizing the numbers "2" and "3". In the example shown in FIG7 , both K1 and K2 are set to 28, and the sum of the number of encoding layers and the number of decoding layers is 3.

As can be seen from FIG7 , when the initial generation network is trained for 10,000 or more iterations, the obtained simulated samples can well simulate the training samples for the first task.

Similar to the sample generation network acquisition unit 206 in the second embodiment, the sample generation network acquisition unit 506 can add noise to each simulated sample or the features of each simulated sample in the second simulated sample set when acquiring the prediction result of the pre-trained first model for the second simulated sample set during the process of training the initial generation network. In this way, the sample generation network obtained by the sample generation network acquisition unit 506 can obtain a simulated sample set that better simulates the training samples for the first task, thereby better retaining the prediction performance for the first task.

8 is a diagram showing a comparison between an information processing device 500 according to a third embodiment of the present disclosure and related art. As can be seen from FIG8, for GEM (e.g., see Lopez-Paz D, Ranzato M A. Gradient episodic memory for continual learning [J]. Advances in neural information processing systems, 2017, 30), A-GEM (e.g., see Arslan Chaudhry, Marc’Aurelio Ranzato, Marcus Rohrbach, and Mohamed Elhoseiny. Efficient lifelong learning with a-gem. In International Conference on Learning Representations, 2019) and EWC (e.g., see Kirkpatrick J, Pascanu R, Rabinowitz N, et al. Overcoming catastrophic forgetting in neural networks [J]. Proceedings of the national academy of sciences, 2017, 114(13): 3521-3526). By introducing the simulated sample set, the average accuracy in the case of increasing categories is improved by 23.11%, 25.60% and 25.84% respectively, and the average accuracy in the case of increasing tasks is improved by 1.08%, 2.85% and 18.62% respectively.

Note that although an application example of the information processing device according to an embodiment of the present disclosure is described, the application of the information processing device is not limited to this, and those skilled in the art can apply the information processing device to various aspects according to actual needs, which will not be repeated here.

<Fourth Embodiment>

Corresponding to the above-mentioned information processing device embodiments, the present disclosure also provides the following information processing method embodiments.

Fig. 9 is a flow chart showing a process example of an information processing method 900 according to the fourth embodiment of the present disclosure. As shown in Fig. 9, the information processing method 900 according to the fourth embodiment of the present disclosure may start at a start step S902, end at an end step S908, and include a sample generation step S904 and a first model training step S906.

In the sample generation step S904, a sample generation network obtained based on the second training sample set for the second task may be used to generate a first simulated sample set for the first task. For example, the sample generation step S904 may be performed by the sample generation units 102, 202, and 502 described above, so the specific details may refer to the above description, and only a brief description is given below.

In the first model training step S906, the first model pre-trained for the first task can be retrained using the second training sample set under the first predetermined condition to obtain a retrained first model for predicting the object to be predicted. For example, the first predetermined condition may include: the difference between the prediction result of the pre-trained first model for the first simulation sample set and the prediction result of the retrained first model for the first simulation sample set is within a predetermined range. In this way, in the process of retraining the pre-trained first model, the prediction performance of the pre-trained first model can be constrained using the first simulation sample set to well retain the prediction performance of the pre-trained first model, thereby obtaining a retrained first model with good prediction performance for both the first task and the second task.

For example, the first model training step S906 can be performed by the first model training units 104, 204 and 504 described above, so the specific details can be found in the above description, and only a brief description is given below.

For example, the pre-trained first model may be a neural network model, such as a convolutional neural network (CNN) model, but is not limited thereto, and those skilled in the art may use a suitable model according to actual needs.

For example, the information processing method may further include a sample generation network acquisition step (not shown). As an example, in the sample generation network acquisition step, the initial generative adversarial network may be trained based on the second training sample set to obtain a trained generative adversarial network, and the trained generative adversarial network may be adjusted to reach a second predetermined condition, thereby obtaining an adjusted generative adversarial network as a sample generation network. The second predetermined condition may include: in the case of using a random vector as the input of the adjusted generative adversarial network, the prediction result of the pre-trained first model for the second simulated sample set as the output of the adjusted generative adversarial network involves the prediction result for the first task.

As another example, in the sample generation network acquisition step, the initial generation network including one or more coding layers can be trained based on the second training sample set to achieve a third predetermined condition, thereby obtaining a trained generation network including one or more coding layers as the sample generation network. The third predetermined condition may include: when the second training sample set is used as the input of the trained generation network, the prediction result of the pre-trained first model for the second simulated sample set as the output of the trained generation network involves the prediction result for the first task.

For example, the sample generation network acquisition step may be performed by the sample generation network acquisition units 206 and 506 described above, so the specific details may refer to the above description, and only a brief description is given below.

