CN116029371B - Federal learning workflow construction method based on pre-training and related equipment - Google Patents
Federal learning workflow construction method based on pre-training and related equipment Download PDFInfo
- Publication number
- CN116029371B CN116029371B CN202310301957.9A CN202310301957A CN116029371B CN 116029371 B CN116029371 B CN 116029371B CN 202310301957 A CN202310301957 A CN 202310301957A CN 116029371 B CN116029371 B CN 116029371B
- Authority
- CN
- China
- Prior art keywords
- target
- training
- training set
- model
- task
- 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.)
- Active
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Image Analysis (AREA)
Abstract
The application provides a federal learning workflow construction method based on pre-training and related equipment, wherein the method comprises the following steps: acquiring a local data set; determining a target training set according to the local data set, and determining a target calculation task and a pre-training model according to the target training set; initiating participation requests of other participants; responding to the rejection request, and determining a first federal learning model according to the target training set, the target calculation task and the pre-training model; and responding to the receiving request, acquiring an auxiliary training set of the alliance member, and determining a second federation learning model according to the target training set, the auxiliary training set, the target calculation task and the pre-training model. According to the method, the target training set and the pre-training model are applied to federal learning to create the federal learning model, the method is not limited by computing resources, is low in image input limit, is used for accelerating fitting of complex functions, reduces model training time, and achieves aggregation of global models with minimum communication expenditure, so that model performance is improved.
Description
Technical Field
The application relates to the technical field of deep learning models, in particular to a federal learning workflow construction method based on pre-training and related equipment.
Background
In practical federal learning application, because the computing resource is limited, only a single lightweight network structure can be constructed, because the lightweight network structure is single, the image input is limited, and complex functions cannot be fitted, so that the model training time is prolonged, and the model performance is difficult to improve. To improve the performance of the model, a large number of model parameters need to be exchanged between the client and the server to train the model, but because of limited communication, network blocking is easily caused when a large number of model parameters are exchanged, and thus model training fails.
Disclosure of Invention
In view of this, the present application aims to provide a federal learning workflow construction method and related equipment based on pre-training, so as to solve the problems of single lightweight network structure and difficult improvement of model performance in federal learning application.
With the above object in view, a first aspect of the present application provides a federal learning workflow construction method based on pre-training, applied to one of a plurality of participants in the federal learning federation, the method including:
A task creation request is sent, and a local data set is acquired in response to receiving an allowable task creation request associated with the task creation request;
determining a target training set according to the local data set, generating a calculation task creating request according to the target training set, and sending the calculation task creating request and the target training set;
in response to receiving a request to allow creation of a computation associated with the request to create a computation task, determining a pre-training model and a target computation task from the target training set;
sending invitations to other participants in the federal learning federation, and determining a first federal learning model according to the target training set, the target computing task and the pre-training model in response to receiving feedback that the other participants reject the invitations; or alternatively, the first and second heat exchangers may be,
and responding to the feedback of receiving the invitation by the other participants, acquiring an auxiliary training set of the other participants, and determining a second linkage learning model according to the target training set, the auxiliary training set, the target calculation task and the pre-training model.
Further, the determining a target training set according to the local data set includes:
Labeling the local data set to obtain a labeled data set;
numbering the marked data set to obtain a target training set.
Further, the determining a target computing task according to the target training set includes:
selecting a preset calculation mode corresponding to the target training set as a target calculation mode according to the target training set;
and creating a target computing task according to the target computing mode.
Further, the determining a pre-training model according to the target training set includes:
and selecting the pre-training model corresponding to the target training set from a pre-constructed model library according to the target training set.
Further, the determining a first federal learning model according to the target training set, the target computing task, and the pre-training model includes:
and training the pre-training model based on the target training set and the target calculation task to obtain a first federal learning model.
Further, the determining a second linkage learning model according to the target training set, the auxiliary training set, the computing task and the pre-training model includes:
And training the pre-training model based on the target training set, the auxiliary training set and the target computing task to obtain a second linkage learning model.
Further, the method further comprises:
acquiring a verification image, and inputting the verification image into the first federal learning model or the second federal learning model to obtain a prediction result;
and evaluating the prediction result according to a preset evaluation index, and determining an evaluation result.
A second aspect of the present application provides a federal learning workflow construction apparatus based on pre-training, comprising:
the first request module is used for sending a task creation request and acquiring a local data set in response to receiving an allowable task creation request associated with the task creation request;
the first determining module is used for determining a target training set according to the local data set, generating a calculation task creating request according to the target training set, and sending the calculation task creating request and the target training set;
the second determining module is used for determining a target computing task and a pre-training model according to the target training set in response to receiving a request for allowing creation of the computing task associated with the request for creating the computing task;
A third determining module, configured to send an invitation to other participants in the federal learning federation, and determine a first federal learning model according to the target training set, the target computing task, and the pre-training model in response to receiving feedback that the other participants reject the invitation; or alternatively, the first and second heat exchangers may be,
and responding to the feedback of receiving the invitation by the other participants, acquiring an auxiliary training set of the other participants, and determining a second linkage learning model according to the target training set, the auxiliary training set, the target calculation task and the pre-training model.
