CN114330446A - Arrhythmia detection algorithm and system based on curriculum meta-learning - Google Patents
Arrhythmia detection algorithm and system based on curriculum meta-learning Download PDFInfo
- Publication number
- CN114330446A CN114330446A CN202111667670.5A CN202111667670A CN114330446A CN 114330446 A CN114330446 A CN 114330446A CN 202111667670 A CN202111667670 A CN 202111667670A CN 114330446 A CN114330446 A CN 114330446A
- Authority
- CN
- China
- Prior art keywords
- meta
- training
- sample
- learning
- model
- 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.)
- Granted
Links
Images
Landscapes
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention relates to the field of intelligent medical treatment, and discloses an arrhythmia detection algorithm and system based on curriculum meta-learning, wherein the algorithm comprises the following steps: s1, pre-training the existing neural network model in a pre-training data set to obtain a pre-training model; s2, transferring the pre-training model obtained in the S1 to a new target set for fine adjustment; according to the method, the original pre-training model is trained through the combination of the meta-learning method and the course learning method, and then the model is moved to a new target set for fine adjustment, so that parameters which can be rapidly adapted to different individuals can be obtained, and the problem of large data difference of the same individual in different periods can be solved.
Description
Technical Field
The invention relates to the field of intelligent medical treatment, in particular to an arrhythmia detection algorithm and system based on curriculum meta-learning.
Background
At present, the deep learning method is widely applied to the classification task of electrocardiosignals. Although deep learning models can perform well, this approach has two major limitations:
(1) the electrocardiographic signals vary from person to person. In other words, it is uncertain whether a model learned from one person can be promoted to another person. This is mainly due to differences in some conditions (such as channel, experimental environment and type of subject), which limits the applicability of machine learning models in clinical environments.
(2) In addition to generalization ability between different individuals, deep learning models may perform poorly in one period and well in other periods of the same individual. People do different activities at different stages, which undoubtedly affect different cardiac activity indicators including heart rate. Thus, models trained by data for certain periods of time may perform poorly in other phases.
The existing deep learning method has the above-mentioned limitations.
Disclosure of Invention
In order to solve the limitation problem of the existing deep learning method, the invention provides an arrhythmia detection algorithm and system based on curriculum meta-learning, and the specific scheme is as follows:
an arrhythmia detection algorithm based on curriculum meta-learning, comprising the following steps:
s1, pre-training an existing neural network model in a pre-training data set to obtain a pre-training model, wherein in the pre-training data set, arrhythmia detection is performed on each individual to obtain a meta-task Tb, the meta-task corresponds to a plurality of samples, and one sample is rhythm detection data; the pre-training data set comprises a pre-training set and a pre-training verification set;
s2, transferring the pre-training model obtained in the S1 to a new target set for fine adjustment, wherein each subset in the target set is an individual, each subset corresponds to a plurality of samples, and one sample is heart rhythm detection data.
Preferably, the pre-training in step S1 includes a meta learning method and a course learning method; the meta-learning method is used for learning parameters which can be fast adapted to new individuals for the pre-training model, and the course learning method is used for providing an easy-to-difficult learning sequence for the meta-learning method.
Preferably, the pre-training comprises the following specific steps:
s1.1, calculating cross entropy loss of a sample, and determining a difficulty value of each meta task; the difficulty value is used for measuring the complexity of each meta-task or the difficulty degree of the model for learning the meta-task, and as long as the model is determined, a fixed value can be set for each meta-task to serve as the 'difficulty value' of the meta-tasks;
s1.2, selecting a plurality of meta-tasks for training at each step of the pre-training;
s1.3, initial weight theta of pre-training model0Updating to obtain an updated meta-learning model parameter theta'; the pre-training model obtained after training by the meta-learning method and the course learning method is called a meta-learning model.