As an example, the prediction results of the pre-trained first model for the second simulated sample set are obtained in the following manner: noise is added to each simulated sample in the second simulated sample set, and the prediction results of the pre-trained first model for each simulated sample with added noise are obtained as the prediction results of the pre-trained first model for the second simulated sample set. The sample generation network obtained in this manner can be used to obtain a simulated sample set that better simulates the training samples for the first task, thereby better retaining the prediction performance of the pre-trained first model for the first task.

As another example, the prediction result of the pre-trained first model for the second simulated sample set is obtained in the following manner: adding noise to the features of each simulated sample in the second simulated sample set extracted by the feature extraction layer included in the pre-trained first model, and obtaining the prediction result based on the feature with added noise as the prediction result of the pre-trained first model for the second simulated sample set. The sample generation network obtained in this manner can be used to obtain a simulated sample set that better simulates the training samples for the first task, thereby better retaining the prediction performance of the pre-trained first model for the first task.

It should be pointed out that although the functional configuration and operation of the information processing device and information processing method according to the embodiments of the present disclosure are described above, this is only an example and not a limitation, and those skilled in the art may modify the above embodiments according to the principles of the present disclosure, for example, the functional modules in each embodiment may be added, deleted or combined, and such modifications shall fall within the scope of the present disclosure.

In addition, it should be pointed out that the method embodiment here corresponds to the above-mentioned device embodiment. Therefore, for the contents not described in detail in the method embodiment, please refer to the description of the corresponding parts in the device embodiment, and the description will not be repeated here.

It should be understood that the machine-executable instructions in the storage medium and program products according to the embodiments of the present disclosure can also be configured to execute the above-mentioned information processing method. Therefore, the contents not described in detail here can refer to the description of the previous corresponding parts and will not be repeated here.

Accordingly, the storage medium for carrying the program product including the machine executable instructions is also included in the disclosure of the present invention, including but not limited to a floppy disk, an optical disk, a magneto-optical disk, a memory card, a memory stick, and the like.

In addition, it should be noted that the above series of processes and systems can also be implemented by software and/or firmware. In the case of implementation by software and/or firmware, the program constituting the software is installed from a storage medium or a network to a computer with a dedicated hardware structure, such as a general-purpose personal computer 1000 shown in FIG. 10 , and the computer can perform various functions, etc. when various programs are installed.

10 , a central processing unit (CPU) 1001 executes various processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1008 to a random access memory (RAM) 1003. In the RAM 1003, data required when the CPU 1001 executes various processes and the like is also stored as needed.

The CPU 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. To the bus 1004, an input/output interface 1005 is also connected.

The following components are connected to the input/output interface 1005: an input device 1006 including a keyboard, a mouse, etc.; an output device 1007 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage device 1008 including a hard disk, etc.; and a communication device 1009 including a network interface card such as a LAN card, a modem, etc. The communication device 1009 performs communication processing via a network such as the Internet.

A drive 1010 is also connected to the input/output interface 1005 as needed. A removable medium 1011 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is mounted on the drive 1010 as needed so that a computer program read therefrom is installed into the storage device 1008 as needed.

In the case where the above-described series of processing is realized by software, a program constituting the software is installed from a network such as the Internet or a storage medium such as the removable medium 1011 .

It should be understood by those skilled in the art that such storage medium is not limited to the removable medium 1011 shown in FIG. 10 in which the program is stored and distributed separately from the device to provide the program to the user. Examples of the removable medium 1011 include magnetic disks (including floppy disks (registered trademark)), optical disks (including compact disk read-only memory (CD-ROM) and digital versatile disks (DVD)), magneto-optical disks (including minidiscs (MD) (registered trademark)), and semiconductor memories. Alternatively, the storage medium may be a ROM 1002, a hard disk included in the storage device 1008, or the like, in which the program is stored and distributed to the user together with the device containing them.

The preferred embodiments of the present disclosure are described above with reference to the accompanying drawings, but the present disclosure is certainly not limited to the above examples. Those skilled in the art may obtain various changes and modifications within the scope of the appended claims, and it should be understood that these changes and modifications will naturally fall within the technical scope of the present disclosure.

For example, a plurality of functions included in one unit in the above embodiments may be implemented by separate devices. Alternatively, a plurality of functions implemented by a plurality of units in the above embodiments may be implemented by separate devices, respectively. In addition, one of the above functions may be implemented by a plurality of units. Needless to say, such a configuration is included in the technical scope of the present disclosure.

In this specification, the steps described in the flowchart include not only the processing performed in time series in the order described, but also the processing performed in parallel or individually rather than necessarily in time series. In addition, even in the steps processed in time series, it goes without saying that the order can be appropriately changed.

Additionally, the technology according to the present disclosure may also be configured as follows.