A third aspect of the present application provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing a method as claimed in any one of the preceding claims when executing the program.
A fourth aspect of the present application provides a non-transitory computer readable storage medium storing computer instructions for causing a computer to perform a method as claimed in any one of the preceding claims.
From the above, it can be seen that, according to the federal learning workflow construction method based on pre-training, the target training set is determined according to the local training set, the target calculation task and the pre-training model are determined according to the target training set, the target training set and the pre-training model are applied to federal learning, the deep learning model is created in federal learning, the deep learning model is not limited by calculation resources and low image input, complex functions can be fitted in an accelerated manner, the model training duration is reduced, when other participants reject invitations in the federal learning alliance, the participants train the pre-training model based on their own target training model, without exchanging a large number of model parameters with the server, train the pre-training model with the minimum communication overhead, the problem of limited communication is solved, other participants are invited to participate in training the pre-training model, a large number of model parameters are not required to be exchanged between a plurality of participants and the server, the participants are trained by adding the auxiliary training set, and the model is jointly trained, and thus the model performance is improved.
Drawings
In order to more clearly illustrate the technical solutions of the present application or related art, the drawings that are required to be used in the description of the embodiments or related art will be briefly described below, and it is apparent that the drawings in the following description are only embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort to those of ordinary skill in the art.
FIG. 1 is a schematic flow chart of a method for constructing a federal learning workflow based on pre-training in an embodiment of the present application;
FIG. 2 is a schematic diagram of a structure of a device for constructing a federal learning workflow based on pre-training according to an embodiment of the present application;
fig. 3 is a schematic diagram of a hardware structure of an electronic device according to an embodiment of the present application.
Detailed Description
For the purposes of making the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail below with reference to the accompanying drawings.
It should be noted that unless otherwise defined, technical or scientific terms used in the embodiments of the present application should be given the ordinary meaning as understood by one of ordinary skill in the art to which the present application belongs. The terms "first," "second," and the like, as used in embodiments of the present application, do not denote any order, quantity, or importance, but rather are used to distinguish one element from another. The word "comprising" or "comprises", and the like, means that elements or items preceding the word are included in the element or item listed after the word and equivalents thereof, but does not exclude other elements or items. The terms "connected" or "connected," and the like, are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "upper", "lower", "left", "right", etc. are used merely to indicate relative positional relationships, which may also be changed when the absolute position of the object to be described is changed.
The federal learning is applied to a distributed network, a branch network comprises client nodes, and a local client can only build a lightweight network structure with a single model due to the limitation of computing resources, and the lightweight network has the following problems: 1. because the lightweight network structure is shallow, full-size images cannot be satisfied as input to the model, either from network width or depth. The local model, when processing full size images, needs to achieve the network-met image input size by cropping the original image. If the target area in the original image is too small, it is difficult for the model to learn the features in the input image after cropping. 2. Because the lightweight network structure cannot fit complex functions, in the training process, even if the training round number is increased, the model is difficult to capture data characteristics. Furthermore, during the course of increasing training rounds, the model tends to repeatedly learn the same features, resulting in a phenomenon in which the model is over-fitted. Because of the limited amount of local data at the client, a large number of model parameters need to be continuously exchanged between the client and the server in order to improve the performance of the model. Some nodes in a network are generally distributed in an environment with limited communication bandwidth, frequent communication not only increases communication overhead, but also easily causes a client to drop due to network congestion, so that model training fails.
When scientific researchers use federal learning to solve practical problems, a solid computer foundation is needed. According to the actual scene, the data set is arranged by using the computer programming language, and a deep learning model is built, so that the complicated process increases the difficulty of federal learning application.
The federal learning is built by multi-party cooperation, so that the training difficulty is increased. Although the federal framework is deployed through the workflow engine to uniformly monitor the request of the client and the federal training process, in a large-scale deep learning task, particularly under the federal framework, how to build a uniform interaction platform to realize that a depth model is built without coding, how to build a distributed task by the client nodes and how to track the task state of the client nodes participating in training by the server are the problems to be solved by the workflow engine at present.
