Streaming federal learning method based on human-like key learning period plasticity

Abstract

The invention discloses a streaming federal learning method, system and device based on human-like key learning period plasticity, and the method comprises the steps: carrying out the online training of each local deep learning model through a client, and calculating a Fischer information matrix trace of a model parameter; the trained local deep learning models and the corresponding Fischer information matrix traces are uploaded; the server carries out asynchronous aggregation on each local deep learning model, and carries out weighted aggregation based on lag attenuation on all Fischer information matrix traces to obtain a federated Fischer information trace of an asynchronous streaming federated model; the method comprises the steps of determining whether an asynchronous streaming federated model is in a key learning period or not on the basis of a federated Fischer information trace, dynamically adjusting training hyper-parameters of the asynchronous streaming federated model on the basis of the federated Fischer information trace, and training the asynchronous streaming federated model. The model is enabled to fully learn diversified category knowledge, and the problem of time distribution over-fitting in streaming data is effectively relieved.

Description

Technical Field

[0001] The present invention relates to the field of streaming federated learning technology, and more specifically to a streaming federated learning method, system, and device based on human-like critical learning period plasticity. Background Art

[0002] Streaming federated learning significantly reduces memory requirements and training latency by performing just-in-time local training and asynchronous weight aggregation on the data stream. However, just-in-time training in streaming federated learning can easily lead to model overfitting of the temporal data distribution, especially when there is a temporal class imbalance in the data stream. This overfitting phenomenon is particularly severe during the model's critical learning period (CLP), as this is when the model is most malleable and absorbs new data most quickly, but is also most susceptible to overfitting.

[0003] To address overfitting in streaming federated learning, existing methods primarily employ strategies such as historical data retrieval and weight regularization. Historical data retrieval mitigates overfitting to specific time distributions by storing data locally during training and combining it with past data. However, these methods impose additional storage requirements, making them difficult to implement on resource-constrained devices. Weight regularization, on the other hand, aggregates generalization updates by discarding updates prone to overfitting, thereby mitigating overfitting without increasing storage overhead. The critical learning period (CLP) is a specific stage in deep neural network (DNN) training that is particularly susceptible to overfitting, especially in the presence of non-independent and identically distributed (non-IID) data. DNNs are particularly prone to overfitting during the early training phase, which can lead to irreversible accuracy degradation. CLP was originally introduced for centralized training and has recently been extended to federated learning (FL). However, existing methods primarily focus on quantifying and leveraging CLP in synchronous FL settings, while asynchronous scenarios such as SFL have not been fully explored. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the first aspect of the present invention proposes a streaming federated learning method based on human-like critical learning period plasticity, comprising: the client performs online training on each local deep learning model based on real-time streaming data, and calculates the Fisher information matrix trace of the current parameters of each local deep learning model; the client uploads the trained local deep learning models and the corresponding Fisher information matrix trace to the server; the server asynchronously aggregates the received local deep learning models to obtain an asynchronous streaming federated model, and performs weighted aggregation based on hysteresis decay on all Fisher information matrix traces to obtain the federated Fisher information trace of the asynchronous streaming federated model; the server determines whether the asynchronous streaming federated model is in a critical learning period based on the federated Fisher information trace, and if it is in the critical learning period, dynamically adjusts the training hyperparameters of the asynchronous streaming federated model based on the federated Fisher information trace, and trains the asynchronous streaming federated model based on the dynamically adjusted hyperparameters.

[0005] Optionally, the method performs weighted aggregation based on hysteresis decay on all Fisher information matrix traces to obtain the federated Fisher information trace of the asynchronous streaming federated model, including: the server assigns the current version index to each asynchronous streaming federated model; the client associates the corresponding version index when uploading each local deep learning model; the server filters the local deep learning models corresponding to the latest version index based on the sliding window, and performs exponential decay penalty on the Fisher information matrix trace of the local deep learning model that exceeds the window to obtain each local deep learning model with a penalty factor; the server calculates the weighted sum of the Fisher information matrix traces of each local deep learning model with the penalty factor to obtain the federated Fisher information trace of the asynchronous streaming federated model.

[0006] Optionally, the dynamic adjustment of the training hyperparameters of the asynchronous streaming federated model based on the federated Fisher information trace includes: during the critical learning period of the asynchronous streaming federated model, the server uses the constructed federated Fisher information trace of the asynchronous streaming federated model to dynamically reduce the learning rate, dynamically reduce the batch size, and dynamically introduce random drop operations.