Preferably, the specific steps in step S1.1 of calculating said difficulty value are as follows:
s1.1.1, training by using K samples in each meta-task, and testing by using the rest samples in each meta-task;
s1.1.2, calculating the average cross entropy loss lossb of all the samples corresponding to each metatask,
wherein N is the number of samples, M is the number of types of arrhythmia, and when the type of the sample i is c, the real label yicIs 1, otherwise is 0, picThe probability that the sample i given for the meta-learning model is the class c, the difficulty value of the meta-task Tb is
Preferably, the basis for selecting the meta-tasks in step S1.2 is to set a threshold value for the difficulty value of each of the meta-tasks, the meta-tasks having difficulty values smaller than the threshold value are selected with equal probability, and the meta-tasks having difficulty values larger than the threshold value are not selected.
Preferably, the initial weights θ of the model are pre-trained in step S1.30The specific steps of updating include:
s1.3.1, randomly dividing the set of samples in the pre-training data set into a sample support set and a sample query set;
s1.3.2, if the pre-training model parameter is theta at this time0Reproduction of thetab＝θ0Calculating the parameter thetabCross-entropy loss over the sample query set:
where M is the number of classes of arrhythmia and when the class of sample i is c, the true label yicIs 1, otherwise is 0, picThe probability that a sample i given for the meta-learning model is of the class c;
s1.3.3, performing a further gradient descent on the sample support set
Where a is the learning rate, where a is,
is given a parameter of thetabThe cross-entropy loss over the set of queries,
is a cross entropy loss pair thetabA gradient of (a);
s1.3.4, query cross entropy loss on set with the sample, pair theta0Updating is carried out;
s1.3.5, calculating parameter as theta0If the cross entropy loss of the pre-training verification set is not reduced for a plurality of continuous times, stopping updating to obtain an updated meta-learning model parameter theta'.
Preferably, the fine adjustment in step S2 includes the steps of:
s2.1, dividing the sample set of the target data set into a target sample training set, a target sample verification set and a target sample test set;
s2.2, to theta0Carrying out transition; wherein the transition method comprises the following steps:
s2.2.1, parameter to be copied theta0' performing a further gradient descent on the target sample training set T:
s2.2.2, calculating cross entropy loss on the target sample training set;
s2.2.3, using calculated cross entropy loss pairs theta0Updating is carried out;
s2.2.4, repeating the steps of 2.2.1-2.2.3 for S times;
s2.2.5, completing the transition.
And S2.3, training on the target sample training set by using a common random gradient descent method.
And S2.4, sequentially trying each hyper-parameter combination according to the hyper-parameters including the learning rate alpha, the learning rate beta and the S, judging whether the inherent classification label on the current sample is correct or not according to the probability of the class of the current sample given in the trained meta-learning model, calculating the classification accuracy of the current sample on a target sample verification set, wherein the classification accuracy on the target sample verification set is the number of classified correct samples/the total number of samples, and then selecting a group of hyper-parameters with the highest accuracy.
S.2.5, calculating performance indexes of the model on the target sample test set, such as area under the ROC (receiver Operating characterization) curve (AUC), accuracy, average Accuracy (AP), accuracy, recall, specificity and F1-score.
The invention also discloses a computer readable storage medium, which stores a computer program, and executes the arrhythmia detection algorithm after the computer program runs.
The invention also discloses a computer system, which comprises a processor and a storage medium, wherein the storage medium is stored with a computer program, and the processor reads the computer program from the storage medium and runs the computer program to execute the arrhythmia detection algorithm.
Preferably, the system for arrhythmia detection algorithm based on curriculum meta-learning comprises a data processing module, and a heart rate detection module and a data storage module which are electrically connected with the data processing module in sequence.
The invention has the beneficial effects that:
according to the method, the original pre-training model is trained through the combination of the meta-learning method and the course learning method, and then the model is moved to a new target set for fine adjustment, so that parameters which can be rapidly adapted to different individuals can be obtained, and the problem of large data difference of the same individual in different periods can be solved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
FIG. 1 is a corresponding abbreviation list when different methods are selected for performance index testing;
FIG. 2 is a comparison list of calculated results of selecting performance indicators corresponding to different methods;
FIG. 3 is a histogram of the number of individuals corresponding to cross entropy losses in each interval after ordinary pre-training, training with only the meta-learning method, and pre-training with the algorithm of the present invention;
FIG. 4 is a flow chart of the algorithm of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
An arrhythmia detection algorithm based on curriculum meta-learning, comprising the following steps:
s1, pre-training the existing neural network model in a pre-training data set to obtain a pre-training model, wherein in the pre-training data set, arrhythmia detection is performed on each individual to obtain a meta-task Tb, the meta-task corresponds to a plurality of samples, and one sample is heart rhythm detection data; the pre-training data set includes a pre-training set and a pre-training validation set.