According to the method, a workflow scheme is utilized, a federal learning interaction platform and a distributed engine are constructed, task states are asynchronously updated on the premise of protecting data privacy, a visual interface is provided for researchers, and circulation of tasks of the flow engine is monitored. Further, the method and the device apply the target training set and the pre-training model to federal learning, create a deep learning model in federal learning, are not limited by computing resources, are low in image input limit, accelerate fitting of complex functions, reduce model training time, and realize aggregation of global models with minimum communication overhead, so that model performance is improved. Meanwhile, the automatic modeling of the code-removed federal learning is realized by means of a federal learning interaction platform.
Embodiments of the present application are described in detail below with reference to the accompanying drawings.
Referring to fig. 1, one embodiment of the present application provides a federal learning workflow construction method based on pre-training, applied to one of a plurality of participants in the federal learning federation, the method comprising:
step S100, a task creation request is sent, and a local data set is acquired in response to receiving an allowable task creation request associated with the task creation request.
Specifically, in this embodiment, the participant may be a client, where the task creation request is a task request for building a deep learning model sent to the server by a user through a client, and after the server receives the task request, it is determined whether the client has a right to create a task, and it is determined whether the client has an abnormality, if it is determined that the client has a right to create a task, and the client does not have an abnormality, feedback information allowing creation is sent to the client, and after the client receives the feedback information allowing creation, the client invokes local data stored in the server.
Step S101, determining a target training set according to the local data set, generating a calculation task creating request according to the target training set, and sending the calculation task creating request and the target training set.
Specifically, the client obtains a target training set for training by processing a local data set, trains the model by using the target training set, and is more beneficial to improving the performance of the model, wherein the creating a calculation task request according to the target training set comprises the following steps: judging whether the target training set is a qualified target training set, and creating a calculation task request in response to the target training set being the qualified training set; and the client side sends the request for creating the computing task and the target training set to the server side.
Step S102, in response to receiving the request for allowing creation of the calculation task associated with the request for creation of the calculation task, a pre-training model and a target calculation task are determined according to the target training set.
Specifically, the server judges whether the client has a right to create a computing task according to the received request for creating the computing task, judges whether the client has an abnormality, and when judging that the client has the right to create the computing task and the client does not have the abnormality, sends feedback information allowing the creation of the computing task to the client, after receiving the feedback information allowing the creation of the computing task, the client selects a corresponding pre-training model according to a target training set and feeds back the selected information to the server, after receiving the information of the pre-training model, the server judges whether the type of the target training set corresponds to the type of the pre-training model, if the type of the target training set corresponds to the type of the pre-training model, the client is successfully matched in a feedback manner, the client is allowed to perform the next operation, and the client selects the corresponding computing mode according to the target training set and feeds back the target training set to the server, and the server performs the creation of the task according to the computing mode.
Step S103, sending invitations to other participants in the federal learning federation.
Specifically, steps S100-S102 may be steps that are executed simultaneously by multiple participants in the federal learning federation, where the multiple participants respectively determine their own corresponding target training sets, target computing tasks, and pre-training models, and the multiple participants may send participation requests to other participants to invite the other participants to participate in training, and the other participants may choose to accept the requests or reject the requests, where during model training, the client may not initiate an invitation to the other participants, and train the pre-training models using their own target training sets.
And in response to receiving feedback that the other participants reject the invitation, determining a first federal learning model according to the target training set, the target computing task and the pre-training model.
Specifically, after receiving feedback that other participants reject the invitation, the client does not need to call data of other participants to participate in model training, and directly trains the pre-training model based on the target training set and the target calculation task to obtain a first federal learning model, wherein the first federal learning model is a model trained by only one client.
And responding to the feedback of receiving the invitation by the other participants, acquiring an auxiliary training set of the other participants, and determining a second linkage learning model according to the target training set, the auxiliary training set, the target calculation task and the pre-training model.
Specifically, after receiving feedback that the other participants accept the invitation, the client needs to call a target training set of the client corresponding to the acceptance request, take the target training set as an auxiliary training set, train the pre-training model based on the target training set, the auxiliary training set and the target calculation task, and obtain a second combined learning model, wherein the second combined learning model is a model in which a plurality of clients participate in training.
In step S100-step S103, the target training set and the pre-training model are applied to federal learning, a deep learning model is created in federal learning, the problem of model singleness is solved, the limitation of computing resources is avoided, the limitation of image input is low, complex functions are accelerated to fit, the model training time is shortened, in federal learning alliance, when other participants reject invitations, the participants train the pre-training model based on the own target training model, a large number of model parameters are not required to be exchanged with a server, the pre-training model is trained with the minimum communication expense, the problem of network blockage and model training failure is avoided, other participants are invited to participate in training the pre-training model, a large number of model parameters are not required to be exchanged between a plurality of participants and the server, the problem of limited communication is solved, and the participants participate in model training, so that the model performance is improved.