[0007] Optionally, the expressions for dynamically reducing the learning rate, dynamically reducing the batch size, and dynamically introducing the random dropout operation using the constructed asynchronous streaming federated model's federated Fisher information trace are:

[0008]

[0009] D=D0·[1-α(F G (T)-F G (T-1))]

[0010] Among them, F G(T) represents the Fisher information matrix trace of the asynchronous streaming federation model at time T, F G (T-1) represents the Fisher information matrix trace of the asynchronous streaming federated model at time T-1, η0 represents the initial learning rate, Β0 represents the initial batch size, D0 represents the initial dropout rate, α represents the preset parameter, η represents the reduced learning rate, Β represents the reduced batch size, and D represents the dropout rate after the introduction of random dropout.

[0011] Optionally, before the client performs online training on each local deep learning model based on real-time streaming data, the method further includes: the client collects real-time data streams and processes the collected data streams using a sliding window mechanism.

[0012] Optionally, after training the asynchronous streaming federated model based on the dynamically adjusted hyperparameters, the method further includes: presetting a maximum number of critical learning period rounds. If the current number of training rounds of the asynchronous streaming federated model based on the dynamically adjusted hyperparameters exceeds R, then stopping extending the critical learning period and accelerating the model convergence according to the default hyperparameters.

[0013] Optionally, the method further includes: if in a non-critical learning period, training the asynchronous streaming federated model according to default hyperparameters.

[0014] Optionally, the Fisher information matrix trace is obtained by calculating the trace of the second-order derivative matrix of the local deep learning model parameters.

[0015] To achieve the above-mentioned purpose, the second aspect of the present application also provides a streaming federated learning system based on human-like critical learning period plasticity, including: a client and a server; wherein the client is used to perform online training on each local deep learning model based on real-time streaming data, and calculate the Fisher information matrix trace of the current parameters of each local deep learning model; and upload the trained local deep learning models and the corresponding Fisher information matrix trace to the server; the server is used to asynchronously aggregate the received local deep learning models to obtain an asynchronous streaming federated model, and perform weighted aggregation based on hysteresis attenuation on all Fisher information matrix traces to obtain a federated Fisher information trace of the asynchronous streaming federated model; and, based on the federated Fisher information trace, determine whether the asynchronous streaming federated model is in a critical learning period. If it is in the critical learning period, dynamically adjust the training hyperparameters of the asynchronous streaming federated model based on the federated Fisher information trace, and train the asynchronous streaming federated model based on the dynamically adjusted hyperparameters.

[0016] To achieve the above-mentioned purpose, the third aspect of the present application also provides an electronic device, characterized in that the electronic device includes a processor and a memory, and the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by the processor to implement the streaming federated learning method based on human-like critical learning period plasticity described in the first aspect.

[0017] The embodiment of the present invention provides a streaming federated learning method, system and device based on the plasticity of the human-like critical learning period. Compared with the existing technology, the beneficial effects of the method are as follows: the client performs online training on each local deep learning model based on real-time streaming data, and calculates the Fisher information matrix trace of the current parameters of each local deep learning model; the client uploads the trained local deep learning models and the corresponding Fisher information matrix trace to the server; the server asynchronously aggregates the received local deep learning models to obtain an asynchronous streaming federated model, and performs weighted aggregation based on hysteresis attenuation on all Fisher information matrix traces to obtain the federated Fisher information trace of the asynchronous streaming federated model; the server determines whether the asynchronous streaming federated model is in the critical learning period based on the federated Fisher information trace. If it is in the critical learning period, the server determines whether the asynchronous streaming federated model is in the critical learning period based on the federated Fisher information trace. The Federated Fisher Information Trace dynamically adjusts the training hyperparameters of the asynchronous streaming federated model, and trains the asynchronous streaming federated model based on the dynamically adjusted hyperparameters. This application accurately measures and dynamically extends the critical learning period of the global model, so that the model can fully learn diverse category knowledge in a high plasticity state, effectively alleviate the problem of time distribution overfitting in streaming data, and enhance the model's adaptability to heterogeneous data. Through version index tracking and lag-attenuated weighted aggregation mechanism, it is compatible with asynchronous client scenarios, ensures the accuracy of global model evaluation, and improves the stability and convergence speed of the training process. It can effectively process non-independent and identically distributed (non-IID) data streams and periodic data, and is especially suitable for multimodal streaming scenarios such as real-time sensor data and video streams in edge computing environments, expanding the application scope of federated learning in the field of Internet of Things. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] To more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present invention, and those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0019] Figure 1 A flowchart of a streaming federated learning method based on human-like critical learning period plasticity provided by an embodiment of the present invention;

[0020] Figure 2 A conceptual diagram of adjusting CLP in a streaming federated learning method based on human-like critical learning period plasticity provided by an embodiment of the present invention;

[0021] Figure 3 A schematic diagram of a streaming federated learning method based on human-like critical learning period plasticity to accurately measure CLP under asynchronous federation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0023] This specification provides method operation steps as described in the embodiments or flowcharts, but more or fewer operation steps may be included based on routine or non-creative work. When implemented in an actual system or server product, the methods shown in the embodiments or figures may be executed sequentially or in parallel (for example, in a parallel processor or multi-threaded processing environment).