Note 1. An information processing device, comprising:

a sample generating unit configured to generate a first simulated sample set for the first task using a sample generating network obtained based on a second training sample set for the second task; and

The first model training unit is configured to retrain the first model pre-trained for the first task using the second training sample set under a first predetermined condition to obtain a retrained first model for predicting the object to be predicted,

The first predetermined condition includes: the difference between the prediction result of the pre-trained first model on the first simulation sample set and the prediction result of the retrained first model on the first simulation sample set is within a predetermined range,

wherein the first task is different from the second task,

The object to be predicted involves the first task and/or the second task.

Note 2. The information processing device according to Note 1 further comprises a sample generation network acquisition unit configured to:

Training an initial generative adversarial network based on the second training sample set to obtain a trained generative adversarial network; and

The trained generative adversarial network is adjusted to meet a second predetermined condition, thereby obtaining an adjusted generative adversarial network as the sample generation network,

The second predetermined condition includes: when using a random vector as the input of the adjusted generative adversarial network, the prediction results of the pre-trained first model for the second simulated sample set as the output of the adjusted generative adversarial network involve the prediction results for the first task.

Note 3. The information processing device according to Note 1 further comprises a sample generation network acquisition unit configured to: train an initial generation network including one or more coding layers based on the second training sample set to achieve a third predetermined condition, thereby obtaining a trained generation network including one or more coding layers as the sample generation network,

Among them, the third predetermined condition includes: when using the second training sample set as the input of the trained generation network, the prediction result of the pre-trained first model for the second simulated sample set as the output of the trained generation network involves the prediction result for the first task.

Note 4. An information processing device according to Note 3, wherein the trained generation network further includes one or more decoding layers, the output of the last encoding layer among the one or more encoding layers is used as the input of the first decoding layer of the one or more decoding layers, and the output of the last decoding layer of the one or more decoding layers is used as the output of the trained generation network.

Note 5. An information processing device according to any one of Notes 2 to 4, wherein the prediction result of the pre-trained first model for the second simulated sample set is obtained by adding noise to each simulated sample in the second simulated sample set, and obtaining the prediction result of the pre-trained first model for each simulated sample with added noise as the prediction result of the pre-trained first model for the second simulated sample set.

Note 6. An information processing device according to any one of Notes 2 to 4, wherein the prediction result of the pre-trained first model for the second simulated sample set is obtained by adding noise to the features of each simulated sample in the second simulated sample set extracted by the feature extraction layer included in the pre-trained first model, and obtaining the prediction result based on the feature with added noise as the prediction result of the pre-trained first model for the second simulated sample set.

Note 7. An information processing device according to any one of Notes 1 to 4, wherein the difference between the prediction results of the pre-trained first model for the first simulated sample set and the prediction results of the retrained first model for the first simulated sample set is represented by the output vector of the loss layer included in the pre-trained first model.

Note 8. An information processing device according to any one of Notes 1 to 4, wherein the pre-trained first model is a neural network model.

Note 9. An information processing method comprising:

generating a first simulated sample set for the first task using a sample generation network obtained based on a second training sample set for the second task; and

Under a first predetermined condition, retraining a first model pre-trained for the first task using the second training sample set to obtain a retrained first model for predicting the object to be predicted,

The first predetermined condition includes: the difference between the prediction result of the pre-trained first model on the first simulation sample set and the prediction result of the retrained first model on the first simulation sample set is within a predetermined range,

wherein the first task is different from the second task,

The object to be predicted involves the first task and/or the second task.

Note 10. The information processing method according to Note 9 further comprises:

Training an initial generative adversarial network based on the second training sample set to obtain a trained generative adversarial network; and

The trained generative adversarial network is adjusted to meet a second predetermined condition, thereby obtaining an adjusted generative adversarial network as the sample generation network,

The second predetermined condition includes: when using a random vector as the input of the adjusted generative adversarial network, the prediction results of the pre-trained first model for the second simulated sample set as the output of the adjusted generative adversarial network involve the prediction results for the first task.

Note 11. The information processing method according to Note 9 further comprises: training an initial generation network including one or more coding layers based on the second training sample set to achieve a third predetermined condition, thereby obtaining a trained generation network including one or more coding layers as the sample generation network,

Among them, the third predetermined condition includes: when using the second training sample set as the input of the trained generation network, the prediction result of the pre-trained first model for the second simulated sample set as the output of the trained generation network involves the prediction result for the first task.

Note 12. An information processing method according to Note 11, wherein the trained generation network further includes one or more decoding layers, the output of the last encoding layer among the one or more encoding layers is used as the input of the first decoding layer of the one or more decoding layers, and the output of the last decoding layer of the one or more decoding layers is used as the output of the trained generation network.