Illustratively, the above-described clients include, but are not limited to, PCs and portable mobile devices, and the clients execute on a workflow visualization interface, which may be a web browser, upon sending a request. Wherein, in the workflow visualization interface, a plurality of tabs are included, such as my initiation, my participation, my invitation, etc.; wherein, my sponsor includes the task creation record of all sponsors of the client, my participation includes the model training record of all participants of the client, and my invitation includes the sponsor record of the client inviting other participants to participate. The workflow visualization interface also comprises task creation, task introduction and other tabs, wherein the task creation tab comprises a task creation initiating connection, and when a user triggers the task creation connection, a dialog box for filling in the task introduction is popped up on the workflow visualization interface, so that the user is required to fill in the introduction of the task creation for other participants to view.
In some embodiments, in step S101, the determining a target training set according to the local data set includes:
labeling the local data set to obtain a labeled data set;
Numbering the marked data set to obtain a target training set.
Specifically, labeling a local data set to obtain a labeling data set, dividing the labeling data set by a client, dividing the labeling data set according to different labeling types, and selecting a corresponding pre-training model for training according to the divided data set by a user to improve the training performance of the model, wherein the labeling process can label according to a preset task type label, and the task type label comprises: classification, detection, segmentation, etc. The labeling process can also label according to preset personalized labeling labels, and the personalized labeling labels comprise multiple labels, entity relations and the like, and the data set can be automatically labeled by using a labeling tool embedded in a browser during labeling.
Further, if the local data set includes the original labeling information, the original labeling information can be modified, and the original labeling information can be modified according to the task type label or the personalized labeling label.
Further, before labeling the local data set, the method may further include: and preprocessing the local data set to obtain a preprocessed data set.
The pretreatment comprises the following steps: data feature sampling, data feature cleaning, missing value filling, outlier processing, feature normalization and other processing processes.
Specifically, when numbering the marked data sets, the server side can automatically assign the marked data sets to number, or the marked data sets are numbered by the client side in a self-behavior mode, so that a target training set is obtained, and when model training is performed, the corresponding numbered target training set can be selected according to training requirements to perform model training, so that training performance of the model is improved.
In some embodiments, in step S102, the determining a target computing task according to the target training set includes:
selecting a preset calculation mode corresponding to the target training set as a target calculation mode according to the target training set;
and creating a target computing task according to the target computing mode.
Specifically, the preset calculation modes include horizontal federal learning and vertical federal learning, when the data in the target training set has a shared feature space but has different sample conditions, the horizontal federal learning is selected, when the data in the target training set has the same sample space but has different feature spaces, the vertical federal learning is selected, and a target calculation task is created according to the horizontal federal learning or the vertical federal learning, wherein the appropriate calculation mode is selected according to the data distribution conditions in the target training set, so that the training speed of the model is increased, and the performance of the model is improved.
In some embodiments, in step S102, the determining a pre-training model according to the target training set includes:
and selecting the pre-training model corresponding to the target training set from a pre-constructed model library according to the target training set.
Specifically, the Model database is illustratively a Model zoo database, and a plurality of pre-training models are stored in the Model zoo database and are divided into a natural image pre-training Model and a medical image pre-training Model. In the medical image pre-training model, the medical image pre-training model can be divided into a pre-training model based on a nuclear magnetic image according to different medical images, and a pre-training model based on CT; both pre-trained models contained images of all medical conditions. The plurality of pre-training models can be further divided into classified task class pre-training models according to task types, and the task class pre-training models and the segmentation task class pre-training models are detected.
The Model zoo database is trained by selecting a required pre-training Model in the Model zoo database, and the pre-training Model in the database can be updated in real time after each time of training.
The pre-training model is applied to federal learning, so that a deep learning model can be automatically built, the problem of limited image input caused by limited computing resources in federal learning application is solved, meanwhile, the pre-training model is trained based on a target training set, fitting of complex functions is accelerated, model training time is shortened, aggregation of global models is realized with minimum communication overhead, and model performance is improved.
In some embodiments, in step S103, the determining a first federal learning model according to the target training set, the target computing task, and the pre-training model includes:
and training the pre-training model based on the target training set and the target calculation task to obtain a first federal learning model.
Specifically, the first federal learning model is trained based on a task initiator in the federal learning alliance, the task initiator marks or preprocesses a local data set after initiating a task to obtain a target training set, the target training set is input into a pre-training model, the pre-training model is trained based on the target training set according to a target calculation task, a deep learning model trained by the task initiator is obtained after training is completed, and the pre-training model is applied to federal learning, so that the deep learning model can be automatically built.
In some embodiments, in step S103, the determining a second linkage learning model according to the target training set, the auxiliary training set, the computing task, and the pre-training model includes:
and training the pre-training model based on the target training set, the auxiliary training set and the target computing task to obtain a second linkage learning model.
