CN116540876A - Human body action recognition method based on individual information personalized federal learning - Google Patents
Human body action recognition method based on individual information personalized federal learning Download PDFInfo
- Publication number
- CN116540876A CN116540876A CN202310516751.8A CN202310516751A CN116540876A CN 116540876 A CN116540876 A CN 116540876A CN 202310516751 A CN202310516751 A CN 202310516751A CN 116540876 A CN116540876 A CN 116540876A
- Authority
- CN
- China
- Prior art keywords
- training
- personalized
- global
- client
- local
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000009471 action Effects 0.000 title claims abstract description 41
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000012549 training Methods 0.000 claims abstract description 114
- 239000011159 matrix material Substances 0.000 claims abstract description 7
- 125000003275 alpha amino acid group Chemical group 0.000 claims description 9
- 150000001875 compounds Chemical class 0.000 claims description 9
- 230000002776 aggregation Effects 0.000 claims description 7
- 238000004220 aggregation Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000004891 communication Methods 0.000 abstract description 5
- 238000010223 real-time analysis Methods 0.000 abstract 1
- 230000000694 effects Effects 0.000 description 5
- 238000013459 approach Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000003814 drug Substances 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 208000018737 Parkinson disease Diseases 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000007786 learning performance Effects 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
- G06F3/011—Arrangements for interaction with the human body, e.g. for user immersion in virtual reality
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/62—Protecting access to data via a platform, e.g. using keys or access control rules
- G06F21/6218—Protecting access to data via a platform, e.g. using keys or access control rules to a system of files or objects, e.g. local or distributed file system or database
- G06F21/6245—Protecting personal data, e.g. for financial or medical purposes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Bioethics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Databases & Information Systems (AREA)
- Computer Hardware Design (AREA)
- Computer Security & Cryptography (AREA)
- Human Computer Interaction (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a human body action recognition method based on individual information personalized federal learning, which comprises the following steps: 1, calculating the similarity between clients according to individual information; 2, the server deploys the model to the client; 3, the client performs model training according to the local data set; and 4, the server aggregates the personalized model for each client according to the similarity matrix until the model converges to deploy a final model and performs real-time analysis. The invention can still ensure the classification accuracy under the condition of heterogeneous client data distribution, thereby reducing the communication cost of federal learning.
Description
Technical Field
The invention belongs to the field of sensor data processing and analysis, and particularly relates to a personalized federal learning human body action recognition method based on individual information.
Background
The development of artificial intelligence and mobile technology has greatly facilitated smart medicine, and human motion recognition based on wearable sensors is a classical application of smart medicine. With wearable sensors such as accelerometers, a human motion recognition system can collect a large amount of real-time data and then employ advanced AI models such as deep learning to recognize a person's activities such as standing, walking, etc. Human motion recognition systems based on wearable sensors improve the quality of life of people. However, traditional sensor-based wearable human motion recognition is centralized, i.e., sensor data collected by multiple individuals is transmitted to a cloud or data center for aggregation, training a centralized AI model for the human motion recognition system. This approach has three limitations, firstly, it greatly increases security and privacy concerns because unauthorized access or data leakage may occur during the transmission of data to a remote location. Furthermore, if a hacker breaks a centralized data system storing sensor data, all data may be illegally transmitted to an unauthorized external party and accessed. Second, the transmission of large amounts of wearable sensor data increases network traffic and may create potential bottlenecks that constitute a significant obstacle to network communication. Third, sensor data is highly sensitive in that it records detailed activities of people's daily living and from which many health conditions (e.g., risk of parkinson's disease) can be inferred. Thus, the circulation and intrusion of sensor data is severely limited by regulations.