[0024] refer to Figure 1 , Figure 1 A flowchart of a streaming federated learning method based on human-like critical learning period plasticity is provided in an embodiment of the present invention. The streaming federated learning method based on human-like critical learning period plasticity can be executed by processors on the server and client sides. The streaming federated learning method based on human-like critical learning period plasticity can include:

[0025] S10. The client performs online training on each local deep learning model based on real-time streaming data, and calculates the Fisher information matrix trace of each current local deep learning model parameter.

[0026] In one embodiment of the present application, before step S10, the streaming federated learning method based on the human-like critical learning period plasticity may further include the following process:

[0027] The client collects real-time data streams and processes the collected data streams using a sliding window mechanism.

[0028] Specifically, each client continuously receives real-time streaming data (such as sensor data, video streams, etc.) through IoT devices, uses a sliding window mechanism to process the data stream, and maintains a dynamically updated local training dataset.

[0029] The client performs online streaming data training locally using a local deep learning model (such as LSTM).

[0030] The S20 client uploads the trained local deep learning models and the corresponding Fisher information matrix traces to the server.

[0031] It is worth noting that after each training round, the client calculates the Fisher information matrix trace of the local model parameters: The local deep learning model and Fisher information matrix trace are uploaded to the server and asynchronously aggregated on the server.

[0032] S30. The server asynchronously aggregates the received local deep learning models to obtain an asynchronous streaming federated model, and performs hysteresis-attenuation-based weighted aggregation on all Fisher information matrix traces to obtain a federated Fisher information trace of the asynchronous streaming federated model.

[0033] In an embodiment of the present application, step S30 may include the following specific execution process:

[0034] S301. The server distributes the current version index to each asynchronous streaming federated model.

[0035] S302. When the client uploads each local deep learning model, it associates the corresponding version index.

[0036] S303. The server filters the local deep learning models corresponding to multiple latest version indexes based on the sliding window, performs exponential decay penalty on the Fisher information matrix trace of the local deep learning models beyond the window, and obtains each local deep learning model with a penalty factor.

[0037] That is, the server caches and tracks the version of the global model, and the clients receive the global model with the version index, which enables them to trace back to the corresponding global model when uploading the local model for critical learning period calculation. The critical learning period is then calculated based on the sliding window, and the outdated local Fisher information trace is penalized by exponential decay.

[0038] S304. The server calculates the weighted sum of the Fisher information matrix traces of each local deep learning model with a penalty factor to obtain the federated Fisher information trace of the asynchronous streaming federated model.

[0039] Specifically, for asynchronous client scenarios, each global model on the server is assigned a version index T = 1, 2, ..., which is incremented after each new aggregation. Clients receive global models with version indices, which enables them to trace back to the corresponding global model when uploading local models for CLP calculation. The server can calculate CLP based on a sliding window and penalize outdated local FIM traces through exponential decay. The Fisher information matrix trace of the global model (asynchronous streaming federated model) is estimated as:

[0040]

[0041] Where m represents the window size, I(T) represents the original version index of the local model aggregated in the Tth round, which is a traceable index, and λ represents a hyperparameter used to penalize the staleness of the Fisher information matrix trace of the local deep learning model.

[0042] The Fisher information matrix trace is obtained by calculating the trace of the second-order derivative matrix of the local deep learning model parameters.

[0043] S40. The server determines whether the asynchronous streaming federated model is in a critical learning period based on the Federated Fisher Information Trace. If it is in the critical learning period, the server dynamically adjusts the training hyperparameters of the asynchronous streaming federated model based on the Federated Fisher Information Trace, and trains the asynchronous streaming federated model based on the dynamically adjusted hyperparameters.

[0044] It should be noted that deep models have been proven to have a critical learning period, which can be measured by the trace of Fisher information. Models in the critical learning period have stronger plasticity and a stronger ability to learn knowledge. Figure 2 To adjust the concept graph of CLP, this application applies it to asynchronous streaming federation. While accurately measuring the critical learning period of the global model, it aims to extend its critical learning period as much as possible, allowing the model to learn more category knowledge and improve its generalization.

[0045] Among them, extending the critical learning period during federated learning can enable the model to learn from a wider range of categories in a highly plastic state. This approach is particularly suitable for processing heterogeneous and periodic data streams. Figure 3 , Figure 3 Schematic diagram of accurately measuring CLP under asynchronous federation. This application prolongs the critical learning period by carefully controlling key hyperparameters during local training.

[0046] In an embodiment of the present application, step S40 may include the following specific execution process:

[0047] S401. During the critical learning period of the asynchronous streaming federated model, the server utilizes the constructed federated Fisher information trace of the asynchronous streaming federated model to dynamically reduce the learning rate, dynamically reduce the batch size, and dynamically introduce random dropping operations.

[0048] Specifically, the expressions for dynamically reducing the learning rate, dynamically reducing the batch size, and dynamically introducing random dropout operations using the constructed asynchronous streaming federated model's federated Fisher information trace are:

[0049]

[0050] D=D0·[1-α(F G (T)-F G (T-1))]

[0051] Among them, F G (T) represents the Fisher information matrix trace of the asynchronous streaming federation model at time T, F G (T-1) represents the Fisher information matrix trace of the asynchronous streaming federated model at time T-1, η0 represents the initial learning rate, Β0 represents the initial batch size, D0 represents the initial dropout rate, α represents the preset parameter, η represents the reduced learning rate, Β represents the reduced batch size, and D represents the dropout rate after the introduction of random dropout.

[0052] In an embodiment of the present application, step S40 may further include the following specific execution process:

[0053] If it is in the non-critical learning period, the asynchronous streaming federated model is trained according to the default hyperparameters.

[0054] In one embodiment of the present application, after step S40, the streaming federated learning method based on the human-like critical learning period plasticity may further include the following process:

[0055] The server receives the preset maximum critical learning period rounds. If the current training rounds of the asynchronous streaming federated model based on the dynamically adjusted hyperparameters exceed R, the server stops extending the critical learning period and accelerates model convergence according to the default hyperparameters.

[0056] In the actual implementation process, this application first empirically stipulates a maximum CLP round R. When this round is exceeded, the critical learning period (CLP) will no longer be extended and training will proceed normally. Within R rounds, the CLP is extended by dynamically reducing the learning rate, reducing the batch size, and introducing the dropout operation.

[0057] Based on the above embodiments, the present application further provides a streaming federated learning system based on human-like critical learning period plasticity, including:

[0058] Client and server side.

[0059] Among them, the client is used to perform online training of each local deep learning model based on real-time streaming data, and calculate the Fisher information matrix trace of the current local deep learning model parameters.

[0060] In addition, the trained local deep learning models and the corresponding Fisher information matrix traces are uploaded to the server.

[0061] The server is used to asynchronously aggregate the received local deep learning models to obtain an asynchronous streaming federated model, and perform weighted aggregation based on hysteresis decay on all Fisher information matrix traces to obtain the federated Fisher information trace of the asynchronous streaming federated model.

[0062] Furthermore, based on the Federated Fisher Information Trace, it is determined whether the asynchronous streaming federated model is in a critical learning period. If it is in the critical learning period, the training hyperparameters of the asynchronous streaming federated model are dynamically adjusted based on the Federated Fisher Information Trace, and the asynchronous streaming federated model is trained based on the dynamically adjusted hyperparameters.

[0063] Based on the above embodiments, the present application also provides an electronic device, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the streaming federated learning method based on human-like critical learning period plasticity of any of the previous method embodiments.

[0064] In the above embodiments, they can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented in whole or in part in the form of a computer program product. The computer program product includes multiple computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates multiple available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0065] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0066] Each embodiment in this specification is described in a related manner. Similar portions between the various embodiments can be referenced to each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are described briefly because they are generally similar to the method embodiments. For related portions, reference can be made to the description of the method embodiments.

[0067] The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention are included in the scope of protection of the present invention.

Claims

1. A streaming federated learning method based on human-like critical learning period plasticity, characterized by: include: The client performs online training on each local deep learning model based on real-time streaming data and calculates the Fisher information matrix trace of each local deep learning model parameter. The client uploads the trained local deep learning models and the corresponding Fisher information matrix traces to the server; The server asynchronously aggregates the received local deep learning models to obtain an asynchronous streaming federated model, and performs hysteresis-attenuated weighted aggregation on all Fisher information matrix traces to obtain the federated Fisher information trace of the asynchronous streaming federated model. Based on the Federated Fisher Information Trace, the server determines whether the asynchronous streaming federated model is in a critical learning period. If it is in the critical learning period, the server dynamically adjusts the training hyperparameters of the asynchronous streaming federated model based on the Federated Fisher Information Trace, and trains the asynchronous streaming federated model based on the dynamically adjusted hyperparameters.

