Information processing device and information processing method

Abstract

To allow preferable VFL to be executed.SOLUTION: An information processing device includes a control unit that specifies one or more features and one or more sample IDs for use in VFL from collected data, filters the specified one or more features and one or more sample IDs on the basis of at least one of a privacy requirement, a minimum number of requirements for at least either the features or the sample IDs, and a maximum number of requirements for at least either the features or the sample IDs, and selects one or more features and one or more sample IDs on the basis of filtering result.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to an information processing device and an information processing method. [Background technology]

[0002] Federated Learning (FL) is a machine learning concept in which multiple entities jointly train a machine learning model in a distributed manner without sharing samples. There are two types of federated learning: horizontal federated learning (HFL) and vertical federated learning (VFL).

[0003] Regarding HFL and VFL, 3GPP specifies the HFL procedure between Network Data Analysis Functions (NWDAFs) (e.g., Non-Patent Document 1). 3GPP also specifies HFL, VFL, VFL Active Participant, and VFL Passive Participant in 5G systems (e.g., For example, Non-Patent Document 2). [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] 3GPP TS 23.288 V18.4.0, “Architecture enhancements for 5G System (5GS) to support network data analytics services.” https: / / www.3gpp.org / ftp / Specs / archive / 23_series / 23.288 / 23288-i40.zip [Non-patent document 2] 3GPP TR 23.700-84 V0.2.0, “Study on Core Network Enhanced Support for Artificial Intelligence (AI) / Machine Learning (ML).” https: / / www.3gpp.org / ftp / Specs / archive / 23_series / 23.700-84 / 23700-84-020.zip Summary of the Invention [Problem to be solved by the invention]

[0005] An object of the present disclosure is to provide an information processing device and an information processing method that enable suitable vertical association learning. [Means for solving the problem]

[0006] One aspect of the present disclosure is an information processing device that includes a control unit that identifies one or more features and one or more sample IDs to be used for vertical federated learning (VFL) from collected data, filters the identified one or more features and one or more sample IDs based on at least one of a privacy requirement, a minimum number requirement for at least one of the features and sample IDs, and a maximum number requirement for at least one of the features and sample IDs, and selects one or more features and one or more sample IDs from the filtering results.

[0007] One aspect of the present disclosure is to provide a method for acquiring one or more features and one or more sample IDs from each of a plurality of vertically federated learning (VFL) participants, filtering the acquired one or more features and one or more sample IDs based on at least one of a privacy requirement, a minimum number requirement for at least one of the features and the sample IDs, and a maximum number requirement for at least one of the features and the sample IDs, and transmitting one or more features and one or more sample IDs to be used in the VFL selected from the filtering results to the plurality of VFL participants; The information processing device includes a control unit that executes the above.

[0008] In addition, as another aspect, an information processing method corresponding to the above-mentioned information processing device, Examples of the program include a program to be executed by a computer, and a computer-readable storage medium that non-temporarily stores the program. [Effects of the Invention]

[0009] According to the present disclosure, it is possible to perform suitable vertical association learning. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1A is an explanatory diagram of horizontal associative learning, and FIG. 1B is an explanatory diagram of vertical associative learning. [Figure 2] FIG. 2 is a sequence diagram illustrating a communication system and an example of an operation in the communication system. [Figure 3] FIG. 3 is a flowchart illustrating an example of processing in the communication system. [Figure 4] FIG. 4 shows an example of the configuration of an information processing device that can operate as an NF that constitutes a communication system. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. The configurations of the following embodiments are examples, and the present disclosure is not limited to the configurations of the embodiments. Figure 1A is an explanatory diagram of horizontal associative learning (HFL), and Figure 1B is an explanatory diagram of vertical associative learning (VFL).

[0012] HFL is a method for identifying entities that share the same feature, as shown in Figure 1A. VFL is a federated learning technique for when multiple entities have different features but share the same samples, as shown in Figure 1B. Among the multiple participants in VFL, participants (also called parties) who have the correct label in machine learning are called active participants, and participants other than active participants (participants who do not have the correct label) are called passive participants.