The pre-training in step S1 includes a meta learning method and a course learning method; the meta-learning method is used for learning the pre-training model and can be quickly adapted to the parameters of a new individual, and the course learning method is used for providing an easy-to-difficult learning sequence for the meta-learning method.
The pre-training comprises the following specific steps:
s1.1, calculating cross entropy loss of a sample, and determining a difficulty value of each element task; the difficulty value is a measure of the complexity of each meta-task or the difficulty degree of the model for learning the meta-task, and as long as the model determines, a fixed value can be set for each meta-task as the "difficulty value" of the meta-tasks.
The specific steps in step S1.1 to calculate the difficulty value are as follows:
s1.1.1, training by using K samples in each meta-task, and testing by using the rest samples in each meta-task.
S1.1.2, calculating the average cross entropy loss of all samples corresponding to each meta taskb，
Wherein N is the number of samples, M is the number of types of arrhythmia, and when the type of the sample i is c, the real label yicIs 1, otherwise is 0, picThe probability that the sample i given for the meta-learning model is the class c, the difficulty value of the meta-task Tb is
S1.2, selecting a plurality of meta tasks for training at each step of pre-training. The basis for selecting the meta-tasks is the difficulty value of each meta-task, a threshold value of the difficulty value is set, the meta-tasks with the difficulty values smaller than the threshold value are selected with equal probability, and the meta-tasks with the difficulty values larger than the threshold value are not selected. The application of the course learning method is embodied in the step, the threshold value is set to be increased along with the increase of the number of pre-training steps, and the number of the pre-training steps is the pair meta learning model parameter theta0The number of updates performed. Firstly setting a threshold with a lower difficulty value, then screening each meta task according to the threshold, after the pre-training of the step is completed, setting the threshold to be higher than the threshold of the previous difficulty value, then screening the rest meta tasks, and sequentially finishing the threshold setting and the meta task screening for a plurality of times, wherein the threshold setting is higher than the difficulty value of the threshold setting for the previous time each time until the screening of all the meta tasks is completed, and the difficulty value V in the step S1.1.2 is used as the difficulty valuebThe formula (2) can be used for obtaining that the difficulty value of all the meta-tasks is less than 1.
S1.3, initial weight theta of pre-training model0Updating to obtain an updated meta-learning model parameter theta'; wherein the learning is obtained by training with a meta learning method and a course learning methodThe pre-trained model is referred to as a meta-learning model. Wherein the initial weight θ of the pre-trained model0The specific steps of updating include:
s1.3.1, randomly dividing the set of samples in the pre-training data set into a sample support set and a sample query set.
S1.3.2, if the pre-training model parameter is theta at this time0Reproduction of thetab＝θ0Calculating the parameter thetabCross-entropy loss over sample query set:
where M is the number of classes of arrhythmia and when the class of sample i is c, the true label yicIs 1, otherwise is 0, picThe probability that sample i is of class c is given for the meta-learning model.
S1.3.3, performing further gradient descent on the sample support set
Where a is the learning rate, where a is,
is given a parameter of thetabThe cross-entropy loss of the pre-trained model over the sample query set,
is a cross entropy loss pair thetabOf the gradient of (c).
S1.3.4 Cross entropy loss on sample query set, Pair θ0And (6) updating.
S1.3.5, calculating parameter as theta0If the cross entropy loss of the pre-training verification set is not reduced for a plurality of continuous times, the updating is stopped, and an updated meta-learning model parameter theta' is obtained.