Specifically, the second federation learning model is a model for training together with other participants based on a task initiator in a federation learning alliance, the task initiator invites a plurality of other participants in the alliance to participate in the training of the model, any other participant in the alliance accepts the invitation, a target training set of the other participant is used as an auxiliary training set, the target training set of the task initiator and the auxiliary training set of any other participant are input into a pre-training model, the pre-training model calculates the target training set and the auxiliary training set according to a target calculation task created by a selected target calculation mode, namely, training of the pre-training model is achieved, a deep learning model trained together by the task initiator and any other participant is obtained after the training is completed, and the pre-training model is applied to federation learning, so that the deep learning model can be automatically built.
After any other party in the alliance accepts the invitation, the server side judges whether the training set of the other party and the training set of the task initiator belong to the same type, if so, the training set of the other party is fed back to the client side, and the client side trains the pre-training model by utilizing the training of the target training set and the other party; if the training sets do not belong to the same type, feedback information mismatch prompts to other participants, and refusing the training sets of the other participants to add training.
In some embodiments, the method further comprises:
acquiring a verification image, and inputting the verification image into the first federal learning model or the second federal learning model to obtain a prediction result;
and evaluating the prediction result according to a preset evaluation index, and determining an evaluation result.
Specifically, the verification image is a locally stored image for testing the deep learning model, the verification image is input into the first federal learning model or the second federal learning model to obtain a prediction result, the prediction result is matched with a preset evaluation index to obtain an evaluation result, and the constructed deep learning model is evaluated through the verification image, so that the performance of the deep learning model is verified.
Illustratively, the preset evaluation index includes an image resolution, an image brightness, an image color, and the like. Taking image resolution as an example, the image resolution output by the model is 960×540, the corresponding preset evaluation index is 960×540, and the evaluation result is excellent.
In some embodiments, the federal learning model construction method may also be described by:
the method comprises the steps of combining a workflow scheme to construct a federal learning interaction platform, wherein the federal learning interaction platform comprises a plurality of clients and a server; and constructing a deep learning model based on the federal learning interaction platform.
Clients include, but are not limited to, PCs and portable mobile devices, clients interact with a server based on a web browser that includes a client home page that includes: my initiated, my participated, my invitation tab; and tabs for task creation, task introduction, and the like.
When a user triggers a task creation tab on a first page of a client, a new workflow task is started, at this time, the client sends a task creation request to a server, after the server receives the task request, the server judges whether the client has a right creation task or not and judges whether the client has an abnormality, and when the client has a task creation right and the client does not have an abnormality, the server sends a request for allowing creation to the client. When the client receives the request for allowing creation, the interface pops up an uploading data dialog box, the uploading data dialog box comprises all file list option cards, and the user clicks the file list option cards to select files to be uploaded, so that uploading is completed, and a local data set is obtained. The user carries out preprocessing operation and labeling on the local data set through the client to obtain a labeled data set, classifies and numbers the labeled data set to obtain a target training set, and if the file is not uploaded successfully, the user triggers to continuously upload the tab until the labeling and classification of the data are completed to obtain the target training set; and the target training set is sent to the server, the server judges whether the target training set can be used for training the model or not, the result is fed back to the client, and the user performs the next operation according to the feedback result. The receiving target training set can be used for training a model, the interface pops up a calculation mode selection dialog box, a calculation mode is selected according to the target training set, (wherein the calculation mode selection dialog box further comprises a historical task list, the historical task list stores a selection record of the historical calculation mode for reference by a user), the client sends the calculation mode selected by the user and a calculation task creation request to the server, and after the server receives the calculation task request, the server judges whether the client has rights to create the calculation task and feeds a result back to the client. The client receives a right creation calculation task, the interface pops up a pre-training model selection dialog box, after a user selects a pre-training model, the selected pre-training model is fed back to the server, and the server judges whether the pre-training model is matched with a target training set or not; if the matching feedback is sent to the client, the interface pops up a participation dialog box of the inviting participant, and the user can select to invite or not invite; after the invitation is not selected, the selection result is fed back to the server, the server feeds back the feedback result to the client, the client performs training confirmation according to the feedback result, and the client trains the pre-training model after training confirmation to obtain a first federal learning model. After invitation is selected, a selection result is fed back to a server, the server invokes an auxiliary training set of a participant, judges whether the auxiliary training set is of the same type as a target training set of the client, feeds back the auxiliary training set to the client for training confirmation if the auxiliary training set is of the same type, trains a pre-training Model based on the auxiliary training set and the target training set after the client confirms training to obtain a second federation learning Model, stores the first federation learning Model or the second federation learning Model into a Model zoo database for downloading by the client, completes construction of a deep learning Model, and verifies that the Model is optimal by inputting a verification image into the first federation learning Model or the second learning Model to complete verification. The pre-training model is applied to the combination of federal learning and an integrated workflow interface, so that researchers and business personnel can automatically construct a deep learning model and a full-flow interface is visualized. The method comprises the steps of introducing a pre-training model into distributed computation to automatically execute a machine learning task, introducing federal learning into a workflow, ensuring that distributed nodes output different computation states, and protecting the data privacy of the distributed nodes.