[0013] In 3GPP, HFL, VFL, etc. are defined as follows: · HVL: A federated learning method without exchanging / sharing local datasets, where the local datasets of different FL clients for local model training have the same feature space for different samples (e.g., UE IDs). · VFL: A federated learning technique that does not exchange / share local datasets, and the local datasets of different VFL participants for local model training have different feature spaces for the same sample (e.g., UE ID). Labels: Labels are features of interest that are learned in supervised machine learning. However, the definition of a label will be revised and expanded as needed. VFL active participant: A network function (NF) with labels for VFL training tasks that may have associated input data. VFL passive participant: An NF that does not have the labels required for the VFL training task but has the input data required for the VFL training task. There can be multiple passive participants in a VFL.

[0014] One of the technical challenges of VFL in 5G systems is to align samples and features between NFs before training a machine learning (ML) model. To do this, it is necessary to determine a common sample ID between NFs without sharing samples between NFs. For example, the following points must be considered when aligning samples and features: Different samples or features may have different privacy requirements to meet. To set an upper bound on the computational complexity of VFL training and inference, it may be necessary to set an upper bound on the number of features and / or samples. To ensure minimum performance, it may be necessary to set a minimum number of features and / or samples that must be met.

[0015] The VFL control method according to the embodiment enables the following: First, each VFL active or passive NF can remove one or more features and / or one or more sample IDs depending on privacy requirements. Also, each VFL active or passive NF can remove one or more features and / or one or more sample IDs to meet the minimum and / or maximum number of features / sample IDs. At this time, each VFL active / passive NF sends the selected features and sample IDs to the VFL coordinator NF.

[0016] Second, the VFL coordinator NF selects an appropriate feature set for each VFL active / passive NF to meet the minimum and / or maximum number of features while ensuring that there are no overlapping features between the VFL active / passive NFs and removing one or more (unnecessary) features.

[0017] Figure 2 shows an example of the filtering and sample and feature alignment procedures proposed for VFL. The procedures shown in Figure 2 are performed by multiple network functions (NFs) that make up a 5G system (5G Core: 5GC), which is an example of a communication system. NFs are network nodes (entities) that make up the 5GC. However, the filtering and alignment procedures described below can also be applied to systems other than 5G systems.

[0018] The communication system (5G system) shown in Figure 2 includes a VFL coordinator NF11, N (N is an integer of 1 or greater) VFL active / passive NFs 12, an NRF (Network Repository Function) 13, and an NF 14. NF 12 is an NF that can become a VFL active participant or a VFL passive participant depending on the VFL coordinator NF 11. VFL active / passive NF 12 is an example of a "VFL participant," and VFL coordinator NF 11 is an example of a "VFL coordinator."

[0019] The VFL coordinator NF11 and the VFL active / passive NF12 are, for example, NWDAFs (Network Data Analytics Functions). The VFL Coordinator NF 11 and the VFL Active / Passive NF 12 may be existing or new NFs other than the NWDAF.

[0020] The NRF 11 stores and manages information about NFs within the 5GC. In response to an inquiry about an NF desired to be used, the NRF 11 can return multiple candidate NFs to the inquiry source. The NF 14 is an NF that serves as a data source, such as, but not limited to, an AMF, an SMF, an UPF, a PCF, or an UDM. The NF 14 may be an existing NF or a new NF. The data collection source may be an OAM terminal or an external server.

[0021] An example of the VFL feature and sample filtering and alignment procedure is described below using FIG. 2.

[0022] <Step 1> In step 1 of Figure 2, each VFL active / passive NF12 collects data from NF14 (data source) via NRF13.

[0023] <Step 2> In step 2 of Figure 2, each VFL active / passive NF12 determines a set of features and sample IDs to share with the VFL coordinator NF11. That is, each VFL active / passive NF12 identifies one or more features and one or more sample IDs to use for the VFL.

[0024] (2-1) First form As a first aspect related to step 2, each VFL active / passive NF12 can perform filtering based on privacy requirements. That is, each VFL active / passive NF12 can perform at least one of removing one or more features and removing one or more sample IDs to satisfy the privacy requirements. The privacy requirements can be associated with at least one of the features, the sample IDs, or the NFs. Regarding the first aspect, the configuration of (2-1-1) described below can be applied.