And S2, transferring the pre-training model obtained in the S1 to a new target set for fine adjustment, wherein each subset in the target set is an individual and corresponds to a plurality of samples, and one sample is heart rhythm detection data. Wherein, the fine tuning comprises the following steps:
s2.1, dividing a set of samples of a target data set into a target sample training set, a target sample testing set and a target sample verifying set;
s2.2, to theta0Carrying out transition; the transition method comprises the following steps:
s2.2.1, parameter to be copied theta0' further gradient descent over the target sample training set T:
s2.2.2, calculating cross entropy loss on the target sample training set;
s2.2.3, using calculated cross entropy loss pairs theta0Updating is carried out;
s2.2.4, repeating the steps of 2.2.1-2.2.3 for S times;
s2.2.5, completing the transition.
And S2.3, training on the target sample training set by using a common random gradient descent method.
And S2.4, sequentially trying each hyper-parameter combination according to the hyper-parameters including the learning rate alpha, the learning rate beta and the S, judging whether the inherent classification label on the current sample is correct or not according to the probability of the class of the current sample given in the trained meta-learning model, calculating the classification accuracy of the current sample on a target sample verification set, wherein the classification accuracy on the target sample verification set is the number of classified correct samples/the total number of samples, and then selecting a group of hyper-parameters with the highest accuracy.
S.2.5, calculating performance indexes of the model on a target sample test set, such as the area under the ROC (receiver Operating characterization) curve (AUC), the accuracy, the average Accuracy (AP), the accuracy, the recall rate, the specificity and the F1-score.
The first embodiment is as follows: the pre-training stage comprises a pre-training stage and a pre-training verification stage, and in the pre-training stage, the number of the selected batch meta-tasks in each step of pre-training is set to be 9. In the pre-training verification stage, 9 individuals are randomly selected as a pre-training verification set. And performing verification once in each step of the pre-training, stopping the training of the pre-training model until the verification of the rising trend of the cross entropy loss occurs, and storing the parameter with the minimum verification cross entropy loss, wherein the parameter is theta'.
The over-parameters appearing in the fine tuning process comprise a learning rate alpha, a learning rate beta and S, a training step number S epsilon {50,100,200} is set, and the learning rate alpha epsilon {10 }-2,5*10-3,10-3And a learning rate β ∈ {10 }-2,10-3,10-4Try each combination of superparameters in turn, for a total of 27 combinations. And judging whether the inherent classification label on the current sample is correct or not according to the probability of the class of the current sample given in the trained meta-learning model, calculating the classification accuracy of the current sample on a target sample verification set, wherein the classification accuracy on the target sample verification set is the number of classified correct samples/the total number of samples, and then selecting a group of hyper-parameters with the highest accuracy.
Referring to fig. 1 to 3, performance indexes of the model, such as area under roc (receiver Operating characterization) curve (AUC), accuracy, average Accuracy (AP), accuracy, recall, specificity, and F1-score, are calculated on the target sample test set. In fig. 1, the MAML, i.e., the meta learning method, the course learning method, and the fine tuning transition stage, are arbitrarily selected and named, for example, all three are named as "MC + PF", only both the meta learning method and the course learning method are selected as "MC + F", only both the meta learning method and the fine tuning transition stage are selected as "M + PF", only the course learning method is selected as "M + F", and none of the three are named as "Vanilla".
After naming is completed, performance indexes of the models on the test set, such as area under ROC (receiver Operating characterization) curve (AUC), accuracy, average Accuracy (AP), accuracy, recall, specificity and F1-score, are calculated. The calculation results are shown in fig. 2.
As shown in FIG. 3, the average cross entropy loss (loss) of the current individual is calculated for each individual i in the pre-training setiThe specific method is to calculate all samples of the current individualCross entropy loss and averaging. Then will loseiBy all lossiThe maximum value maxloss in (1) is normalized:
therein, lossiIs cross entropy loss after normalization, maxloss is the maximum value of cross entropy loss, and loss can be obtained by a formulai' is between 0 and 1. Later, we will losei' dividing into 10 intervals, and counting the number of individuals corresponding to each interval. It can be seen that the method of the present invention can control the cross-entropy loss of the model in a lower range on most individuals. In the figure, the MAML is a meta-learning method.
