CN111553485A - View display method, device, equipment and medium based on federal learning model - Google Patents

View display method, device, equipment and medium based on federal learning model Download PDF

Info

Publication number
CN111553485A
CN111553485A CN202010370699.6A CN202010370699A CN111553485A CN 111553485 A CN111553485 A CN 111553485A CN 202010370699 A CN202010370699 A CN 202010370699A CN 111553485 A CN111553485 A CN 111553485A
Authority
CN
China
Prior art keywords
learning model
view
data
client
federated learning
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010370699.6A
Other languages
Chinese (zh)
Inventor
李�权
魏锡光
林焕彬
陈天健
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
WeBank Co Ltd
Original Assignee
WeBank Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by WeBank Co Ltd filed Critical WeBank Co Ltd
Priority to CN202010370699.6A priority Critical patent/CN111553485A/en
Publication of CN111553485A publication Critical patent/CN111553485A/en
Priority to PCT/CN2021/090983 priority patent/WO2021219080A1/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/32—Monitoring with visual or acoustical indication of the functioning of the machine
    • G06F11/323—Visualisation of programs or trace data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3452—Performance evaluation by statistical analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/40—Software arrangements specially adapted for pattern recognition, e.g. user interfaces or toolboxes therefor

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Quality & Reliability (AREA)
  • Evolutionary Biology (AREA)
  • Medical Informatics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Human Computer Interaction (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computer Hardware Design (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a view display method, a device, equipment and a medium based on a federal learning model, which relate to the field of financial science and technology, and the view display method based on the federal learning model comprises the following steps: acquiring operation data of the transverse federated learning model corresponding to each client in the iterative training process of the transverse federated learning model; constructing a visual view corresponding to the transverse federated learning model according to the operating data, and determining a training process of the transverse federated learning model; determining content displayed in the visual view according to the training process. The invention improves the success rate of the training of the transverse federal learning model and the accuracy of the identification data of the transverse federal learning model obtained by training.