[0025] (2-1-1) Each feature or sample ID is associated with a corresponding privacy indicator. Each VFL active / passive NF12 determines whether to delete one or more features and / or one or more sample IDs based on the privacy indicator associated with each feature and / or sample ID. The privacy indicator may be indicated by the data source NF14, may be (pre)configured, or may be predefined (e.g., defined in a standard).

[0026] (2-2) Second form As a second aspect of step 2, each VFL active / passive NF12 filters the identified one or more features and one or more sample IDs based on at least one of a minimum number requirement for at least one of features and sample IDs and a maximum number requirement for at least one of features and sample IDs. That is, each VFL active / passive NF12 deletes one or more features and / or one or more sample IDs while satisfying the minimum and / or maximum numbers of features and / or sample IDs. The second aspect may apply the configurations according to the following (2-2-1) to (2-2-3).

[0027] (2-2-1) Each VFL active / passive NF12 can remove (filter) one or more features based on an importance metric (e.g., a Shapley value or an approximation thereof) associated with each feature and an importance threshold. For example, a feature is removed if its corresponding importance metric is below the importance threshold. The importance metric for each feature can be determined by pre-training using a partial training dataset (e.g., a validation dataset), (pre-)configuration, or pre-definition (e.g., defined in a standard). The initial importance threshold value may be (pre-)configured or pre-defined. If the number of features after removal falls below the minimum number of features, the importance threshold is lowered and the feature set is initialized. The removal process is then repeated until the minimum number of features is met.

[0028] 3 is a flowchart showing an example of the procedure shown in 2-2-1. In step S01, each VFL active / passive NF12 and / or VFL coordinator NF11 determines the metric of each feature. In step S02, each VFL active / passive NF12 and / or VFL coordinator NF11 initializes a set of features.

[0029] In step S03, it is determined whether the importance metric is below the importance threshold. If it is determined that the importance metric is below the importance threshold, each VFL active / passive NF12 and / or VFL coordinator NF11 determines whether the importance metric is below the importance threshold. The features below are deleted (step S04).

[0030] In step S05, it is determined whether the number of features after the deletion is less than the minimum number of features. If it is determined that the number of features after the deletion is less than the minimum number of features, the importance metric threshold is lowered (step S06) and the process returns to step S02.

[0031] (2-2-2) The minimum and / or maximum number of features and / or sample IDs (at least one of the minimum number related to the minimum number requirement and the maximum number related to the maximum number requirement) can be obtained by pre-training, (pre-)configuring, or pre-defining using a partial training dataset (e.g., a validation dataset). For example, the minimum number of features or sample IDs may be determined based on pre-training so that a minimum requirement for ML (machine learning) performance (e.g., accuracy, loss) is achieved. Alternatively, the maximum number of features and / or sample IDs may be determined so that the time required for training and / or inference does not exceed the maximum time allocated for training and / or inference.

[0032] (2-2-3) The minimum and / or maximum number of features and / or sample IDs are all It can be made equal for all VFL active / passive NF12s, or , the minimum and / or maximum number of features and / or sample IDs may be different for each VFL active / passive NF12.

[0033] <Step 3> In step 3 of FIG. 2, each VFL active / passive NF12 sends the selected features and sample IDs to the VFL coordinator NF11.

[0034] <Step 4> In step 4 of FIG. 2, the VFL coordinator NF11 determines a set of features and common sample IDs for each VFL active / passive NF12. A common sample ID is a sample ID shared between the VFL active / passive NF12 and the VFL coordinator NF11 for the VFL. The VFL coordinator NF11 removes one or more (unnecessary) features to ensure that there are no duplicated features between the VFL active / passive NF12. That is, the VFL coordinator NF11 filters the obtained one or more features and one or more sample IDs based on at least one of a privacy requirement, a minimum number requirement for at least one of features and sample IDs, and a maximum number requirement for at least one of features and sample IDs.

[0035] (4-1) First form As a first aspect of step 4, the VFL coordinator NF11 can remove one or more features and / or one or more sample IDs to satisfy a privacy requirement. The privacy requirement may be associated with at least one of the features or sample IDs. Regarding the first aspect of step 4, the following configurations (4-1-1) to (4-1-5) are applicable.