2. The streaming federated learning method based on human-like critical learning period plasticity according to claim 1, characterized in that: The weighted aggregation based on hysteresis decay is performed on all Fisher information matrix traces to obtain the federated Fisher information trace of the asynchronous streaming federated model, including: The server assigns the current version index to each asynchronous streaming federated model; When the client uploads each local deep learning model, it associates the corresponding version index; The server filters the local deep learning models corresponding to multiple latest version indexes based on a sliding window, and performs an exponential decay penalty on the Fisher information matrix trace of the local deep learning models that exceed the window to obtain each local deep learning model with a penalty factor. The server calculates the weighted sum of the Fisher information matrix traces of each local deep learning model with a penalty factor to obtain the federated Fisher information trace of the asynchronous streaming federated model.

3. The streaming federated learning method based on human-like critical learning period plasticity according to claim 1, characterized in that: The method of dynamically adjusting the training hyperparameters of the asynchronous streaming federated model based on the Federated Fisher Information Trace includes: During the critical learning period of the asynchronous streaming federated model, the server uses the constructed federated Fisher information trace of the asynchronous streaming federated model to dynamically reduce the learning rate, dynamically reduce the batch size, and dynamically introduce random drop operations.

4. The streaming federated learning method based on human-like critical learning period plasticity according to claim 1, characterized in that: The expressions for dynamically reducing the learning rate, dynamically reducing the batch size, and dynamically introducing the random dropout operation using the Federated Fisher Information Trace of the asynchronous streaming federated model are: D=D0·[1-α(F G (T)-F G (T-1))] Among them, F G (T) represents the Fisher information matrix trace of the asynchronous streaming federation model at time T, F G (T-1) represents the Fisher information matrix trace of the asynchronous streaming federated model at time T-1, η0 represents the initial learning rate, Β0 represents the initial batch size, D0 represents the initial dropout rate, α represents the preset parameter, η represents the reduced learning rate, Β represents the reduced batch size, and D represents the dropout rate after the introduction of random dropout.

5. The streaming federated learning method based on human-like critical learning period plasticity according to claim 1, characterized in that: Before the client performs online training on each local deep learning model based on real-time streaming data, the method further includes: The client collects real-time data streams and processes the collected data streams using a sliding window mechanism.

6. The streaming federated learning method based on human-like critical learning period plasticity according to claim 1, characterized in that: After training the asynchronous streaming federated model based on the dynamically adjusted hyperparameters, the method further includes: The maximum number of critical learning rounds is preset. If the number of training rounds for the asynchronous streaming federated model based on the dynamically adjusted hyperparameters exceeds R, the critical learning period will be stopped from being extended, and the model convergence will be accelerated according to the default hyperparameters.

7. The streaming federated learning method based on human-like critical learning period plasticity according to claim 1, characterized in that: The method further comprises: If it is in the non-critical learning period, the asynchronous streaming federated model is trained according to the default hyperparameters.

8. The streaming federated learning method based on human-like critical learning period plasticity according to claim 1, characterized in that: The Fisher information matrix trace is obtained by calculating the trace of the second-order derivative matrix of the local deep learning model parameters.

9. A streaming federated learning system based on human-like critical learning period plasticity, characterized by: include: Client and server; The client is used to perform online training on each local deep learning model based on real-time streaming data and calculate the Fisher information matrix trace of each local deep learning model parameter. In addition, the trained local deep learning models and the corresponding Fisher information matrix traces are uploaded to the server; The server is used to asynchronously aggregate the received local deep learning models to obtain an asynchronous streaming federated model, and perform hysteresis-attenuation-based weighted aggregation on all Fisher information matrix traces to obtain a federated Fisher information trace of the asynchronous streaming federated model. Furthermore, based on the Federated Fisher Information Trace, it is determined whether the asynchronous streaming federated model is in a critical learning period. If it is in the critical learning period, the training hyperparameters of the asynchronous streaming federated model are dynamically adjusted based on the Federated Fisher Information Trace, and the asynchronous streaming federated model is trained based on the dynamically adjusted hyperparameters.

10. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the streaming federated learning method based on human-like critical learning period plasticity as described in any one of claims 1-8.

Images