Federal learning has been proposed to address the challenges described above. Using federal learning, sensor data is collected and saved on personal devices, and a decentralized AI model (referred to as a local model) is trained using each person's data. Federal learning offers the following benefits over the centralized paradigm. First, since personal data is stored locally, the risk of data leakage during transmission is greatly reduced. At the same time, hackers must hack all the decentralized devices to access all the data, which makes this task more difficult than attacking a centralized data system. Second, less communication resources are required to transmit the local model parameters than to transmit the original data, as the size of the parameters is typically fixed and significantly smaller than the original data. Third, federal learning does not involve data circulation or aggregation, and therefore it complies with legal regulations. Furthermore, since the global model summarizes useful information from all clients, federally learning-based models can generally achieve accuracy comparable to AI models trained in traditional centralized settings. The above advantages motivate numerous federal learning innovative approaches to wearable human motion recognition based on sensors. However, one significant challenge faced by these approaches is data heterogeneity, as different patterns of activity for different individuals tend to result in different local data. The problem of heterogeneity prevents federal learning from generalizing between individuals. To solve this problem, existing methods can be classified into three types. The first type uses the idea of FedAvg to weight average all local models to obtain a global model for each client. While these methods alleviate the data heterogeneity problem to some extent, they do not explicitly consider heterogeneity in the design and are therefore of limited effectiveness. The second type of study is particularly focused on heterogeneity issues, aimed at designing a global model that is more general and suited to different individuals. However, a single model may be difficult to adapt to all individuals, limiting its generalization ability. The third type overcomes this limitation by building a personalized global model for each individual, these methods typically rely on similarities between individuals to achieve personalization. However, measuring similarity by comparing the local models from each individual in each round introduces additional computational costs. Furthermore, the parameters of the local model will change during federal learning training, so that the similarity relationship calculated in one round may not hold in the next round. Thus, stability and performance of federal learning may be hindered.
Disclosure of Invention
The invention aims to solve the defects in the prior art, and provides a human body action recognition method based on individual information personalized federal learning, so that the accuracy and the stability of a human body action recognition system can be ensured under the condition of client data heterogeneity by using fewer communication resources, thereby providing more accurate human body action recognition results for users by using fewer privacy data, ensuring the privacy safety of the users and reducing the risk of data privacy leakage.
The invention aims at achieving the aim of the invention and adopts the following technical scheme:
the invention relates to a human body action recognition method based on individual information personalized federal learning, which is characterized by comprising the following steps:
step 1. Suppose that the kth client has own individual information d k And use d k,i Representing individual information d k Information attribute of the ith dimension;
let the motion data of the local sensor preprocessed by the kth client be ζ k And xi k ＝x k,1 ,y k,1 ),(x k,2 ,y k,2 ),…(x k,s ,y k,s ),…(x k,S ,y k,S ) The method comprises the steps of carrying out a first treatment on the surface of the Wherein x is k,s And y k,s Respectively representing the action samples of the S-th local sensor after pretreatment of the kth client and the corresponding categories thereof, wherein k=1, 2 …, K and K are the total number of clients, s=1, 2 …, S and S are the total number of the action samples of the local sensor of the kth client;
for individual information d k After normalization, the kth client and the kth client are calculated by using a cosine similarity formulaIndividual information similarity between individual clients +.>Thereby obtaining a similarity matrix S, if +.>Representing that the two clients are completely similar, +.>Representing two clients orthogonal;
step 2, defining a global training round as t, and initializing t=1; defining a global maximum training round as Tmax;
in the process of the t-th global training, the server deploys personalized models of the t-th global training to K clients respectively;
step 3, each client updates and trains the personalized model of the t-th global training according to the motion data of the client in a local training stage to obtain the updated personalized model of the t-th global training;
step 4, the server performs personalized aggregation on the personalized models updated on the K clients under the t-th global training according to the similarity matrix S to obtain K personalized models aggregated under the t-th global training and serve as personalized models of the t+1th global training, and the K personalized models are deployed on the K clients respectively;
and 5, after assigning t+1 to t, returning to the step 3 for sequential execution until t > Tmax, wherein the K clients are completely updated, trained and aggregated locally, so that a personalized model of Tmax+1 global training is obtained and used for identifying human body actions of action data acquired by a local sensor in real time.
The human motion recognition method of the present invention is also characterized in that the step 2 includes:
definition of the definitionInitial parameters representing personalized model on kth client at t-th global training,/->Parameters of the personalized model aggregated on the kth client during the t-th global training are represented;
the server willAssign to->And taking the personalized model aggregated in the t-th global training as the personalized model on the kth client and deploying the personalized model on the kth client.