Description

View display method, device, equipment and medium based on federal learning model
Technical Field
The invention relates to the technical field of federal learning of financial science and technology (Fintech), in particular to a view display method, a device, equipment and a medium based on a federal learning model.
Background
With the development of computer technology, more and more technologies are applied in the financial field, the traditional financial industry is gradually changing to financial technology (Fintech), and the artificial intelligence technology is no exception, but because of the requirements of the financial industry on safety and real-time performance, higher requirements are also put forward on the artificial intelligence technology.
Traditional machine learning adopts a centralized method, and data with different sources are aggregated in a computer or a data center for training, however, the data privacy is easily exposed by the centralized traditional machine learning method, and users have to sacrifice the privacy of themselves by sharing personal data to train a better machine learning model. In recent years, federal Learning (fed Learning) has enabled users to collaboratively train machine Learning models while retaining their own data, particularly private data containing private information, locally, in which case users can benefit from trained machine Learning models without sharing their sensitive personal data. Currently, the focus of horizontal federal Learning (horizontal federal Learning) in federal Learning is that data sets of different clients have the same feature space, but data samples are different, the running mechanism of horizontal federal Learning is more similar to a distributed Learning framework, and a security aggregation scheme is used to protect the privacy of users.
While federal learning works well in industrial and medical applications, among others, the following problems are encountered by the practices of federal learning when attempting to jointly model in their own setting: (1) the data available for viewing is limited, in the traditional centralized machine learning framework, a data center or a server side almost knows everything of the whole system, but the horizontal federal learning framework has no right to access the data of the client side due to the design of a data privacy mechanism, and cannot completely control the behavior of the client side. Therefore, potential risks such as malicious information in the client data are not visible to the federated learning server and may cause unexpected results to the federated learning server, thereby reducing the success rate of the lateral federated learning model training and resulting in poor accuracy of the trained lateral federated learning model identification data.
Therefore, the success rate is low in the process of training the horizontal federal learning model at present, and the accuracy of the recognition data of the horizontal federal learning model obtained by training is low.
Disclosure of Invention
The invention mainly aims to provide a view display method, a view display device, view display equipment and a view display medium based on a federated learning model, and aims to solve the technical problems that the success rate is low and the accuracy of the trained horizontal federated learning model identification data is low in the existing process of training the horizontal federated learning model.
In order to achieve the above object, the present invention provides a view display method based on a federal learning model, which comprises the steps of:
acquiring operation data of the transverse federated learning model corresponding to each client in the iterative training process of the transverse federated learning model;
constructing a visual view corresponding to the transverse federated learning model according to the operating data, and determining a training process of the transverse federated learning model;
determining content displayed in the visual view according to the training process.
Optionally, the operational data includes at least one of: the method comprises the steps that client identification, a starting timestamp of each iterative training, the number of training samples corresponding to the client, a loss value corresponding to a local model, and a starting time point and an ending time point of each client for training the local model through local data are obtained, wherein a visual view comprises an overview view;
the step of constructing a visual view corresponding to the lateral federated learning model based on the operational data includes:
performing visual coding on the operation data to obtain visualized operation data;
and constructing a summary view corresponding to the transverse federated learning model according to the visualized operating data.
Optionally, the visual view includes a projection view, and the step of constructing the visual view corresponding to the lateral federal learning model according to the operating data includes:
determining index data corresponding to each client according to the operation data;
and constructing a projection view corresponding to the transverse federated learning model according to the index data, wherein each node in the projection view represents a mapping relation between a client identifier and the iteration times respectively.
Optionally, the visual view includes a summary view, and the step of constructing the visual view corresponding to the lateral federated learning model according to the operation data includes:
determining statistical data corresponding to the operation data, wherein the statistical data at least comprises one of the following data: the number of clients corresponding to the transverse federated learning model, the number of iterations, the number of changes of the number of samples to be trained for training the transverse federated learning model, a reduction value corresponding to a loss value corresponding to the transverse federated learning model, and an increase value of the recognition accuracy of a local model corresponding to each client;
and constructing a summary view corresponding to the transverse federated learning model according to the statistical data.
Optionally, the visual view includes a comparison view, and the step of constructing the visual view corresponding to the lateral federated learning model according to the operation data includes:
determining index data corresponding to each client according to the operation data, and determining a target index corresponding to the last iterative training of the transverse federated learning model in the index data;
and constructing a comparison view corresponding to the transverse federated learning model according to the target indexes.
Optionally, the visual view includes a contribution ranking view, and the step of constructing the visual view corresponding to the lateral federated learning model according to the operation data includes:
determining the ranking sequence of the running data of the client in each iterative training;
and displaying the ranking training in a box and whisker graph mode to construct a contribution degree sequencing view corresponding to the transverse federated learning model.
Optionally, after the step of constructing a visual view corresponding to the lateral federal learning model according to the operating data and determining a training process of the lateral federal learning model, the method further includes:
detecting whether an operation instruction for operating the visual view is received;
if the operation instruction is received, the step of determining the content displayed in the visual view according to the training process includes:
and determining the content displayed in the visual view according to the operation instruction and the training process.
In addition, to achieve the above object, the present invention provides a view display apparatus based on a federal learning model, including:
the acquisition module is used for acquiring the running data of each client corresponding to the transverse federated learning model in the iterative training process of the transverse federated learning model;
the construction module is used for constructing a visual view corresponding to the transverse federated learning model according to the operating data;
a determining module for determining a training process of the lateral federated learning model; determining content displayed in the visual view according to the training process.
In addition, in order to achieve the above object, the present invention further provides a view display device based on a federal learning model, which includes a memory, a processor and a view display program based on a federal learning model stored in the memory and operable on the processor, wherein the view display program based on the federal learning model, when executed by the processor, implements the steps of the view display method based on the federal learning model corresponding to the federal learning server.
In addition, to achieve the above object, the present invention further provides a computer readable storage medium, on which a view display program based on the federal learning model is stored, and when being executed by a processor, the view display program based on the federal learning model implements the steps of the view display method based on the federal learning model as described above.
According to the method, the operation data of the transverse federated learning model corresponding to each client side in the iterative training process of the transverse federated learning model are obtained, the visual view corresponding to the transverse federated learning model is constructed according to the operation data, the training process of the transverse federated learning model is determined, and the content displayed in the visual view is controlled according to the training process. The method and the device have the advantages that the training process data corresponding to the iterative training of the transverse federated learning model are determined through the content displayed by the visual view, various influence factors in the iterative training process are determined through the training process data, and the unexpected result of the federated learning server caused by the potential risks of malicious information and the like in the client data is avoided, so that the success rate of the transverse federated learning model training is improved, and the accuracy of the identification data of the transverse federated learning model obtained through training is improved.
Drawings
FIG. 1 is a schematic flow chart diagram of a first embodiment of a view display method based on a federated learning model in accordance with the present invention;
FIG. 2 is a schematic illustration of an overview view in an embodiment of the invention;
FIG. 3 is a diagram illustrating loss values, recognition accuracy, and number of training samples after visualization in an embodiment of the present invention;
FIG. 4 is a schematic view of a projection view in an embodiment of the invention;
FIG. 5 is a diagram illustrating a ranking view of the contribution score in accordance with an embodiment of the present invention;
FIG. 6 is a block diagram of a preferred embodiment of a federated learning model-based view display apparatus in accordance with the present invention;
fig. 7 is a schematic structural diagram of a hardware operating environment according to an embodiment of the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention provides a view display method based on a federal learning model, and referring to fig. 1, fig. 1 is a flow diagram of a first embodiment of the view display method based on the federal learning model.
While a logical order is shown in the flow chart, in some cases, the steps shown or described may be performed in a different order than presented herein.
The view display method based on the federal learning model is applied to the federal learning server, and for convenience of description, an executive body is omitted for explaining various embodiments. The view display method based on the federal learning model comprises the following steps:
and step S10, acquiring the operation data of the transverse federated learning model corresponding to each client in the iterative training process of the transverse federated learning model.