Note 13. An information processing method according to any one of Notes 10 to 12, wherein the prediction result of the pre-trained first model for the second simulated sample set is obtained by adding noise to each simulated sample in the second simulated sample set, and obtaining the prediction result of the pre-trained first model for each simulated sample with added noise as the prediction result of the pre-trained first model for the second simulated sample set.

Note 14. An information processing method according to any one of Notes 10 to 12, wherein the prediction result of the pre-trained first model for the second simulated sample set is obtained by adding noise to the features of each simulated sample in the second simulated sample set extracted by the feature extraction layer included in the pre-trained first model, and obtaining the prediction result based on the feature with added noise as the prediction result of the pre-trained first model for the second simulated sample set.

Note 15. An information processing method according to any one of Notes 9 to 12, wherein the difference between the prediction results of the pre-trained first model for the first simulated sample set and the prediction results of the retrained first model for the first simulated sample set is represented by the output vector of the loss layer included in the pre-trained first model.

Note 16. An information processing method according to any one of Notes 9 to 12, wherein the pre-trained first model is a neural network model.

Note 17. A computer-readable storage medium storing a program, wherein when the program is executed by a computer, the computer executes the information processing method according to any one of Notes 9 to 16.

Claims (10)

1. An information processing device, comprising: a sample generating unit configured to generate a first simulated sample set for the first task using a sample generating network obtained based on a second training sample set for the second task; and The first model training unit is configured to retrain the first model pre-trained for the first task using the second training sample set under a first predetermined condition to obtain a retrained first model for predicting the object to be predicted, The first predetermined condition includes: the difference between the prediction result of the pre-trained first model on the first simulation sample set and the prediction result of the retrained first model on the first simulation sample set is within a predetermined range, wherein the first task is different from the second task, The object to be predicted involves the first task and/or the second task. 2. The information processing device according to claim 1, further comprising a sample generation network acquisition unit configured to: Training an initial generative adversarial network based on the second training sample set to obtain a trained generative adversarial network; and The trained generative adversarial network is adjusted to meet a second predetermined condition, thereby obtaining an adjusted generative adversarial network as the sample generation network, The second predetermined condition includes: when using a random vector as the input of the adjusted generative adversarial network, the prediction results of the pre-trained first model for the second simulated sample set as the output of the adjusted generative adversarial network involve the prediction results for the first task. 3. The information processing device according to claim 1, further comprising a sample generation network acquisition unit, configured to: train an initial generation network including one or more coding layers based on the second training sample set to achieve a third predetermined condition, thereby obtaining a trained generation network including one or more coding layers as the sample generation network, Among them, the third predetermined condition includes: when using the second training sample set as the input of the trained generation network, the prediction result of the pre-trained first model for the second simulated sample set as the output of the trained generation network involves the prediction result for the first task. 4. The information processing device according to claim 3, wherein the trained generation network further includes one or more decoding layers, the output of the last encoding layer of the one or more encoding layers is used as the input of the first decoding layer of the one or more decoding layers, and the output of the last decoding layer of the one or more decoding layers is used as the output of the trained generation network. 5. An information processing device according to any one of claims 2 to 4, wherein the prediction result of the pre-trained first model for the second simulated sample set is obtained by adding noise to each simulated sample in the second simulated sample set, and obtaining the prediction result of the pre-trained first model for each simulated sample with added noise as the prediction result of the pre-trained first model for the second simulated sample set. 6. An information processing device according to any one of claims 2 to 4, wherein the prediction result of the pre-trained first model for the second simulation sample set is obtained by adding noise to the features of each simulation sample in the second simulation sample set extracted by using the feature extraction layer included in the pre-trained first model, and obtaining the prediction result based on the feature with added noise as the prediction result of the pre-trained first model for the second simulation sample set. 7. An information processing device according to any one of claims 1 to 4, wherein the difference between the prediction results of the pre-trained first model for the first simulated sample set and the prediction results of the retrained first model for the first simulated sample set is represented by the output vector of the loss layer included in the pre-trained first model. 8. The information processing device according to any one of claims 1 to 4, wherein the pre-trained first model is a neural network model. 9. An information processing method, comprising: generating a first simulated sample set for the first task using a sample generation network obtained based on a second training sample set for the second task; and Under a first predetermined condition, retraining a first model pre-trained for the first task using the second training sample set to obtain a retrained first model for predicting the object to be predicted, The first predetermined condition includes: the difference between the prediction result of the pre-trained first model on the first simulation sample set and the prediction result of the retrained first model on the first simulation sample set is within a predetermined range, wherein the first task is different from the second task, The object to be predicted involves the first task and/or the second task. 10 . A computer-readable storage medium storing a program which, when executed by a computer, causes the computer to execute the information processing method according to claim 9 .