[0036] (4-1-1) Each feature or sample ID is associated with a corresponding privacy indicator. The VFL coordinator NF11 assigns a privacy indicator to each feature and / or sample ID. Based on the associated privacy indicators, one or more characteristics and / or one or more Decide whether to delete the sample ID above. The privacy indicator is set for each VFL active / passive NF12 or data source, (pre-)configured or predefined (e.g., defined in the standard) as indicated by NF14. Good too.

[0037] (4-1-2) The privacy indicator may be a binary value. For example, "0" means that the feature or sample ID is available to the VFL, and "1" means that the feature or sample ID is not available to the VFL. The meanings of "0" and "1" may be reversed. The privacy indicator may also be a function of the VFL priority or VFL use case. For example, each VFL active / passive NF12 removes the feature or sample ID if the associated privacy indicator value indicates that the feature or sample ID is not available to the VFL.

[0038] (4-1-3) Privacy indicators are used to indicate the privacy of VFL and / or HFL. For example, the privacy indicator may indicate at least one of the following: 1. Feature or sample IDs can be used in both VFL and HFL. 2. Feature or sample IDs can only be used in VFL. 3. Feature or sample IDs can only be used in HFL. 4. Feature or sample IDs are not available in VFL. 5. Feature or sample IDs are not available in HFL.

[0039] (4-1-4) A privacy indicator may be associated with a specific combination of one or more features and / or one or more sample IDs. For example, the privacy indicator may indicate at least one of the following 1 to 8: 1. In VFL, the combination of feature x1 and feature x2 is available. 2. The combination of Feature x1 and Feature x3 cannot be used in VFL. 3. The combination of 4 features and 2 features cannot be used in VFL. 4. In VFL, a combination of feature x4 and feature x3 can be used. 5. The combination of feature x1 and sample ID y1 is available in VFL. 6. The combination of feature x1 and sample ID y2 is not available in VFL. 7. The combination of feature x2 and sample ID y1 is not available in VFL. 8. The combination of feature x2 and sample ID y2 is available in VFL.

[0040] (4-1-5) The privacy indicator may be an integer value indicating a privacy level. For example, a higher value may mean that the feature or sample ID has stricter privacy requirements for the VFL. Alternatively, a lower value may mean that the feature or sample ID has less strict privacy requirements for the VFL. For example, if the privacy indicator value exceeds the privacy threshold, each VFL active / passive NF12 deletes the feature or sample ID. The privacy threshold may be a function of the VFL priority or the VFL use case. The privacy threshold may be (pre)configured or predefined (e.g., defined in a standard).

[0041] (4-2) Second Form As a second mode of step 4, the VFL coordinator NF11 can delete one or more features and / or one or more sample IDs while satisfying the minimum and / or maximum numbers of features and / or sample IDs. The second mode of step 4 can apply the following configurations (4-2-1) to (4-2-4).

[0042] (4-2-1) As an example, if the same feature is displayed by multiple VFL active / passive NFs 12, the VFL coordinator NF11 may remove the feature from one or more VFL active / passive NFs 12 randomly with equal probability or based on a priority associated with each VFL active / passive NF 12. The priority of each VFL active / passive NF 12 may be obtained, for example, by (pre)configuration or pre-definition. can be done.

[0043] (4-2-2) As an example, the VFL coordinator NF11 may remove one or more features based on an importance threshold and an importance metric (e.g., a Shapley value or an approximation thereof) associated with each feature. Here, the importance metric for each feature may be pre-trained using a partial training dataset (e.g., a validation dataset), (pre-)configured, or pre-defined (i.e., defined in a standard). The initial importance threshold may be (pre-)configured or pre-defined. If the number of features after removal falls below the minimum number of features, the importance threshold is lowered and the feature set is initialized. The removal process is then repeated until the minimum number of features is met. In this regard, a process similar to that shown in FIG. 3 may be performed.

[0044] (4-2-3) In one example, the minimum and / or maximum number of features and / or sample IDs may be obtained by pre-training using a partial training dataset (e.g., a validation dataset), a (pre-)configured dataset, or a pre-defined dataset (e.g., a dataset defined by a standard). For example, the minimum number of features and / or sample IDs may be determined based on pre-training so that minimum requirements for machine learning (ML) performance (accuracy, loss, etc.) are achieved. As another example, the maximum number of features and / or sample IDs may be determined so that the time required for training and / or inference does not exceed the maximum time allocated for training and / or inference.

[0045] (4-2-4) In one example, the minimum and / or maximum number of features and / or sample IDs may be the same for all VFL active / passive NFs 12. In another example, the minimum and / or maximum number of features and / or sample IDs may be different for each VFL active / passive NF.

[0046] <Step 5> 5, the VFL coordinator NF11 returns the selected function list and sample ID list (including the sample ID order) to each VFL active / passive NF12. The VFL coordinator NF11 also sends instructions (or orders) regarding the VFL to each VFL active / passive NF12 as necessary.

[0047] In one embodiment, either the VFL active / passive NF12 or the VFL coordinator NF11 can perform the filtering and selection of features and sample IDs. Alternatively, as shown in the example of FIG. 2, both the VFL active / passive NF12 or the VFL coordinator NF11 can perform the filtering and selection of features and / or sample IDs.

[0048] According to the communication system of the embodiment, if there is a privacy requirement regarding any of the features, samples, and NFs, the privacy requirement can be satisfied in the selection of the features and sample IDs. Furthermore, unnecessary features are filtered (removed), thereby reducing the computational load of VFL training and inference. Furthermore, by setting an upper limit on the number of samples or features, an upper limit on the computational load can be set. These features enable an optimal VFL.

[0049] FIG. 4 shows a configuration example of an information processing device 20 that can operate as the NF (NWDAF, etc.) shown in FIG. 2. As an example, a 5G system can be configured by one information processing device 20, or two or more information processing devices 20 (clouds) that are communicatively connected. In FIG. 4, the information processing device 20 includes a processor (controller) interconnected by a bus 127. The computer includes a memory 121, a storage device 122, a communication interface (communication IF) 124, an input device 125, and a display 126.

[0050] The storage device 122 is used as a storage area for data and programs, a program loading area, a work area for the processor 121, a buffer area for communication data, etc. The storage device 122 includes a main storage device and an auxiliary storage device. The main storage device is, for example, a random access memory (RAM), a read-only memory (ROM), or a combination of a RAM and a ROM. The auxiliary storage device is, for example, a hard disk drive (HDD), a solid state drive (SSD), or an EEPROM. The auxiliary storage device can store, for example, an operating system (OS) and multiple types of application programs (apps). The apps include programs for implementing various functions, such as a communication control program. The processor 121 is an example of a "control unit" and executes various programs stored in the storage device 122. This allows the information processing device 20 to operate as each NF constituting a 5GC, such as a VFL coordinator NF11, an NF (e.g., an NWDAF) operating as a VFL active / passive NF12, an NRF13, and an NF14.

[0051] The communication interface (communication IF) 124 is a circuit that controls communication processing, and performs transmission and reception of control information and user data (packets), format (protocol) conversion, etc. The input device 125 is a button, key, touch panel, etc. that are used for inputting and setting information, etc. The display 126 is used to display information. The input device 125 and display 126 are optional.

[0052] The processor 121 may be a processor other than a CPU, such as a DSP or a GPU, or may be combined with a CPU. Furthermore, the processing and operations by the processor 121 (the processing and operations as each NF performed by executing a program by the processor 121) may be performed using a semiconductor device (hardware) such as an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit), or may be performed using a combination of a processor and hardware such as an SoC (System On a Chip). Each of the processor 121 (controller), FPGA, ASIC, and SoC is an example of a "circuitry."

[0053] The above-described embodiment and modifications are merely examples, and the present disclosure may be modified as appropriate within the scope of the present disclosure. Furthermore, the processes and means described in the present disclosure may be freely combined and implemented as long as no technical contradiction occurs.

[0054] Furthermore, a process described as being performed by one device may be shared and executed by multiple devices. Alternatively, a process described as being performed by different devices may be executed by one device. In a computer system, the hardware configuration (server configuration) by which each function is realized can be flexibly changed.