When a construction instruction for constructing a visual view corresponding to the transverse federated learning model is detected, the operation data of the transverse federated learning model corresponding to each client side in the iterative training process of the transverse federated learning model is obtained according to the construction instruction. The construction instruction can be triggered by a user according to specific needs, and can also be triggered when the transverse federated learning model is started to be constructed, namely when the transverse federated learning model is subjected to iterative training for the first time, the construction instruction is automatically triggered. Each client has corresponding operation data, and in this embodiment, the operation data includes at least one of the following: the method comprises the steps of client identification, client names, a starting timestamp of each time iterative training is started on a transverse federated learning model, the training times of a local model of the client, a loss value corresponding to the local model after each iterative training, the recognition accuracy of the local model for data recognition after each iterative training, and a starting time point and an ending time point of each time local model is trained by the client through local data. The local model is a model which is sent to each client after the horizontal federal learning model is obtained by the federal learning server and is stored by each client. After each client receives the transverse federal learning model, the received transverse federal learning model is used as a local model of the client, and the local model can be adjusted by adopting local data of the client. It will be appreciated that the operational data does not relate to the private data of the respective client user.
Specifically, the client identifier is used for uniquely representing a certain client, and the client identifier and the client name can be transmitted together in the transverse federated learning iterative training process; the starting timestamp of each time the horizontal federated learning model starts iterative training can be obtained by a timer when the iterative training starts; the training times of the local model of the client can be sent to the federated learning server in each iterative training process, and it needs to be stated that the training times can be equal to or unequal to the iterative times; each iterative training can obtain a corresponding loss value; the identification precision can be determined by obtaining preset data to be tested after each iterative training to obtain a local model, inputting the data to be tested into the local model to obtain an output result of the transverse local model, and comparing the output result with a correct result corresponding to the data to be tested.
And S20, constructing a visual view corresponding to the transverse federated learning model according to the running data, and determining the training process of the transverse federated learning model.
And after the operation data are obtained, constructing a visual view corresponding to the transverse federated learning model according to the operation data, and determining a training process of the transverse federated learning model. Wherein the visual view includes at least one of: the system comprises a summary view, an abstract view, a projection view, a contrast view and a contribution ranking view, wherein the summary view can display the whole operation process of each client in the process of iteratively training a transverse federated learning model; the projection view can display the mapping relation between the client terminal identification and each iterative training in two-dimensional distribution; the abstract view is used for displaying statistical information corresponding to various data in the iterative training process of the transverse federated learning model; the comparison view is used for displaying the comparison condition of the corresponding index data of any two clients in the iterative training process of the transverse federated learning model; and the contribution degree ranking view is used for displaying the contribution degree of each client to the transverse federated learning model in different dimensions. The training process of the transverse federated learning model can be represented by the number of iterations of iterative training performed by the transverse federated learning model, and it can be understood that when the transverse federated learning model is trained to a convergence state, a certain number of iterative training can be performed, and each iterative training changes model parameters corresponding to the transverse federated learning model, correspondingly, in each iterative training process, part of operating data corresponding to each client also changes, such as a loss value and recognition accuracy change, and therefore, in this embodiment, the training process of the transverse federated learning model can be determined by the number of iterations. Specifically, in the iterative training process of the transverse federated learning model, the iteration times of the transverse federated learning model are calculated through a timer, the transverse federated learning model is iteratively trained once, and the value corresponding to the timer is added by 1, so that when the training process of the transverse federated learning model needs to be determined, the value corresponding to the timer is obtained, the iteration times of transverse federated learning are determined according to the value, and the training process of the transverse federated learning model is determined.
Further, the operational data includes at least one of: the method comprises the steps that client identification, a starting timestamp of each iterative training, the number of training samples corresponding to the client, a loss value corresponding to a local model, and a starting time point and an ending time point of each client for training the local model through local data are obtained, wherein a visual view comprises an overview view; the step of constructing a visual view corresponding to the lateral federated learning model based on the operational data includes:
step a, performing visual coding on the operation data to obtain visualized operation data.
Further, when the visual view is the overview view, each piece of operation data is visually encoded to obtain the visualized operation data. It should be noted that, the processing procedures of the operation data of each client are the same, so for convenience of description, the embodiment takes the operation data of one client as an example for description. Specifically, in order to measure the change of the 'network structure' of the horizontal federal learning network caused by the addition and the exit of a client in the iterative training process of the horizontal federal learning model, the change process of the horizontal federal learning network corresponding to the visual horizontal federal learning model needs to be explained.
In this embodiment, the position of each client in the horizontal federated learning network is determined in each iterative training process, specifically, the position of the corresponding client in the horizontal federated learning network can be determined through the client identifier and the start timestamp of the current iterative training, specifically, the position of each client in the horizontal federated learning network can be preset in each iterative training, and the mapping relationship among the client identifier, the start timestamp, and the position identifier is preset. After the starting timestamp and the client identifier of the current iterative training are determined, the position of the corresponding client in the transverse federated learning network can be determined through the starting timestamp, the client identifier and the mapping relation.
The method comprises the steps that a start timestamp of a visual horizontal federated learning model when iterative training is started each time, a loss value corresponding to a local model after iterative training each time, identification precision of the local model for data identification after iterative training each time, a start time point and an end time point of a client side through local data when the local model is trained each time and the like are correspondingly obtained, and the start timestamp, the loss value, the identification precision, the start time point and the end time point after visualization are correspondingly obtained. Specifically, a box whisker graph is adopted to represent the distribution of the recognition accuracy and the loss value of each client in each iterative training process, and curves are used to connect the average values corresponding to the recognition accuracy and the loss value in each iterative training process so as to obtain the visualized recognition accuracy and the visualized loss value, wherein the box whisker graph is a statistical graph used for displaying a group of data dispersion situation data. The number of training samples corresponding to each client in each iterative training process is visualized as a curve to obtain the number of visualized training samples, further, a corresponding bar graph can be added in a region corresponding to the curve, the total number of training samples in the corresponding iterative training process is represented by the bar graph, and it can be understood that the total number of training samples is equal to the sum of the number of training samples corresponding to each client. The training time of each client for training the local model can be calculated through the starting time point and the ending time point of the local model trained through the local data of the client, the training time of each client is represented through the slope of the connecting line between the starting time point and the ending time point, and therefore the visualized ending time point and the visualized starting time point are obtained. Specifically, referring to fig. 3, fig. 3 is a schematic diagram of a loss value, a recognition Accuracy and a Number of training samples after visualization in an embodiment of the present invention, where, from left to right in fig. 3, a first graph represents a loss value (loss) after visualization, a second graph represents a recognition Accuracy (Accuracy) after visualization, and a third graph represents a Number of training samples (sample Number) after visualization.
And b, constructing a summary view corresponding to the transverse federated learning model according to the visualized operating data.
And after the visualized operation data is obtained, constructing a summary view corresponding to the transverse federated learning model according to the visualized operation data. It can be understood that the overview view is composed of various visualized operation data, and the operation condition of each client in the horizontal federated learning model iterative training process can be determined through the overview view.
It should be noted that, through the overview view, it can be found that the client participating in the horizontal federated learning model iterative training changes as the iterative training process evolves. The change of the federated learning network can be seen through the overview view, and as can be determined through the overview view, in each iterative training process, the number of samples provided by each client participating in the iterative training is generally uniformly distributed, that is, the difference value between the number of samples provided by each client is within a preset range, and the preset range can be set according to specific needs; through the overview view, the starting time point and the ending time point of the local model training performed by each client can also be determined, and it can be understood that the starting time point of the local model training of each client may be different, which may be caused by network delay. Because the local data quantity of each client for training the local model by adopting the local data is different, the training time of each client for training the local model is different. It can be understood that the training duration for each client to train the local model is determined through the overview view, and the starting time for the next iterative training is adjusted according to the training duration, that is, the waiting duration between two adjacent iterative training is adjusted, so as to better adapt to the training duration corresponding to each client. It can be understood that the waiting duration is greater than or equal to the maximum training duration corresponding to the client. Specifically, through the overview view, it can be determined that as the number of iterations increases, the loss value is continuously reduced, and the recognition accuracy is continuously improved. Furthermore, through the overview view, it can also be determined which iterative training corresponds to a larger recognition accuracy change amplitude, and which iterative training corresponds to a smaller recognition accuracy change amplitude.
Specifically, referring to fig. 2, fig. 2 is a schematic diagram of an overview view in an embodiment of the present invention, in fig. 2, each rounded rectangle box represents one iteration training, a small circle in each rounded rectangle box represents a client participating in the current iteration training, as can be seen from fig. 2, the iteration times are aligned in the y (vertical axis) direction, and in each rounded rectangle box, the appearance sequence of each small circle represents the appearance sequence of each client in the iteration training process; the coordinates in the x (horizontal axis) direction in fig. 2 are adjusted so that the respective links in fig. 2 do not intersect to minimize the utilization of the entire space. Note that the small solid circles in fig. 2 represent clients that have participated in only part of the iterative training. As can be seen from fig. 2, the clients involved in each iterative training process are visualized through an overview view.
Step S30, determining the content displayed in the visual view according to the training process.
When the training process of the transverse federated learning model is determined, the content displayed in the visual view is controlled according to the training process, so that the training process data of the iterative training of the transverse federated learning model is determined through the content displayed in the visual view. It is understood that the training process data obtained from the different visual views are also different, for example, the training process data corresponding to the overview view is another representation of the operation data of each client in the iterative training process of the horizontal federal learning model of each client. With the change of the training process of the transverse federated learning model, the model parameters of the transverse federated learning model are changed, and the operation data of the client are also changed, so that the contents displayed in the corresponding visual views are different, that is, the contents displayed in the visual views are changed with the change of the training process of the transverse federated learning model, and the operation condition of the client in the training process of the transverse federated learning model can be checked through the visual views. It can be understood that, in the embodiment, whether each client has an abnormal condition may be determined according to the content displayed by the corresponding visual view of each client, and if there is a large difference between the visual view of one client and the visual views of other clients, it may be determined that the client may have an abnormal condition.
In the embodiment, in the iterative training process of the transverse federated learning model, the operating data of the transverse federated learning model corresponding to each client is obtained, the visual view corresponding to the transverse federated learning model is constructed according to the operating data, the training process of the transverse federated learning model is determined, and the content displayed in the visual view is controlled according to the training process. The method and the device have the advantages that the training process data corresponding to the iterative training of the transverse federated learning model are determined through the content displayed by the visual view, various influence factors in the iterative training process are determined through the training process data, and the unexpected result of the federated learning server caused by the potential risks of malicious information and the like in the client data is avoided, so that the success rate of the transverse federated learning model training is improved, and the accuracy of the identification data of the transverse federated learning model obtained through training is improved.
Further, traditional centralized machine learning typically performs model training and inference in a separate manner, whereas federated learning servers typically couple the training and inference processes. In other words, the federal learning server is a data distribution that can be continuously updated to accommodate changes that may occur, and it is not sufficient for federal learning server maintenance personnel to rely solely on simple logs and metrics to interpret information at a given stage or time and to make quick and informed decisions in a short amount of time. Therefore, there is a strong need for a method to efficiently express "spatiotemporal data" that varies over time from different clients. This is done to facilitate phased adjustment of the federated learning aggregation strategy and timely review of the iterative training process of the lateral federated learning model for more effective intervention. In the embodiment, the content displayed in the visual view is controlled according to the training process, the training process data corresponding to the iterative training of the horizontal federal learning model is determined according to the content displayed in the visual view, the content displayed in the visual view is changed along with the change of the training process of the horizontal federal learning model, and the change condition of the client corresponding to the horizontal federal learning model in the iterative training process is achieved through the visual view, so that relevant operation and maintenance personnel can timely review the iterative training process of the horizontal federal learning model, so that the intervention can be more effectively performed, and the horizontal federal learning model obtained by training can be optimized.
Further, the invention provides a view display method based on the federal learning model in a second embodiment. The second embodiment of the view display method based on the federal learning model is different from the first embodiment of the view display method based on the federal learning model in that the visual view comprises a projection view, and the step of constructing the visual view corresponding to the transverse federal learning model according to the operating data comprises the following steps:
and c, determining index data corresponding to each client according to the operation data.
And when the visual view required to be constructed is a projection view and the operation data is acquired, determining the index data corresponding to each client according to the operation data. It should be noted that, when the projection view is constructed, the operation data may further include a gradient histogram corresponding to the local model and a weight histogram corresponding to the local model, and it should be noted that, after the loss value is obtained, the gradient of the model parameter corresponding to each local model may be determined according to the loss value, so that the corresponding gradient histogram may be obtained according to the determined gradient. The weight is the weight corresponding to each model parameter in the local model. In this embodiment, the index data includes at least one of: loss values, recognition accuracy, training evolution number, weight histogram, and gradient value histogram.
And d, constructing a projection view corresponding to the transverse federated learning model according to the index data, wherein each node in the projection view represents a mapping relation between a client identifier and the iteration times respectively.
After the index data is obtained, a projection view corresponding to the transverse federated learning model is constructed based on t-SNE (t-distributed stored probabilistic neighbor embedding) projection according to the index data, wherein the projection view is a 2D (two-dimensional) view, and the t-SNE is a dimension reduction technology used for creating a low-dimensional representation and retaining local similarity to convey a neighborhood structure. It is to be understood that the projection view may also be constructed by using dimensionality reduction techniques such as PCA (Principal Component Analysis) and MDS (multidimensional scaling). Potential clustering and abnormal values in the process of the iterative training of the transverse federated learning model can be checked through the projection view, so that the abnormal client side in the process of the iterative training of the transverse federated learning model is determined. In the projection view, there is at least one node, and each node represents a mapping relationship between a stack of "client identifier-iteration number", that is, in the present embodiment, the "client identifier-iteration number" is projected into the two-dimensional view. The corresponding projection views are different for different iteration times. In the projection view, the federal learning server is a special client in the projection view.
In the projection view, the federal learning server corresponding to the first iterative training is used as a starting point of the projection view, the federal learning server corresponding to the last iterative training is used as an end point of the projection view, the nodes appearing in the middle are clients involved in the iterative training process, and then all the nodes are connected, so that the evolution process of the clients in the iterative training process of the transverse federal learning model can be determined through the projection view.
Specifically, referring to fig. 4, fig. 4 is a schematic diagram of a projection view in an embodiment of the invention. In fig. 4, two solid small circles before and after the connection line represent the first iterative training and the last iterative training of the horizontal federated learning model, and a hollow small circle in the middle of the connection line represents the mapping relationship between the client identifier and the iteration number of each client participating in the horizontal federated learning model, and if a certain client identifier is a and the iteration number is 10 th, a hollow small circle represents "a-10" in fig. 4. It should be noted that, if a small circle deviates from the curve more, it is indicated that the contribution degree of the client corresponding to the small circle in the iterative training process of the horizontal federal learning model is smaller, and it is more likely that an abnormal client exists, that is, the client corresponding to the small circle may be determined as the abnormal client.
Further, the visual views comprise abstract views, and the step of constructing the visual view corresponding to the lateral federated learning model according to the operation data comprises the following steps:
step e, determining statistical data corresponding to the operating data, wherein the statistical data at least comprises one of the following data: the number of clients corresponding to the transverse federated learning model, the number of iterations, the number of changes of the number of samples to be trained for training the transverse federated learning model, a reduction value corresponding to a loss value corresponding to the transverse federated learning model, and an increase value of the recognition accuracy of a local model corresponding to each client;
and f, constructing an abstract view corresponding to the transverse federated learning model according to the statistical data.
Further, when a visual view to be constructed is an abstract view and operation data is acquired, determining statistical data corresponding to the operation data, wherein the statistical data at least comprises one of the number of clients corresponding to a transverse federated learning model, iteration times, the number of changes of samples to be trained for training the transverse federated learning model, a reduction value corresponding to a loss value corresponding to a transverse federated learning model and an increase value of identification accuracy of a local model corresponding to each client, the number of changes is a difference value of the number of samples to be trained corresponding to two adjacent iterative trainings, and the data difference value is equal to the number of samples to be trained corresponding to the next iterative training minus the number of samples to be trained corresponding to the previous iterative training; the reduction value is a loss difference value between corresponding loss values of two adjacent iterative training, the loss difference value is equal to a loss value corresponding to the previous iterative training minus a loss value corresponding to the next iterative training, and the loss value of the local model is calculated by the transverse federal learning model; the increment value is equal to the difference between the recognition precision of the local model corresponding to the last iterative training and the recognition precision of the local model corresponding to the previous iterative training.
And after the statistical data are obtained, constructing an abstract view corresponding to the horizontal federal learning model according to the statistical data, wherein the abstract view can display the statistical data in a form of a table or a chart.
Further, the step f includes:
and f1, constructing a summary view of each statistical data corresponding to the transverse federal learning model by taking the iteration times corresponding to the transverse federal learning model as a horizontal coordinate and the corresponding statistical data as a vertical coordinate according to the statistical data.
Specifically, in the process of constructing the abstract view, the iteration times corresponding to the horizontal federal learning model are abscissa, and the corresponding statistical data are ordinate, so that the abstract view of each statistical data corresponding to the horizontal federal learning model is constructed. For example, in the process of constructing the abstract views corresponding to the number of the clients, the number of iterations is used as an abscissa, in each iterative training process, the number of the clients participating in the iterative training is used as an ordinate, the abstract views corresponding to the number of the clients are constructed, and the change of the number of the clients in the iterative training can be seen through the abstract views. It can be understood that the principle of the construction process is similar for various statistical data, and the description is not repeated in this embodiment. It can be understood that, in the embodiment, the abstract view corresponding to the number of clients, the number of changes of the number of samples to be trained, the reduction value corresponding to the loss value, the increase value of the recognition accuracy, and the like can be constructed.
Further, the visual views include comparison views, and the step of constructing the visual view corresponding to the lateral federated learning model according to the operation data includes:
and g, determining index data corresponding to each client according to the operation data, and determining a target index corresponding to the last iterative training of the transverse federated learning model in the index data.
Further, when the visual view to be constructed is a comparison view and the operation data is acquired, determining the index data corresponding to each client according to the operation data, wherein the index data index at least comprises one of the following data: identifying accuracy, loss values, training times of local models, weight histograms and gradient histograms. Note that the index data is a part of the operation data. In this embodiment, which operation data are set in advance as index data corresponding to the client, specifically, a specific index identifier may be added to the index data, after the operation data are obtained, which operation data carry the index identifier is detected, and the operation data carrying the index identifier is determined as the index data, which does not limit the expression form of the index data.
It should be noted that the comparison view is constructed by at least index data between two clients. And after the index data is determined, determining a target index corresponding to the last iterative training of the transverse federated learning model in the index data, wherein the target index is a comparison standard in the comparison view, namely, the target index is used for comparing the related data of the two clients. It is understood that the target index is at least one, and in the present embodiment, the weight histogram and the gradient histogram may be set as the target index.
And h, establishing a comparison view corresponding to the transverse federated learning model according to the target index.
And after the target index is determined, establishing a comparison view corresponding to the transverse federated learning model according to the target index. Specifically, by comparing views, a weight histogram and a gradient histogram corresponding to each client in each iteration number can be obtained, then the weight histogram and the gradient histogram between at least two clients in the same iteration number or different iteration numbers are compared to obtain the similarity between the weight histogram and the gradient histogram between at least two clients, and the similarity is used as a new index value, that is, the weight histogram and the gradient histogram are converted into a numerical value.
Specifically, step h includes:
and h1, acquiring the target indexes corresponding to the adjacent clients in the federated learning network structure in the same iterative training process, and constructing a corresponding comparison view according to the target indexes corresponding to the adjacent clients.
Specifically, in the component view comparison process, each iterative training process can be compared, the target indexes corresponding to the adjacent clients in the federated learning network structure are used as abscissa, and the target indexes corresponding to the adjacent clients are used as ordinate to construct the comparison view corresponding to the adjacent clients. It should be noted that, in the process of training the horizontal federated learning model, due to the nature of the federated learning network, the characteristics of the adjacent clients in the federated learning network are similar, and therefore, if the comparison view between the two adjacent clients indicates that the two clients are greatly different, it may be determined that one of the clients has an abnormal condition in the iterative training process, and at this time, the client having the abnormal condition may be determined by combining the comparison views between the two clients and the other clients. It will be appreciated that except for the first and last clients in the federated learning network architecture, there are two adjacent clients for the remaining clients, one left adjacent and one right adjacent.
Further, in the comparison view in this embodiment, a rectangle in each row represents one client, bars in different colors in the rectangle represent values corresponding to each index data of the client, starting points of the same index data bar corresponding to different clients are the same, and thus, the similarity between each index data can be determined by the end point of the bar. The top row in the comparison view may be used to represent the federated learning server. Further, comparison views corresponding to different iteration times can be obtained, and then the comparison views corresponding to different iteration times are displayed in the same interface, so that the arrangement conditions of the same client in different iteration times can be checked through the comparison views corresponding to different iteration times. It should be noted that, for a comparison view corresponding to the same iteration number, if the comparison view contains all the index data of the iterative training, the ordering condition of the index data corresponding to each client in the comparison view can be determined as needed, and if there is identification accuracy, the clients can be ordered from large to small according to the identification accuracy to obtain the comparison view, so that the identification accuracy corresponding to each client in the current iterative training process can be checked through the comparison view.
Further, after the index data of the client corresponding to the horizontal federal learning model is determined, target indexes of various types of index data are selected, and the similarity between each client index data and the corresponding target index is calculated, wherein the greater the similarity is, the more the ranking in the comparison view is, the smaller the similarity is, and the more the ranking in the comparison view is. Specifically, the similarity between each client index data and the corresponding target index may be calculated by a euclidean distance or a cosine distance. In the comparative view, the similarity may be represented by a curve. It can be understood that if the calculated similarity is greater than the preset similarity, it indicates that the corresponding client has a large change in the iterative training process of the horizontal federated learning model; if the calculated similarity is smaller than or equal to the preset similarity, it indicates that the corresponding client belongs to normal change in the iterative training process of the horizontal federated learning model, where the size of the preset similarity may be set according to specific needs, and the embodiment does not limit the size of the preset similarity.
It should be noted that, in the iterative training process, it may be determined that, in comparison with other normal fluctuating clients, there is an obviously fluctuating client, at this time, nodes corresponding to the obviously fluctuating client and the normal fluctuating client may be selected in the comparison view through an operation instruction, so that the two clients are compared to check a difference between running data such as a loss value and recognition accuracy corresponding to the two clients, and determine the obviously fluctuating client. Further, by comparing the views, whether the gradient change condition of each client deviates from the normal condition can be checked. Therefore, by comparing the views, the abnormal client side of the transverse federated learning model in the iterative training process can be determined.
Further, the visual views comprise contribution degree sequencing views, and the step of constructing the visual view corresponding to the horizontal federated learning model according to the operation data comprises the following steps:
and step i, determining the ranking sequence of the running data of the client in each iterative training.
And j, displaying the ranking training in a box and whisker graph mode to construct a contribution degree sequencing view corresponding to the transverse federated learning model.
Further, when the visual view required to be constructed is a contribution ranking view and the operation data is acquired, determining the ranking sequence of the operation data of each client in each iterative training process. After the ranking sequence of the running data of each client in each iterative training is determined, the ranking sequence is displayed in a box and whisker graph mode to construct a contribution degree ranking view corresponding to the horizontal federal learning model, and it should be noted that in the contribution degree ranking view, the ranking sequence can be ranked from small to large or from large to small. Specifically, in the contribution ranking view, the clients may be ranked according to the lowest ranking, the highest ranking, the median ranking and the number of participating iterations of each client, for example, it is determined that, in the number of participating iterations of a certain client, the running data of several times are ranked the highest, the running data of several times are ranked the lowest, and the running data of several times are ranked in the middle. It should be noted that a client does not necessarily participate in all iterations of the horizontal federated learning model, for example, the number of iterations of the horizontal federated learning model is 100 in total, and a client may participate in only 65 of the iterations. When one of the running data of the client is ranked highest in a certain iterative training process, it can be determined that the client has the highest ranking 1 time in the current iterative training process. It can be understood that, for the same operation data, the corresponding contribution degree can be determined according to the ranking order, for example, if a certain operation data is ranked highest, the maximum or minimum contribution degree of the operation data to the horizontal federal learning model can be determined according to the property of the operation data.
Further, comparing the rankings in the views may also affect the ranking of the contribution ranking views. For example, when selecting the loss values in the comparison view for sorting, each client gets a rank according to the loss values in each round, so that each client has a rank in each round, the box and whisker graph is used to represent the rank distribution of the client, and the rank distribution is shown in the contribution sorting view, and at this time, the contribution sorting view shows the contribution sorting when the attribute selected by the user is the loss value.
It should be noted that the contribution ranking view shows participation of all clients in the iterative training process of the horizontal federal learning model by using the design of the box-whisker graph, and the clients are sorted in a descending order or an ascending order. Just as in joint learning, the local data of the client is completely invisible to the federated learning server, and the contribution of the client to the horizontal federated learning model can be known through the ranking of different running data. The loss rate and the recognition accuracy may reflect the quality of the training data of each client, the number of the training data represents the contribution of the data, and the loss rate represents the proportion of the unused sample data in the sample data to be trained provided by the client to the total sample data to be trained provided by the client. It can be understood that by ranking the clients by using different operation data, it can be determined that the operation data of the client with the abnormality is not all arranged in the last iterative training process, so that the abnormality of the client is allowed to appear only in a few iterative training processes in the iterative training process of the horizontal federal learning model, but the client is normal in most iterative training processes.
Specifically, referring to fig. 5, fig. 5 is a schematic diagram of a contribution ranking view in an embodiment of the present invention, in fig. 5, a y (vertical) axis is a client identifier of a client, an x (horizontal) axis is a ranking distribution, and fig. 5 is a ranking distribution of minimum running data of each client in each iterative training process. As can be seen from fig. 5, the longer the length of the rectangular box corresponding to each client indicates that the minimum values in the running data of the client are more in the iterative training process of the horizontal federal learning model.
It will be appreciated that existing visual analysis tools such as the Turbofan tycoon or face-Board, which help facilitate analysis and improvement of the federal learning model by summarizing logs and performance index data generated by the federal learning process, can be used to convey the advantages of the federal learning model. However, in-depth analysis is lacking, and fine-grained analysis such as analysis of potential client anomalies and contribution assessments is challenging, e.g., design of privacy preserving mechanisms can hinder many basic operations during the course of horizontal federal learning model training. If the effective analysis cannot be carried out to provide support for subsequent optimization adjustment, the effect of the whole horizontal federal learning model training is influenced, namely the accuracy of the recognition data of the horizontal federal learning model obtained by training is low. In the embodiment, different visual views are constructed, namely a comparison view, an abstract view, a projection view, a contribution degree sequencing view and the like, the operation conditions of each client are analyzed from different dimensions through the different visual views in the iterative training process of the transverse federated learning model, the abnormal conditions of each client are found in time, and then the training process of the transverse federated learning model is adjusted in time, so that the accuracy of the identification data of the transverse federated learning model obtained by training is further improved.
Further, a third embodiment of the view display method based on the federal learning model is provided. The third embodiment of the federal learning model-based view display method differs from the first and/or second embodiment of the federal learning model-based view display method in that the federal learning model-based view display method further includes:
and k, detecting whether an operation instruction for operating the visual view is received.
If the operation instruction is received, the step of determining the content displayed in the visual view according to the training process includes:
and step l, determining the content displayed in the visual view according to the operation instruction and the training process.
After the visual view is created, whether an operation instruction for operating the visual view is received or not is detected, wherein the operation instruction is started by a user according to specific needs. After receiving the operation instruction, controlling the content displayed in the visual view according to the operation instruction and the training process; and when the operation instruction is not received, continuously detecting whether the operation instruction for operating the visual view is received or not. If a user can select a certain node in the projection view through an operation instruction, then, relevant data corresponding to the selected node in the current training process can be displayed in the abstract view, the overview view, the comparison view and the contribution degree ranking view; for example, when there are comparison views corresponding to multiple iterative trainings and there are running data corresponding to multiple clients in each comparison view, when a certain client is selected in one of the comparison views, that is, when a rectangle corresponding to the certain client is clicked in one of the comparison views, relevant data corresponding to the client is distinguished and displayed in other comparison views, for example, relevant data corresponding to the client in other comparison views is highlighted.
According to the embodiment, the content displayed in the visual view is controlled according to the operation instruction and the training process, so that the display of the related content according to the user requirement is realized, and the intelligence of the content displayed in the visual view is improved. Further, according to the embodiment of the invention, the overall operation state of each client and the correlation condition of the operation data among the clients in the iterative training process of the transverse federated learning model can be determined through the visual views such as the abstract view, the projection view, the overview view, the comparison view and the contribution degree ranking view, so that the abnormal condition existing in the iterative training process of the transverse federated learning model and the contribution condition of each client to the transverse federated learning model can be detected according to the visual views.
In addition, the present invention also provides a view display apparatus based on the federal learning model, and referring to fig. 6, the view display apparatus based on the federal learning model includes:
the obtaining module 10 is configured to obtain operation data of each client corresponding to a horizontal federal learning model in an iterative training process of the horizontal federal learning model;
a construction module 20, configured to construct a visual view corresponding to the transverse federated learning model according to the operating data;
a determining module 30, configured to determine a training process of the lateral federated learning model; determining content displayed in the visual view according to the training process.
Further, the operational data includes at least one of: the method comprises the steps that client identification, a starting timestamp of each iterative training, the number of training samples corresponding to the client, a loss value corresponding to a local model, and a starting time point and an ending time point of each client for training the local model through local data are obtained, wherein a visual view comprises an overview view;
the building block 20 comprises:
the coding unit is used for carrying out visual coding on the operation data to obtain visualized operation data;
the first construction unit is used for constructing the overview view corresponding to the transverse federated learning model according to the visualized operating data.
Further, the visual view comprises a projection view, and the construction module 20 further comprises:
the first determining unit is used for determining index data corresponding to each client according to the operating data;
and the second construction unit is used for constructing a projection view corresponding to the transverse federated learning model according to the index data, wherein each node in the projection view represents a mapping relation between a client identifier and the iteration times respectively.
Further, the visual view includes a summary view, and the building module 20 further includes:
a second determining unit, configured to determine statistical data corresponding to the operating data, where the statistical data at least includes one of: the number of clients corresponding to the transverse federated learning model, the number of iterations, the number of changes of the number of samples to be trained for training the transverse federated learning model, a reduction value corresponding to a loss value corresponding to the transverse federated learning model, and an increase value of the recognition accuracy of a local model corresponding to each client;
and the third construction unit is used for constructing the abstract view corresponding to the transverse federated learning model according to the statistical data.
Further, the visual view comprises a comparison view, and the construction module 20 further comprises:
the third determining unit is used for determining index data corresponding to each client according to the operating data and determining a target index corresponding to the last iterative training of the transverse federated learning model in the index data;
and the fourth construction unit is used for constructing a comparison view corresponding to the transverse federated learning model according to the target index.
Further, the visual view comprises a contribution ranking view, and the building module 20 further comprises:
the fourth determining unit is used for determining the ranking sequence of the running data of the client in each iterative training;
and the display unit is used for displaying the ranking training in a box and whisker graph mode so as to construct a contribution degree sequencing view corresponding to the transverse federated learning model.
Further, the view display device based on the federal learning model further comprises:
the detection module is used for detecting whether an operation instruction for operating the visual view is received or not;
the determining module 30 is further configured to determine, if the operation instruction is received, content displayed in the visual view according to the operation instruction and the training process.
The specific implementation of the view display device based on the federal learning model of the present invention is basically the same as that of each embodiment of the view display method based on the federal learning model, and is not described herein again.
In addition, the invention also provides view display equipment based on the federal learning model. As shown in fig. 7, fig. 7 is a schematic structural diagram of a hardware operating environment according to an embodiment of the present invention.
It should be noted that fig. 7 is a schematic structural diagram of a hardware operating environment of a view display device based on the federal learning model. The view display device based on the federal learning model in the embodiment of the invention can be a terminal device such as a PC, a portable computer and the like.
As shown in fig. 7, the view display apparatus based on the federal learning model may include: a processor 1001, such as a CPU, a memory 1005, a user interface 1003, a network interface 1004, a communication bus 1002. Wherein a communication bus 1002 is used to enable connective communication between these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1005 may be a high-speed RAM memory or a non-volatile memory (e.g., a magnetic disk memory). The memory 1005 may alternatively be a storage device separate from the processor 1001.
It will be understood by those skilled in the art that the federal learning model based view display device configuration shown in fig. 7 does not constitute a limitation of the federal learning model based view display device, and may include more or fewer components than those shown, or some components in combination, or a different arrangement of components.
As shown in fig. 7, the memory 1005, which is a kind of computer storage medium, may include therein an operating system, a network communication module, a user interface module, and a view display program based on the federal learning model. The operating system is a program for managing and controlling hardware and software resources of the view display device based on the federal learning model, and supports the operation of the view display program based on the federal learning model and other software or programs.
In the view display device based on the federal learning model shown in fig. 7, the user interface 1003 is mainly used for connecting a terminal device and performing data communication with the terminal device, such as receiving an image to be recognized or an image to be trained sent by the terminal device; the network interface 1004 is mainly used for the background server and performs data communication with the background server; the processor 1001 may be configured to invoke the view display program based on the federal learning model stored in the memory 1005 and execute the steps of the view display method based on the federal learning model as described above.
The specific implementation of the view display device based on the federal learning model of the present invention is basically the same as that of each embodiment of the view display method based on the federal learning model, and is not described herein again.
In addition, an embodiment of the present invention further provides a computer-readable storage medium, where a view display program based on a federal learning model is stored on the computer-readable storage medium, and when the view display program based on the federal learning model is executed by a processor, the steps of the view display method based on the federal learning model as described above are implemented.
The specific implementation of the computer-readable storage medium of the present invention is substantially the same as the above-mentioned embodiments of the view display method based on the federal learning model, and will not be described herein again.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal device (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims (10)