The step 3 comprises the following steps:
step 3.1. The kth client side is according to the action data ζ k Calculating a motion dataset sampled on a kth client at a kth global training time using equation (1)
In the formula (1), random represents a Random value k Random decimation r 2 ×|ξ k Action data of the number r 2 Representing the sampling ratio;
step 3.2, defining a local training round as e, and initializing e=1; defining a local maximum training round as E;
step 3.3. Calculating the gradient of the kth client in the ith local training under the jth global training by using the method (2)
In equation (2), SGD represents a random gradient descent optimizer,representing loss of personalized model on kth client at the time of the e-th local training under the t-th global training,/->Parameters representing the personalized model on the kth client during the ith local training under the jth global training;
step 3.4. Calculating parameters of the personalized model on the kth client during the (e+1) th local training under the (t) th global training by using the (4)
In the formula (3), the amino acid sequence of the compound,the learning rate of the kth client in the ith local training under the nth global training is represented;
step 3.5. After assigning e+1 to e, return to step 3.3 for sequential execution until e>E, obtaining parameters of the personalized model on the kth client during the E-th local training under the t-th global trainingAnd uploading the parameters of the personalized model updated under the t-th global training to the server.
The server in the step 4 calculates the parameters of the aggregated personalized model on the kth client during the t+1st global training by using the formula (4)And is used as the initial parameter of the personalized model of the t+1st global training;
in the formula (4), the amino acid sequence of the compound,representing +.o. for the E th local training under the t th global training>Parameters of the personalized model on the individual clients.
The server in the step 5 obtains the parameters of the personalized model of the kth client in the Tmax+1st global training by using the formula (5)
In the formula (5), the amino acid sequence of the compound,represents the +.sup.th in the E-th local training under the Tmax global training>Parameters of the personalized model on the individual clients.
The electronic device of the invention comprises a memory and a processor, wherein the memory is used for storing a program for supporting the processor to execute any human action recognition method, and the processor is configured to execute the program stored in the memory.
The invention relates to a computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, and the computer program is executed by a processor to execute any step of the human body action recognition method.
Compared with the existing federal learning human body action recognition scheme, the invention has the beneficial effects that:
1. the present invention utilizes individual information such as: age, gender, height and weight, similarity between individuals is calculated as a substitute for data similarity, and weights are determined by the similarity when personalizing the model for the client. In this way, local models from individuals having higher similarity values will be more conducive to personalizing the model of the target individual than those less similar. Meanwhile, the similarity is calculated by using the individual information, so that the generalization capability of the model is enhanced, the calculation resources are reduced, the training stability is improved, and the development of personalized federal learning is facilitated. By utilizing personal information which some users are willing to disclose to realize human action recognition tasks based on federal learning, the risk of revealing personal privacy by utilizing private data such as medical records is avoided.
2. The invention realizes higher performance on all indexes of the human body action recognition task of the wearable sensor. Compared with the common method for recognizing human body actions by using federal learning, the method considers the influence of individual information on the human body action mode, such as age influence on standing time, walking speed and the like, thereby improving the accuracy of human body action recognition tasks, and indicating that the recognition effect of the method on the task exceeds that of the prior other methods.
Drawings
FIG. 1 is a schematic overall flow diagram of the method of the present invention;
fig. 2 is a block diagram of the overall algorithm of the present invention.
Detailed Description
On the basis of performing human body action recognition tasks by using federal learning, the method considers the individual information similarity of the client to perform personalized model aggregation, so that communication resources are reduced, and the human body action recognition tasks based on federal learning achieve better classification effects on the aspect of more realistic data heterogeneity. In order to better understand the above technical solutions, the following detailed description will refer to the accompanying drawings and specific embodiments.
In this embodiment, as shown in fig. 1 and 2, a personalized federal learning human motion recognition method based on individual information is performed according to the following steps:
step 1. Suppose that the kth client has own individual information d k And use d k,i Representing individual information d k Information attribute of the ith dimension;
let the motion data of the local sensor preprocessed by the kth client be xi k And xi k ＝x k,1 ,y k,1 ),(x k,2 ,y k,2 ),…(x k,s ,y k,s ),…(x k,S ,y k,S ) The method comprises the steps of carrying out a first treatment on the surface of the Wherein x is k,s And y k,s Respectively representing the action samples of the S-th local sensor after pretreatment of the kth client and the corresponding categories thereof, wherein k=1, 2 …, K and K are the total number of clients, s=1, 2 …, S and S are the total number of the action samples of the local sensor of the kth client;
for individual information d k After normalization, the kth client and the kth client are calculated by using a cosine similarity formulaIndividual information similarity between individual clients +.>Thereby obtaining a similarity matrix S, if +.>Representing that the two clients are completely similar, +.>Representing two clients orthogonal;
standardized individual information: taking a human motion recognition data set RealWorld as an example, the RealWorld data set contains motion data from 15 subjects and personal information thereof, each subject is taken as a client, and the personal information mainly comprises gender, age, height and attribute, which are respectively expressed as d k,g ，d k,a ，d k,w ，d k,h Since different attributes have different scales and ranges, respectively, they are standardized:
wherein n is k,i Representing the normalized individual information attribute values.
Calculating the similarity relation between individuals by using cosine similarity: individual k and individualAfter the information vector of the formula (1) is normalized, individual k and individual +.>Similarity->In particular implementations, individual k and individual +.>Calculate similarity +.>
Wherein n is the L2 norm of the vector n,where 1 represents perfect similarity and 0 represents orthogonality.
Step 2, defining a global training round as t, and initializing t=1; defining a global maximum training round as Tmax;
in the process of the t-th global training, the server deploys personalized models of the t-th global training to K clients respectively;
definition of the definitionInitial parameters representing personalized model on kth client at t-th global training,/->Parameters of the personalized model aggregated on the kth client during the t-th global training are represented;
the server willAssign to->And taking the personalized model aggregated in the t-th global training as the personalized model on the kth client and deploying the personalized model on the kth client. In specific implementation, the global model may select a deep learning model such as CNN or RNN to perform an action recognition task.
Step 3, each client updates and trains the personalized model of the t-th global training according to the motion data of the client in a local training stage to obtain the updated personalized model of the t-th global training;
step 3.1. The kth client is according toAction data xi k Calculating a motion dataset sampled on a kth client at a kth global training time using equation (3)
In the formula (3), random represents a Random value k Random decimation r 2 ×|ξ k Action data of the number r 2 Representing the sampling ratio;
step 3.2, defining a local training round as e, and initializing e=1; defining a local maximum training round as E;
step 3.3. Calculating the gradient of the kth client in the ith local training under the jth global training by using the method (4)
In equation (4), SGD represents a random gradient descent optimizer,representing loss of personalized model on kth client at the time of the e-th local training under the t-th global training,/->Parameters representing the personalized model on the kth client during the ith local training under the jth global training; in a specific implementation, using an optimizer in a python extension package, the optimizer may select SGD (·) or Adam (·), the loss function uses a multi-class cross entropy loss function, i.e., the local iteration number E may set different parameters according to learning performance of the model, in this case, if CNN is selected as a global modelE may be set to 2; if RNN is selected as the global model, E can be set to 3.
Step 3.4. Calculating parameters of the personalized model on the kth client during the (e+1) th local training under the (t) th global training by using the (4)
In the formula (5), the amino acid sequence of the compound,the learning rate of the kth client in the ith local training under the nth global training is represented;
step 3.5. After assigning e+1 to e, return to step 3.3 for sequential execution until e>E, obtaining parameters of the personalized model on the kth client during the E-th local training under the t-th global trainingAnd uploading the parameters of the personalized model updated under the t-th global training to a server.
Step 4, the server performs personalized aggregation on the personalized models updated on the K clients under the t-th global training according to the similarity matrix S to obtain K personalized models aggregated under the t-th global training and serve as personalized models of the t+1th global training, and the K personalized models are deployed on the K clients respectively;
the server calculates the parameters of the aggregated personalized model on the kth client during the (t+1) th global training by using the formula (6)And is used as the initial parameter of the personalized model of the t+1st global training;
in the formula (6), the amino acid sequence of the compound,representing +.o. for the E th local training under the t th global training>Parameters of the personalized model on the individual clients.
Thus, the local model from the individual with higher similarity contributes more to the client k obtaining the personalized model than the individual with lower similarity. The method emphasizes information from highly similar individuals while utilizing information from other individuals, and is therefore an effective personalized federal learning method. Because for human motion recognition characters, the distribution of the motion data of different clients tends to be unbalanced, the model learned by the target client is more suitable for the motion data of the target client, in specific implementation, the weight of the target client is set to be 0.5, and the sum of the weights calculated by other individual clients according to the similarity is set to be 0.5. And when the information from the highly similar individuals is utilized for personalized aggregation, the proportion occupied by the parameter information is limited, the model parameter information trained by the motion data is focused, and the heterogeneous data migration of the local model to other clients is avoided.
And 5, after assigning t+1 to t, returning to the step 3 for sequential execution until t > Tmax, wherein the K clients are completely updated, trained and aggregated locally, so that a personalized model of Tmax+1 global training is obtained and used for identifying human body actions of action data acquired by a local sensor in real time. In practice, tmax may be 100 rounds.
In the formula (7), the amino acid sequence of the compound,represent the firstTmax time global training E th local training +.>Parameters of the personalized model on the individual clients.
In this embodiment, an electronic device includes a memory for storing a program supporting the processor to execute the above method, and a processor configured to execute the program stored in the memory.
In this embodiment, a computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the method described above.
Claims (7)
1. The human body action recognition method based on individual information personalized federal learning is characterized by comprising the following steps of:
step 1. Suppose that the kth client has own individual information d k And use d k，i Representing individual information d k Information attribute of the ith dimension;
let the motion data of the local sensor preprocessed by the kth client be ζ k And xi k ＝(x k，1 ，y k，1 )，(x k，2 ，y k，2 )，...(x k，s ，y k，s )，...(x k，S ，y k，S ) The method comprises the steps of carrying out a first treatment on the surface of the Wherein x is k，s And y k，s Respectively representing the action samples of the S-th local sensor after pretreatment of the kth client and the corresponding categories thereof, wherein k=1, 2 …, K and K are the total number of clients, s=1, 2 …, S and S are the total number of the action samples of the local sensor of the kth client;
for individual information d k After normalization, the kth client and the kth client are calculated by using a cosine similarity formulaIndividual information similarity between individual clients +.>Thereby obtaining a similarity matrix S, if +.>Representing that the two clients are completely similar, +.>Representing two clients orthogonal;
step 2, defining a global training round as t, and initializing t=1; defining a global maximum training round as Tmax;
in the process of the t-th global training, the server deploys personalized models of the t-th global training to K clients respectively;
step 3, each client updates and trains the personalized model of the t-th global training according to the motion data of the client in a local training stage to obtain the updated personalized model of the t-th global training;
step 4, the server performs personalized aggregation on the personalized models updated on the K clients under the t-th global training according to the similarity matrix S to obtain K personalized models aggregated under the t-th global training and serve as personalized models of the t+1th global training, and the K personalized models are deployed on the K clients respectively;
and 5, after assigning t+1 to t, returning to the step 3 for sequential execution until t is more than Tmax, wherein the K clients are completely updated, trained and aggregated locally, so that a personalized model of Tmax+1 global training is obtained and used for identifying human body actions of action data acquired by a local sensor in real time.
2. The human action recognition method according to claim 1, wherein the step 2 comprises:
definition of the definitionRepresentation ofInitial parameters of personalized model on kth client at t-th global training, +.>Parameters of the personalized model aggregated on the kth client during the t-th global training are represented;
the server willAssign to->And taking the personalized model aggregated in the t-th global training as the personalized model on the kth client and deploying the personalized model on the kth client.
3. The human action recognition method according to claim 2, wherein the step 3 comprises:
step 3.1. The kth client side is according to the action data ζ k Calculating a motion dataset sampled on a kth client at a kth global training time using equation (1)
In the formula (1), random represents a Random value k Random decimation r 2 ×|ξ k Action data of the number r 2 Representing the sampling ratio;
step 3.2, defining a local training round as e, and initializing e=1; defining a local maximum training round as E;
step 3.3. Calculating the gradient of the kth client in the ith local training under the jth global training by using the method (2)
In equation (2), SGD represents a random gradient descent optimizer,representing loss of personalized model on kth client at the time of the e-th local training under the t-th global training,/->Parameters representing the personalized model on the kth client during the ith local training under the jth global training;
step 3.4. Calculating parameters of the personalized model on the kth client during the (e+1) th local training under the (t) th global training by using the (4)
In the formula (3), the amino acid sequence of the compound,the learning rate of the kth client in the ith local training under the nth global training is represented;
step 3.5. After e+1 is assigned to E, returning to step 3.3 for sequential execution until E > E, thereby obtaining parameters of the personalized model on the kth client during the E-th local training under the t-th global trainingAnd uploading the parameters of the personalized model updated under the t-th global training to the server.
4. The human motion recognition method according to claim 3, wherein the server in the step 4 calculates parameters of the aggregated personalized model on the kth client at the time of the (t+1) th global training by using the formula (4)And is used as the initial parameter of the personalized model of the t+1st global training;
in the formula (4), the amino acid sequence of the compound,representing +.o. for the E th local training under the t th global training>Parameters of the personalized model on the individual clients.
5. The human motion recognition method according to claim 4, wherein the server in step 5 obtains parameters of the personalized model of the kth client in the time of the Tmax+1th global training by using equation (5)
In the formula (5), the amino acid sequence of the compound,represents the +.sup.th in the E-th local training under the Tmax global training>Parameters of the personalized model on the individual clients.
6. An electronic device comprising a memory and a processor, wherein the memory is configured to store a program that supports the processor to perform the human action recognition method of any one of claims 1-5, the processor being configured to execute the program stored in the memory.
7. A computer readable storage medium having stored thereon a computer program, characterized in that the computer program when executed by a processor performs the steps of the human action recognition method according to any of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310516751.8A CN116540876A (en) | 2023-05-09 | 2023-05-09 | Human body action recognition method based on individual information personalized federal learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310516751.8A CN116540876A (en) | 2023-05-09 | 2023-05-09 | Human body action recognition method based on individual information personalized federal learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116540876A true CN116540876A (en) | 2023-08-04 |
Family
ID=87450141
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310516751.8A Pending CN116540876A (en) | 2023-05-09 | 2023-05-09 | Human body action recognition method based on individual information personalized federal learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116540876A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117077817A (en) * | 2023-10-13 | 2023-11-17 | 之江实验室 | Personalized federal learning model training method and device based on label distribution |
| CN117275098A (en) * | 2023-11-13 | 2023-12-22 | 南京栢拓视觉科技有限公司 | Federal increment method oriented to action recognition and based on topology data analysis |
-
2023
- 2023-05-09 CN CN202310516751.8A patent/CN116540876A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117077817A (en) * | 2023-10-13 | 2023-11-17 | 之江实验室 | Personalized federal learning model training method and device based on label distribution |
| CN117077817B (en) * | 2023-10-13 | 2024-01-30 | 之江实验室 | Personalized federal learning model training method and device based on label distribution |
| CN117275098A (en) * | 2023-11-13 | 2023-12-22 | 南京栢拓视觉科技有限公司 | Federal increment method oriented to action recognition and based on topology data analysis |
| CN117275098B (en) * | 2023-11-13 | 2024-02-27 | 南京栢拓视觉科技有限公司 | Federal increment method oriented to action recognition and based on topology data analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Al-Waisy et al. | A multi-biometric iris recognition system based on a deep learning approach | |
| US10592783B2 (en) | Risky transaction identification method and apparatus | |
| CN116540876A (en) | Human body action recognition method based on individual information personalized federal learning | |
| WO2019228317A1 (en) | Face recognition method and device, and computer readable medium | |
| JP2018535492A (en) | License plate classification method, system, electronic apparatus and storage medium based on deep learning | |
| US11790222B2 (en) | Methods for learning the parameters of a convolutional neural network, and for classifying an input datum | |
| CN108229532B (en) | Image recognition method and device and electronic equipment | |
| CN112651511A (en) | Model training method, data processing method and device | |
| WO2019232772A1 (en) | Systems and methods for content identification | |
| CN114387486A (en) | Image classification method and device based on continuous learning | |
| CN112418059B (en) | Emotion recognition method and device, computer equipment and storage medium | |
| WO2024001806A1 (en) | Data valuation method based on federated learning and related device therefor | |
| WO2021127982A1 (en) | Speech emotion recognition method, smart device, and computer-readable storage medium | |
| CN113553582A (en) | Malicious attack detection method and device and electronic equipment | |
| US11537750B2 (en) | Image access management device, image access management method, and image access management system | |
| CN110489659A (en) | Data matching method and device | |
| WO2021036397A1 (en) | Method and apparatus for generating target neural network model | |
| CN116664930A (en) | Personalized federal learning image classification method and system based on self-supervision contrast learning | |
| CN114299304B (en) | Image processing method and related equipment | |
| US20160125297A1 (en) | System and method for solving spatiotemporal-based problems | |
| Shen et al. | A classifier based on multiple feature extraction blocks for gait authentication using smartphone sensors | |
| Walse et al. | A study on the effect of adaptive boosting on performance of classifiers for human activity recognition | |
| WO2023185541A1 (en) | Model training method and related device | |
| Xu et al. | An efficient and lightweight method for human ear recognition based on MobileNet | |
| TWI742312B (en) | Machine learning system, machine learning method and non-transitory computer readable medium for operating the same |
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 |