It should be noted that, the method of the embodiments of the present application may be performed by a single device, for example, a computer or a server. The method of the embodiment can also be applied to a distributed scene, and is completed by mutually matching a plurality of devices. In the case of such a distributed scenario, one of the devices may perform only one or more steps of the methods of embodiments of the present application, and the devices may interact with each other to complete the methods.
It should be noted that some embodiments of the present application are described above. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments described above and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
Based on the same inventive concept, the application also provides a federal learning workflow construction device based on pre-training, which corresponds to the method of any embodiment.
Referring to fig. 2, the pre-training-based federal learning workflow construction apparatus includes:
a first request module 200 that sends a task creation request, and that obtains a local data set in response to receiving an allowable task creation request associated with the task creation request;
a first determining module 201, configured to determine a target training set according to the local data set, generate a request for creating a computing task according to the target training set, and send the request for creating the computing task and the target training set;
a second determining module 202, configured to determine, in response to receiving a request for allowing creation of a calculation task associated with the request for creation of a calculation task, a target calculation task and a pre-training model according to the target training set;
a third determining module 203, configured to send an invitation to other participants in the federal learning federation, and determine a first federal learning model according to the target training set, the target computing task, and the pre-training model in response to receiving feedback that the other participants reject the invitation; or alternatively, the first and second heat exchangers may be,
and responding to the feedback of receiving the invitation by the other participants, acquiring an auxiliary training set of the other participants, and determining a second linkage learning model according to the target training set, the auxiliary training set, the target calculation task and the pre-training model.
For convenience of description, the above devices are described as being functionally divided into various modules, respectively. Of course, the functions of each module may be implemented in the same piece or pieces of software and/or hardware when implementing the present application.
The device of the foregoing embodiment is configured to implement the corresponding pre-training-based federal learning workflow construction method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which are not described herein.
Based on the same inventive concept, the application also provides an electronic device corresponding to the method of any embodiment, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the method for constructing the federal learning workflow based on the pre-training according to any embodiment when executing the program.
Fig. 3 shows a more specific hardware architecture of an electronic device according to this embodiment, where the device may include: a processor 1010, a memory 1020, an input/output interface 1030, a communication interface 1040, and a bus 1050. Wherein processor 1010, memory 1020, input/output interface 1030, and communication interface 1040 implement communication connections therebetween within the device via a bus 1050.
The processor 1010 may be implemented by a general-purpose CPU (Central Processing Unit ), microprocessor, application specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, etc. for executing relevant programs to implement the technical solutions provided in the embodiments of the present disclosure.
The Memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory ), static storage device, dynamic storage device, or the like. Memory 1020 may store an operating system and other application programs, and when the embodiments of the present specification are implemented in software or firmware, the associated program code is stored in memory 1020 and executed by processor 1010.
The input/output interface 1030 is used to connect with an input/output module for inputting and outputting information. The input/output module may be configured as a component in a device (not shown) or may be external to the device to provide corresponding functionality. Wherein the input devices may include a keyboard, mouse, touch screen, microphone, various types of sensors, etc., and the output devices may include a display, speaker, vibrator, indicator lights, etc.
It should be noted that although the above-described device only shows processor 1010, memory 1020, input/output interface 1030, communication interface 1040, and bus 1050, in an implementation, the device may include other components necessary to achieve proper operation. Furthermore, it will be understood by those skilled in the art that the above-described apparatus may include only the components necessary to implement the embodiments of the present description, and not all the components shown in the drawings.
The electronic device of the foregoing embodiment is configured to implement the corresponding pre-training-based federal learning workflow construction method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which are not described herein.
Based on the same inventive concept, corresponding to any of the above embodiments of the method, the present application further provides a non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the pre-training based federal learning workflow construction method according to any of the above embodiments.
The computer readable media of the present embodiments, including both permanent and non-permanent, removable and non-removable media, may be used to implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device.
The storage medium of the foregoing embodiments stores computer instructions for causing the computer to execute the pre-training-based federal learning workflow construction method according to any one of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which are not described herein.
Those of ordinary skill in the art will appreciate that: the discussion of any of the embodiments above is merely exemplary and is not intended to suggest that the scope of the application (including the claims) is limited to these examples; the technical features of the above embodiments or in the different embodiments may also be combined within the idea of the present application, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application as described above, which are not provided in detail for the sake of brevity.