A system for arrhythmia detection algorithm based on course meta-learning comprises a data processing module, a heart rate detection module and a data storage module, wherein the heart rate detection module and the data storage module are electrically connected with the data processing module in sequence. The heart rate detection module is used for detecting the heart rate value of an individual, the data storage module is used for storing a pre-training data set, heart rate detection data, a target set and the like, and the data processing module is used for extracting required information from the data storage module to process.
According to the method, the original pre-training model is trained through the combination of the meta-learning method and the course learning method, and then the model is moved to a new target set for fine adjustment, so that parameters which can be rapidly adapted to different individuals can be obtained, and the problem of large data difference of the same individual in different periods can be solved.
The invention also discloses a computer readable storage medium, which stores a computer program, and executes the arrhythmia detection algorithm after the computer program runs.
The invention also discloses a computer system, which comprises a processor and a storage medium, wherein the storage medium is stored with a computer program, and the processor reads the computer program from the storage medium and runs the computer program to execute the arrhythmia detection algorithm.
Those of skill would further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
The various illustrative logical blocks, modules, and circuits described in connection with the embodiments disclosed herein may be implemented or performed with a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In the alternative, the processor and the storage medium may reside as discrete components in a user terminal.
In one or more exemplary embodiments, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media may be any available media that can be accessed by a computer. By way of example, and not limitation, such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a web site, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. Disk (disk) and disc (disc), as used herein, includes Compact Disc (CD), laser disc, optical disc, Digital Versatile Disc (DVD), floppy disk and blu-ray disc where disks (disks) usually reproduce data magnetically, while discs (discs) reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.
Although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. An arrhythmia detection algorithm based on curriculum meta-learning, which is characterized by comprising the following steps:
s1, pre-training an existing neural network model in a pre-training data set to obtain a pre-training model, wherein in the pre-training data set, arrhythmia detection is performed on each individual to obtain a meta-task Tb, the meta-task corresponds to a plurality of samples, and one sample is rhythm detection data; the pre-training data set comprises a pre-training set and a pre-training verification set;
s2, transferring the pre-training model obtained in the S1 to a new target set for fine adjustment, wherein each subset in the target set is an individual, each subset corresponds to a plurality of samples, and one sample is heart rhythm detection data.
2. The algorithm of claim 1, wherein: the pre-training described in step S1 includes a meta learning method and a course learning method; the meta-learning method is used for learning parameters which can be fast adapted to new individuals for the pre-training model, and the course learning method is used for providing an easy-to-difficult learning sequence for the meta-learning method.
3. The algorithm of claim 2, wherein the pre-training comprises:
s1.1, calculating cross entropy loss of a sample, and determining a difficulty value of each meta task; the difficulty value is used for measuring the complexity of each meta-task or the difficulty degree of the model for learning the meta-task, and as long as the model is determined, a fixed value can be set for each meta-task to serve as the 'difficulty value' of the meta-tasks;
s1.2, selecting a plurality of meta-tasks for training at each step of the pre-training;
s1.3, initial weight theta of pre-training model0Updating to obtain an updated meta-learning model parameter theta'; the pre-training model obtained after training by the meta-learning method and the course learning method is called a meta-learning model.
4. The algorithm according to claim 3, characterized in that the specific steps of calculating the difficulty value in step S1.1 are as follows:
s1.1.1, training by using K samples in each meta-task, and testing by using the rest samples in each meta-task;
s1.1.2, calculating the average cross entropy loss of all the samples corresponding to each meta taskb，
Wherein N is the number of samples, M is the number of types of arrhythmia, and when the type of the sample i is c, the real label yicIs 1, otherwise is 0, picThe probability that the sample i given for the meta-learning model is the class c, the difficulty value of the meta-task Tb is
5. The algorithm of claim 3, wherein: the basis for selecting the meta-tasks in step S1.2 is to set a threshold value for the difficulty value of each of the meta-tasks, the meta-tasks having difficulty values smaller than the threshold value are selected with equal probability, and the meta-tasks having difficulty values larger than the threshold value are not selected.
6. The algorithm of claim 3, wherein: initial weight θ of pre-trained model in step S1.30The specific steps of updating include:
s1.3.1, randomly dividing the set of samples in the pre-training data set into a sample support set and a sample query set;
s1.3.2, if the pre-training model parameter is theta at this time0Reproduction of thetab＝θ0Calculating the parameter thetabCross-entropy loss over the sample query set:
where M is the number of classes of arrhythmia and when the class of sample i is c, the true label yicIs 1, otherwise is 0, picThe probability that a sample i given for the meta-learning model is of the class c;
s1.3.3, performing a further gradient descent on the sample support set
Where a is the learning rate, where a is,
is given a parameter of thetabThe cross-entropy loss over the set of queries,
is a cross entropy loss pair thetabA gradient of (a);
s1.3.4, query cross entropy loss on set with the sample, pair theta0Updating is carried out;
s1.3.5, calculating parameter as theta0If the cross entropy loss of the pre-training verification set is not reduced for a plurality of continuous times, stopping updating to obtain an updated meta-learning model parameter theta'.
7. The algorithm according to claim 6, wherein the fine tuning in step S2 comprises the steps of:
s2.1, dividing the sample set of the target data set into a target sample training set, a target sample testing set and a target sample verifying set;
s2.2, to theta0Carrying out transition; wherein the transition method comprises the following steps:
s2.2.1, parameter to be copied theta0' performing a further gradient descent on the target sample training set T:
s2.2.2, calculating cross entropy loss on the target sample training set;
s2.2.3, using calculated cross entropy loss pairs theta0Updating is carried out;
s2.2.4, repeating the steps of 2.2.1-2.2.3 for S times;
s2.2.5, completing the transition.
And S2.3, training on the target sample training set by using a common random gradient descent method.
And S2.4, sequentially trying each hyper-parameter combination according to the hyper-parameters including the learning rate alpha, the learning rate beta and the S, judging whether the inherent classification label on the current sample is correct or not according to the probability of the class of the current sample given in the trained meta-learning model, calculating the classification accuracy of the current sample on a target sample verification set, wherein the classification accuracy on the target sample verification set is the number of classified correct samples/the total number of samples, and then selecting a group of hyper-parameters with the highest accuracy.
S.2.5, calculating performance indexes of the model on the target sample test set, such as area under the ROC (receiver Operating characterization) curve (AUC), accuracy, average Accuracy (AP), accuracy, recall, specificity and F1-score.
8. A computer-readable storage medium characterized by: a computer program stored on a medium, which when executed, performs a curriculum meta-learning based arrhythmia detection algorithm as claimed in any of claims 1 to 7.
9. A computer system, characterized by: comprising a processor, a storage medium having a computer program stored thereon, the processor reading and executing the computer program from the storage medium to perform a curriculum meta-learning based arrhythmia detection algorithm as claimed in any of claims 1 to 7.
10. A system for a curriculum-learnt based arrhythmia detection algorithm as claimed in any of claims 1 to 7, wherein: including data processing module and in proper order with data processing module electric connection's rhythm of the heart detection module and data storage module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111667670.5A CN114330446B (en) | 2021-12-30 | 2021-12-30 | Arrhythmia detection algorithm and system based on curriculum meta-learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111667670.5A CN114330446B (en) | 2021-12-30 | 2021-12-30 | Arrhythmia detection algorithm and system based on curriculum meta-learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114330446A true CN114330446A (en) | 2022-04-12 |
| CN114330446B CN114330446B (en) | 2023-04-07 |
Family
ID=81020133
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111667670.5A Active CN114330446B (en) | 2021-12-30 | 2021-12-30 | Arrhythmia detection algorithm and system based on curriculum meta-learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114330446B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2430975A1 (en) * | 2010-09-17 | 2012-03-21 | Stichting IMEC Nederland | Principal component analysis or independent component analysis applied to ambulatory electrocardiogram signals |
| US20190286986A1 (en) * | 2018-01-11 | 2019-09-19 | Huawei Technologies Co., Ltd. | Machine Learning Model Training Method And Apparatus |
| CN112417319A (en) * | 2020-11-24 | 2021-02-26 | 清华大学 | Site recommendation method and device based on difficulty sampling meta-learning |
| CN113361645A (en) * | 2021-07-03 | 2021-09-07 | 上海理想信息产业（集团）有限公司 | Target detection model construction method and system based on meta-learning and knowledge memory |
| CN113837000A (en) * | 2021-08-16 | 2021-12-24 | 天津大学 | Small sample fault diagnosis method based on task sequencing meta-learning |
-
2021
- 2021-12-30 CN CN202111667670.5A patent/CN114330446B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2430975A1 (en) * | 2010-09-17 | 2012-03-21 | Stichting IMEC Nederland | Principal component analysis or independent component analysis applied to ambulatory electrocardiogram signals |
| US20190286986A1 (en) * | 2018-01-11 | 2019-09-19 | Huawei Technologies Co., Ltd. | Machine Learning Model Training Method And Apparatus |
| CN112417319A (en) * | 2020-11-24 | 2021-02-26 | 清华大学 | Site recommendation method and device based on difficulty sampling meta-learning |
| CN113361645A (en) * | 2021-07-03 | 2021-09-07 | 上海理想信息产业（集团）有限公司 | Target detection model construction method and system based on meta-learning and knowledge memory |
| CN113837000A (en) * | 2021-08-16 | 2021-12-24 | 天津大学 | Small sample fault diagnosis method based on task sequencing meta-learning |
Non-Patent Citations (2)
| Title |
|---|
| YUDONG CHEN ET AL.: "Curriculum Meta-Learning for Next POI Recommendation" * |
| 王志豪: "基于元学习的图像特定的超分辨率技术" * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114330446B (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10130307B2 (en) | Electrocardiogram (ECG) authentication method and apparatus | |
| RU2517286C2 (en) | Classification of samples data | |
| CN106653001A (en) | Baby crying identifying method and system | |
| JP2020036633A (en) | Abnormality determination program, abnormality determination method and abnormality determination device | |
| KR20170082440A (en) | Method and apparatus for electrocardiogram authentication | |
| CN108877947B (en) | Depth sample learning method based on iterative mean clustering | |
| CN112560710B (en) | Method for constructing finger vein recognition system and finger vein recognition system | |
| CN112599218A (en) | Training method and prediction method of drug sensitivity prediction model and related device | |
| CN114330446B (en) | Arrhythmia detection algorithm and system based on curriculum meta-learning | |
| CN111131279B (en) | Safety perception model construction method based on immune theory | |
| Tishby | Statistical physics models of supervised learning | |
| Saleh et al. | Healthcare Embedded System for Predicting Parkinson's Disease Based on AI of Things | |
| Wang et al. | Early Diagnosis of Parkinson's Disease with Speech Pronunciation Features Based on XGBoost Model | |
| JP2021504782A (en) | Artificial intelligence-based insurance claim detection device and method | |
| CN112766403A (en) | Incremental clustering method and device based on information gain weight | |
| CN114300116B (en) | Robust syndrome detection method based on online classification algorithm | |
| CN117312971B (en) | Autism spectrum disorder individual identification device | |
| CN112668362A (en) | Testimony comparison model training method for dynamic optimization class agent | |
| Jiang et al. | Adaptive functional connectivity network using parallel hierarchical BiLSTM for MCI diagnosis | |
| CN114239742B (en) | Medical data classification method based on rule classifier and related equipment | |
| Fombellida et al. | Application of artificial metaplasticity fundamentals to WBCD breast cancer database classification method | |
| KR102617046B1 (en) | Sleep stage prediction method using deep learning model and analysis apparatus | |
| JP7464114B2 (en) | Identification device, identification method, and recording medium | |
| Billah | Symptom analysis of Parkinson disease using SVM-SMO and Ada-Boost classifiers | |
| CN112561188A (en) | People flow prediction method and device |
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 |