1. A view display method based on a federal learning model is characterized by comprising the following steps:
acquiring operation data of the transverse federated learning model corresponding to each client in the iterative training process of the transverse federated learning model;
constructing a visual view corresponding to the transverse federated learning model according to the operating data, and determining a training process of the transverse federated learning model;
determining content displayed in the visual view according to the training process.
2. The federal learning model based view display method as claimed in claim 1, wherein the operating data includes at least one of: the method comprises the steps that client identification, a starting timestamp of each iterative training, the number of training samples corresponding to the client, a loss value corresponding to a local model, and a starting time point and an ending time point of each client for training the local model through local data are obtained, wherein a visual view comprises an overview view;
the step of constructing a visual view corresponding to the lateral federated learning model based on the operational data includes:
performing visual coding on the operation data to obtain visualized operation data;
and constructing a summary view corresponding to the transverse federated learning model according to the visualized operating data.
3. The method for displaying a view based on a federal learning model as claimed in claim 1, wherein the visual view comprises a projection view, and the step of constructing the visual view corresponding to the horizontal federal learning model according to the operating data comprises:
determining index data corresponding to each client according to the operation data;
and constructing a projection view corresponding to the transverse federated learning model according to the index data, wherein each node in the projection view represents a mapping relation between a client identifier and the iteration times respectively.
4. The federal learning model-based view display method as claimed in claim 1, wherein the visual views include abstract views, and the step of constructing the visual view corresponding to the horizontal federal learning model according to the operating data includes:
determining statistical data corresponding to the operation data, wherein the statistical data at least comprises one of the following data: the number of clients corresponding to the transverse federated learning model, the number of iterations, the number of changes of the number of samples to be trained for training the transverse federated learning model, a reduction value corresponding to a loss value corresponding to the transverse federated learning model, and an increase value of the recognition accuracy of a local model corresponding to each client;
and constructing a summary view corresponding to the transverse federated learning model according to the statistical data.
5. The federal learning model-based view display method as claimed in claim 1, wherein the visual views include comparison views, and the step of constructing the visual view corresponding to the horizontal federal learning model based on the operating data comprises:
determining index data corresponding to each client according to the operation data, and determining a target index corresponding to the last iterative training of the transverse federated learning model in the index data;
and constructing a comparison view corresponding to the transverse federated learning model according to the target indexes.
6. The federated learning model-based view display method of claim 1, wherein the visual view comprises a contribution ranking view, and the step of constructing a visual view corresponding to the lateral federated learning model from the operational data comprises:
determining the ranking sequence of the running data of the client in each iterative training;
and displaying the ranking training in a box and whisker graph mode to construct a contribution degree sequencing view corresponding to the transverse federated learning model.
7. The method for displaying views based on a federated learning model as defined in any one of claims 1 to 6, wherein the step of constructing the visual view corresponding to the horizontal federated learning model according to the operational data and determining the training course of the horizontal federated learning model further comprises:
detecting whether an operation instruction for operating the visual view is received;
if the operation instruction is received, the step of determining the content displayed in the visual view according to the training process includes:
and determining the content displayed in the visual view according to the operation instruction and the training process.
8. A view display apparatus based on a federal learning model, the view display apparatus based on the federal learning model comprising:
the acquisition module is used for acquiring the running data of each client corresponding to the transverse federated learning model in the iterative training process of the transverse federated learning model;
the construction module is used for constructing a visual view corresponding to the transverse federated learning model according to the operating data;
a determining module for determining a training process of the lateral federated learning model; determining content displayed in the visual view according to the training process.
9. A view display device based on a federal learning model, which is characterized by comprising a memory, a processor and a view display program based on the federal learning model stored in the memory and capable of running on the processor, wherein the view display program based on the federal learning model realizes the steps of the view display method based on the federal learning model as claimed in any one of claims 1 to 7 when being executed by the processor.
10. A computer-readable storage medium, wherein a view display program based on a federal learning model is stored in the computer-readable storage medium, and the view display program based on the federal learning model realizes the steps of the view display method based on a federal learning model according to any one of claims 1 to 7 when being executed by a processor.
CN202010370699.6A 2020-04-30 2020-04-30 View display method, device, equipment and medium based on federal learning model Pending CN111553485A (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202010370699.6A CN111553485A (en) 2020-04-30 2020-04-30 View display method, device, equipment and medium based on federal learning model
PCT/CN2021/090983 WO2021219080A1 (en) 2020-04-30 2021-04-29 Federated learning model-based view display method, apparatus and device, and medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010370699.6A CN111553485A (en) 2020-04-30 2020-04-30 View display method, device, equipment and medium based on federal learning model