Additionally, well-known power/ground connections to Integrated Circuit (IC) chips and other components may or may not be shown within the provided figures, in order to simplify the illustration and discussion, and so as not to obscure the embodiments of the present application. Furthermore, the devices may be shown in block diagram form in order to avoid obscuring the embodiments of the present application, and this also takes into account the fact that specifics with respect to implementation of such block diagram devices are highly dependent upon the platform on which the embodiments of the present application are to be implemented (i.e., such specifics should be well within purview of one skilled in the art). Where specific details (e.g., circuits) are set forth in order to describe example embodiments of the application, it should be apparent to one skilled in the art that embodiments of the application can be practiced without, or with variation of, these specific details. Accordingly, the description is to be regarded as illustrative in nature and not as restrictive.
While the present application has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of those embodiments will be apparent to those skilled in the art in light of the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.
The present embodiments are intended to embrace all such alternatives, modifications and variances which fall within the broad scope of the appended claims. Accordingly, any omissions, modifications, equivalents, improvements and/or the like which are within the spirit and principles of the embodiments are intended to be included within the scope of the present application.
Claims (10)
1. A method of federal learning workflow construction based on pre-training, applied to one of a plurality of participants in a federal learning federation, the method comprising: a task creation request is sent, and a local data set is acquired in response to receiving an allowable task creation request associated with the task creation request;
determining a target training set according to the local data set, generating a calculation task creating request according to the target training set, and sending the calculation task creating request and the target training set;
In response to receiving a request to allow creation of a computation associated with the request to create a computation task, determining a pre-training model and a target computation task from the target training set;
sending invitations to other participants in the federal learning federation, and determining a first federal learning model according to the target training set, the target computing task and the pre-training model in response to receiving feedback that the other participants reject the invitations; or alternatively, the first and second heat exchangers may be,
and responding to the feedback of receiving the invitation by the other participants, acquiring an auxiliary training set of the other participants, and determining a second linkage learning model according to the target training set, the auxiliary training set, the target calculation task and the pre-training model.
2. The method of claim 1, wherein said determining a target training set from said local data set comprises:
labeling the local data set to obtain a labeled data set;
numbering the marked data set to obtain a target training set.
3. The method of claim 1, wherein the determining a target computing task from the target training set comprises:
Selecting a preset calculation mode corresponding to the target training set as a target calculation mode according to the target training set;
and creating a target computing task according to the target computing mode.
4. The method of claim 1, wherein said determining a pre-training model from said target training set comprises:
and selecting the pre-training model corresponding to the target training set from a pre-constructed model library according to the target training set.
5. The method of claim 1, wherein the determining a first federal learning model from the target training set, the target computing task, the pre-training model comprises:
and training the pre-training model based on the target training set and the target calculation task to obtain a first federal learning model.
6. The method of claim 1, wherein the determining a second linkage learning model from the target training set, the auxiliary training set, the computing task, the pre-training model comprises:
and training the pre-training model based on the target training set, the auxiliary training set and the target computing task to obtain a second linkage learning model.
7. The method according to claim 1, wherein the method further comprises:
acquiring a verification image, and inputting the verification image into the first federal learning model or the second federal learning model to obtain a prediction result;
and evaluating the prediction result according to a preset evaluation index, and determining an evaluation result.
8. A federal learning workflow construction apparatus based on pre-training, comprising:
the first request module is used for sending a task creation request and acquiring a local data set in response to receiving an allowable task creation request associated with the task creation request;
the first determining module is used for determining a target training set according to the local data set, generating a calculation task creating request according to the target training set, and sending the calculation task creating request and the target training set;
the second determining module is used for determining a pre-training model and a target computing task according to the target training set in response to receiving a request for allowing creation of the computing task associated with the request for creating the computing task;
a third determining module, configured to send an invitation to other participants in the federal learning federation, and determine a first federal learning model according to the target training set, the target computing task, and the pre-training model in response to receiving feedback that the other participants reject the invitation; or alternatively, the first and second heat exchangers may be,
And responding to the feedback of receiving the invitation by the other participants, acquiring an auxiliary training set of the other participants, and determining a second linkage learning model according to the target training set, the auxiliary training set, the target calculation task and the pre-training model.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any of claims 1 to 7 when the program is executed by the processor.