[0055] The present disclosure can also be realized by supplying a computer program that implements the functions described in the above embodiments to a computer, and having one or more processors of the computer read and execute the program. Such a computer program may be provided to the computer by a non-transitory computer-readable storage medium connectable to the system bus of the computer, or may be provided to the computer via a network. The non-transitory computer-readable storage medium may be, for example, any type of disk, such as a magnetic disk (e.g., a floppy disk, a hard disk drive (HDD), etc.), an optical disk (e.g., a CD-ROM, a DVD disk, a Blu-ray disk), a read-only memory (ROM), a random access memory (RAM), an EPROM, an EEPROM, a magnetic card, a flash memory, or an optical card, for storing electronic instructions. Any suitable type of medium may be used. [Explanation of symbols]

[0056] 11. VFL Coordinator NF, 12. VFL Active / Passive NF, 13. NRF, 14. NF, 121. Processor, 122. Storage device

Claims

1. Identifying one or more features and one or more sample IDs to be used in vertical federated learning (VFL) from the collected data; filtering the identified one or more features and one or more sample IDs based on at least one of a privacy requirement, a minimum number requirement for features and / or sample IDs, and a maximum number requirement for features and / or sample IDs; Select one or more features and one or more sample IDs from the filtering results. A control unit that executes An information processing device comprising:

2. obtaining one or more features and one or more sample IDs from each of a plurality of vertical federated learning (VFL) participants; filtering the obtained one or more features and one or more sample IDs based on at least one of a privacy requirement, a minimum number requirement for at least one of features and sample IDs, and a maximum number requirement for at least one of features and sample IDs; transmitting one or more features and one or more sample IDs to be used for the VFL selected from the filtering results to the plurality of VFL participants; A control unit that executes An information processing device comprising:

3. The control unit performs the filtering based on a privacy indicator associated with at least one of a feature and a sample ID.

3. The information processing device according to claim 1 or 2.

4. The privacy indicator is a binary value or a specific integer value. The information processing device according to claim 3 .

5. The privacy indicator is associated with a particular combination of feature and sample ID. The information processing device according to claim 3 .

6. The privacy indicator is an integer value indicating a privacy level. The information processing device according to claim 3 .

7. The control unit performs the filtering on features based on an importance threshold and an importance metric associated with the features.

3. The information processing device according to claim 1 or 2.

8. At least one of the minimum number of requirements and the minimum number of requirements is obtained by pre-training, pre-configuration, or pre-definition.

3. The information processing device according to claim 1 or 2.

9. At least one of the minimum number related to the minimum number requirement and the minimum number related to the maximum number requirement is equal or different among the plurality of VFL participants.

3. The information processing device according to claim 1 or 2.

10. The control unit may perform the filtering of the features randomly with equal probability or in a predetermined order. Based on the priority associated with the VFL participant The information processing device according to claim 2 .

11. The information processing device Identifying one or more features and one or more sample IDs to be used in vertical federated learning (VFL) from the collected data; filtering the identified one or more features and one or more sample IDs based on at least one of a privacy requirement, a minimum number requirement for features and / or sample IDs, and a maximum number requirement for features and / or sample IDs; selecting one or more features and one or more sample IDs from the results of the filtering.

12. The information processing device obtaining one or more features and one or more sample IDs from each of a plurality of vertical federated learning (VFL) participants; filtering the obtained one or more features and one or more sample IDs based on at least one of a privacy requirement, a minimum number requirement for features and / or sample IDs, and a maximum number requirement for features and / or sample IDs; transmitting one or more features and one or more sample IDs to be used for the VFL selected from the filtered results to the plurality of VFL participants; An information processing method that performs the above.

13. The information processing device performs the filtering based on a privacy indicator associated with at least one of the features and the sample ID.

13. The information processing method according to claim 11 or 12.

14. The privacy indicator is a binary value or a specific integer value. The information processing method according to claim 13.

15. The privacy indicator is associated with a particular combination of feature and sample ID. The information processing method according to claim 13.

16. The privacy indicator is an integer value indicating a privacy level. The information processing method according to claim 13.

17. The information processing device performs the filtering on the features based on an importance threshold and an importance metric associated with the features.

13. The information processing method according to claim 11 or 12.

18. At least one of the minimum number of requirements and the maximum number of requirements is obtained by pre-training, pre-configuration, or pre-definition.

13. The information processing method according to claim 11 or 12.

19. At least one of the minimum number related to the minimum number requirement and the minimum number related to the maximum number requirement is equal or different among the plurality of VFL participants.

13. The information processing method according to claim 11 or 12.

20. The information processing device performs the filtering of features randomly with equal probability or based on priorities associated with the VFL participants. The information processing method according to claim 12.

Images