Publications (1)

Publication Number Publication Date
CN111553485A true CN111553485A (en) 2020-08-18

Family

ID=72002664

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010370699.6A Pending CN111553485A (en) 2020-04-30 2020-04-30 View display method, device, equipment and medium based on federal learning model

Country Status (2)

Country Link
CN (1) CN111553485A (en)
WO (1) WO2021219080A1 (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111950651A (en) * 2020-08-21 2020-11-17 中国科学院计算机网络信息中心 High-dimensional data processing method and device
CN112416887A (en) * 2020-11-18 2021-02-26 脸萌有限公司 Information interaction method and device and electronic equipment
WO2021219080A1 (en) * 2020-04-30 2021-11-04 深圳前海微众银行股份有限公司 Federated learning model-based view display method, apparatus and device, and medium
CN113723619A (en) * 2021-08-31 2021-11-30 南京大学 Federal learning training method based on training phase perception strategy
CN114281231A (en) * 2021-10-12 2022-04-05 腾讯科技（深圳）有限公司 Information presentation method and device, electronic equipment and storage medium

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114189899B (en) * 2021-12-10 2023-03-31 东南大学 User equipment selection method based on random aggregation beam forming
CN115391734B (en) * 2022-10-11 2023-03-10 广州天维信息技术股份有限公司 Client satisfaction analysis system based on federal learning

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120278659A1 (en) * 2011-04-27 2012-11-01 Microsoft Corporation Analyzing Program Execution
CN108268362A (en) * 2018-02-27 2018-07-10 郑州云海信息技术有限公司 A kind of method and device that curve graph is drawn under NVcaffe frames
CN110825476A (en) * 2019-10-31 2020-02-21 深圳前海微众银行股份有限公司 Display method, device, terminal and medium for federal learning workflow interface
CN111078725A (en) * 2019-12-13 2020-04-28 浙江大学 Visual query method and device for multi-data-source combined data

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110153475A1 (en) * 2009-12-18 2011-06-23 At&T Intellectual Property I, L.P. Federal adjustment tracking system
CN110874648A (en) * 2020-01-16 2020-03-10 支付宝(杭州)信息技术有限公司 Federal model training method and system and electronic equipment
CN111553485A (en) * 2020-04-30 2020-08-18 深圳前海微众银行股份有限公司 View display method, device, equipment and medium based on federal learning model

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120278659A1 (en) * 2011-04-27 2012-11-01 Microsoft Corporation Analyzing Program Execution
CN108268362A (en) * 2018-02-27 2018-07-10 郑州云海信息技术有限公司 A kind of method and device that curve graph is drawn under NVcaffe frames
CN110825476A (en) * 2019-10-31 2020-02-21 深圳前海微众银行股份有限公司 Display method, device, terminal and medium for federal learning workflow interface
CN111078725A (en) * 2019-12-13 2020-04-28 浙江大学 Visual query method and device for multi-data-source combined data

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
WEI XIGUANG ET AL.: "Multi-Agent Visualization for Explaining Federated Learning", 《PROCEEDINGS OF THE TWENTY-EIGHTH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE (IJCAI-19)》 *
潘如晟: "联邦学习可视化: 挑战与框架", 《计算机辅助设计与图形学学报》 *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021219080A1 (en) * 2020-04-30 2021-11-04 深圳前海微众银行股份有限公司 Federated learning model-based view display method, apparatus and device, and medium
CN111950651A (en) * 2020-08-21 2020-11-17 中国科学院计算机网络信息中心 High-dimensional data processing method and device
CN111950651B (en) * 2020-08-21 2024-02-09 中国科学院计算机网络信息中心 High-dimensional data processing method and device
CN112416887A (en) * 2020-11-18 2021-02-26 脸萌有限公司 Information interaction method and device and electronic equipment
CN112416887B (en) * 2020-11-18 2024-01-30 脸萌有限公司 Information interaction method and device and electronic equipment
CN113723619A (en) * 2021-08-31 2021-11-30 南京大学 Federal learning training method based on training phase perception strategy
CN114281231A (en) * 2021-10-12 2022-04-05 腾讯科技（深圳）有限公司 Information presentation method and device, electronic equipment and storage medium
CN114281231B (en) * 2021-10-12 2023-10-20 腾讯科技（深圳）有限公司 Information presentation method, device, electronic equipment and storage medium

Also Published As

Publication number Publication date
WO2021219080A1 (en) 2021-11-04

Similar Documents

Publication Publication Date Title
CN111553485A (en) View display method, device, equipment and medium based on federal learning model
US10409367B2 (en) Predictive graph selection
US20150134554A1 (en) Network visualization for employment profiling
US20160350717A1 (en) Group supporting apparatus for recognizing density of discussions and activity levels of individuals and related computer readable recording medium
US20080163158A1 (en) System and method for efficient development of configurable software systems in a large software development community
US20160217427A1 (en) Systems, methods, and devices for implementing a referral processing engine
CN111161804B (en) Query method and system for species genomics database
CN113313170A (en) Full-time global training big data platform based on artificial intelligence
US20150379112A1 (en) Creating an on-line job function ontology
WO2020190176A1 (en) Recommender system for staff recruitment using machine learning with multivariate data dimension reduction and staff recruitment method using machine learning with multivariate data dimension reduction
Collins et al. Human-in-the-loop mixup
CN111026972B (en) Subscription data pushing method, device, equipment and storage medium in Internet of things
US20180173776A1 (en) Mapping 1:Many Relationships for Elements in a Database System
KR20220046380A (en) System for classifying and managing contents asset
CA2869871C (en) System and method for personality-based formatting of information
US20160217425A1 (en) Systems, methods, and devices for implementing a referral processor
US20160217216A1 (en) Systems, methods, and devices for implementing a referral search
CN110262950A (en) Abnormal movement detection method and device based on many index
US20240104804A1 (en) System for clustering data points
US20230126022A1 (en) Automatically determining table locations and table cell types
CN116089490A (en) Data analysis method, device, terminal and storage medium
CN115631374A (en) Control operation method, control detection model training method, device and equipment
CN111291259B (en) Data screening method and device, electronic equipment and storage medium
US9485315B2 (en) System and method for generating a customized singular activity stream
KR20220046379A (en) Method for providing cooperation solution among multiple subjects specialized in video contents production

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20200818

RJ01 Rejection of invention patent application after publication