10. A non-transitory computer readable storage medium storing computer instructions for causing a computer to perform the method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310301957.9A CN116029371B (en) | 2023-03-27 | 2023-03-27 | Federal learning workflow construction method based on pre-training and related equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310301957.9A CN116029371B (en) | 2023-03-27 | 2023-03-27 | Federal learning workflow construction method based on pre-training and related equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116029371A CN116029371A (en) | 2023-04-28 |
| CN116029371B true CN116029371B (en) | 2023-06-06 |
Family
ID=86074344
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310301957.9A Active CN116029371B (en) | 2023-03-27 | 2023-03-27 | Federal learning workflow construction method based on pre-training and related equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116029371B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117010023A (en) * | 2023-09-28 | 2023-11-07 | 国网信息通信产业集团有限公司 | Computing system, method, terminal and storage medium based on federal learning |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN110874649A (en) * | 2020-01-16 | 2020-03-10 | 支付宝(杭州)信息技术有限公司 | State machine-based federal learning method, system, client and electronic equipment |
| CN114091103A (en) * | 2021-11-25 | 2022-02-25 | 支付宝(杭州)信息技术有限公司 | Method for training federated learning model, method for calling federated learning model and federated learning system |
| CN115130683A (en) * | 2022-07-18 | 2022-09-30 | 山东大学 | Asynchronous federal learning method and system based on multi-agent model |
| CN115222061A (en) * | 2022-07-29 | 2022-10-21 | 平安科技（深圳）有限公司 | Federal learning method based on continuous learning and related equipment |
| CN115525921A (en) * | 2022-01-18 | 2022-12-27 | 富算科技(上海)有限公司 | MPC-based federated learning model training and prediction method, system, device and medium |
-
2023
- 2023-03-27 CN CN202310301957.9A patent/CN116029371B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN110874649A (en) * | 2020-01-16 | 2020-03-10 | 支付宝(杭州)信息技术有限公司 | State machine-based federal learning method, system, client and electronic equipment |
| CN114091103A (en) * | 2021-11-25 | 2022-02-25 | 支付宝(杭州)信息技术有限公司 | Method for training federated learning model, method for calling federated learning model and federated learning system |
| CN115525921A (en) * | 2022-01-18 | 2022-12-27 | 富算科技(上海)有限公司 | MPC-based federated learning model training and prediction method, system, device and medium |
| CN115130683A (en) * | 2022-07-18 | 2022-09-30 | 山东大学 | Asynchronous federal learning method and system based on multi-agent model |
| CN115222061A (en) * | 2022-07-29 | 2022-10-21 | 平安科技（深圳）有限公司 | Federal learning method based on continuous learning and related equipment |
Non-Patent Citations (2)
| Title |
|---|
| Age-aware Communication Strategy in Federated Learning with Energy Harvesting Devices;Xin Liu 等;《2021 IEEE/CIC International Conference on Communications in China (ICCC)》;全文 * |
| Energy-Aware Edge Association for Cluster-Based Personalized Federated Learning;Yixuan Li 等;《IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY》;第71卷(第6期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116029371A (en) | 2023-04-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11587344B2 (en) | Scene understanding and generation using neural networks | |
| CN109313490B (en) | Eye gaze tracking using neural networks | |
| CN107194743B (en) | A kind of network surveying questionnaire generation method and device | |
| Yadav et al. | Evalai: Towards better evaluation systems for ai agents | |
| CN111275784B (en) | Method and device for generating image | |
| CN116029371B (en) | Federal learning workflow construction method based on pre-training and related equipment | |
| CN109272543B (en) | Method and apparatus for generating a model | |
| CN110809175A (en) | Video recommendation method and device | |
| Fan et al. | An Interactive platform for low-cost 3D building modeling from VGI data using convolutional neural network | |
| CN110310299B (en) | Method and apparatus for training optical flow network, and method and apparatus for processing image | |
| Wang et al. | CyberGIS-Toward synergistic advancement of cyberinfrastructure and GIScience: A workshop summary | |
| CN108875539A (en) | Expression matching process, device and system and storage medium | |
| CN110286938A (en) | For exporting the method and apparatus for being directed to the evaluation information of user | |
| CN111782068A (en) | Method, device and system for generating mouse track and data processing method | |
| CN111414842A (en) | Video comparison method and device, computer equipment and storage medium | |
| WO2022228335A1 (en) | Input-encoding with federated learning | |
| CN110415318B (en) | Image processing method and device | |
| CN111382689A (en) | Card punching system and method for online learning by using computer | |
| CN107395663A (en) | Data capture method and device | |
| KR102433646B1 (en) | 3d modeling system based on 2d image recognition and method therefor | |
| CN112818241B (en) | Content promotion method and device, computer equipment and storage medium | |
| CN115686300A (en) | User interaction method, device, equipment and storage medium | |
| CN116402366A (en) | Data contribution evaluation method and device based on joint learning | |
| Tourk et al. | The new industrial revolution and industrial upgrading in China: achievements and challenges | |
| WO2020211429A1 (en) | Government affairs service hall system |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |