CN110797119B - Intelligent health risk monitoring device and transfer learning method - Google Patents
Intelligent health risk monitoring device and transfer learning method Download PDFInfo
- Publication number
- CN110797119B CN110797119B CN201910899722.8A CN201910899722A CN110797119B CN 110797119 B CN110797119 B CN 110797119B CN 201910899722 A CN201910899722 A CN 201910899722A CN 110797119 B CN110797119 B CN 110797119B
- Authority
- CN
- China
- Prior art keywords
- data
- module
- health
- health risk
- monitoring device
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention provides an intelligent health risk monitoring device which comprises a basic module, a transfer learning module, a multi-level weighting parameter learning module, a health data updating module, a super-health parameter processing module and a mixed value attribute data processing module, wherein the basic module is used for storing a plurality of health data; the transfer learning module, the multi-level weighting parameter learning module, the health data updating module, the ultra-health parameter processing module and the mixed value attribute data processing module are respectively connected with the basic module. A health risk assessment model can be generated in the health risk intelligent monitoring device. The invention also provides a transfer learning method. Compared with the prior art, the invention has the beneficial effects that: the device constructs an efficient health risk assessment model capable of processing the hyperparameter and mixed value attribute data, and the model can well generalize newly-added data with different dimensionalities. The model is also capable of self-learning based on new data being acquired continuously. The device has high data utility. The transfer learning method has high efficiency.
Description
Technical Field
The invention relates to the technical field of big data statistical analysis and mining, in particular to an intelligent health risk monitoring device and a transfer learning method.
Background
At present, the health examination of residents is mainly performed by medical institutions and physical examination institutions. The health examination mode is mainly a traditional mode, and is lack of individual pertinence and accuracy, so that some necessary examinations are not performed, and some unnecessary examinations are performed, so that certain resource waste is caused. In addition, the traditional health examination mode lacks individual health condition assessment, and cannot acquire the difference characteristics of individual genetic conditions, living habits and regional characteristics. Most importantly, the traditional health examination mode belongs to a periodic examination or a triggered examination, the results of a plurality of health examinations are relatively independent, and the correlation among data acquired by the examinations is neglected for a great length.
In order to realize health risk prediction, in the prior art, a person skilled in the art tries to analyze health parameters of a part of fixed dimensions by means of big data, cloud computing, intelligent evaluation and the like so as to obtain a prediction result of health risk. However, such an existing health risk intelligent monitoring device lacks multidimensional data processing capability due to problems of data dimension and data utility, lacks prediction accuracy due to its inability to process multidimensional data well, and has problems of poor learning, updating, and adjusting capabilities.
Disclosure of Invention
In view of the above, in order to solve the problems of poor multidimensional data processing capability and poor data utility of the existing health risk intelligent monitoring device in the prior art, the invention provides a health risk intelligent monitoring device, which comprises a basic module, and a migration learning module, a multi-level weighting parameter learning module, a health data updating module, a super-health parameter processing module and a mixed value attribute data processing module which are respectively connected with the basic module; the intelligent health risk monitoring device can generate a health risk assessment model; the transfer learning module is used for filling data; the multi-level weighted parameter learning module is used for executing parameter learning of a health data structure containing high-dimensional parameters and variables; the health data updating module is used for executing updating and expansion of different types of health data; the super-health parameter processing module is used for processing reduction of the super-health parameter; and the mixed value attribute data processing module is used for completing reduction of the mixed value attribute of the health risk assessment model.
Preferably, a multi-hidden-layer feature subspace module is arranged in the multi-level weighting parameter learning module, and the multi-hidden-layer feature subspace module is used for establishing linear mapping by a feature subspace hidden layer method and projecting the grouping features of the high-dimensional data into a low-dimensional space.
Preferably, a nested regular mixing module is arranged in the multi-level weighting parameter learning module, and the nested regular mixing module is used for nesting the regular mixing model into an unsupervised clustering process and clustering data at the same time.
Preferably, the health data update module updates the relative data set D new Health risk assessment model L new Expressed as:
wherein the operation sign
Represents an integration of the different evaluation models,
represent the weights of the base model, an
Representing a health risk assessment model trained on newly added data, L old Representing an original health risk assessment model;
preferably, the relative data set D new Expressed as:
wherein the content of the first and second substances,
for newly added health data, D wld Is the original data set.
Preferably, the weights of the base models
The determination method comprises the following steps:
when the newly added data is longitudinally added data, the newly added health data is obtained by a kernel density estimation method
Probability density function of
Measure of the passing
And p old Similarity determination between them
Value of (a), p old For the original data set D old A probability density function of;
when the newly added data is the data which is transversely added, the weight of the base model is determined by the information amount of the data set
Value of (a), the original data set D old And newly added health data
The amount of information of (a) is calculated by substituting entropy.
Preferably, the super-health parameter processing module is capable of executing a deep neural network weight training method independent of a gradient descent principle, and the deep neural network weight training method independent of the gradient descent principle includes:
step S11, constructing a deep learning neural network containing K hidden layers in the following form:
wherein X is H (0) And is the original super-healthy input matrix, { H } (K) Where K is 1, 2, L, K, X, w (K) The weight of the K layer of the deep neural network;
step S12, determining w in a non-iterative manner (K) ，w (K) Expressed as:
W (K) ＝[H (K) ] T H (K-1)
wherein [ H ] (K) ] T Represents H (K) The transposing of (1).
Preferably, the mixed-value attribute data processing module is configured to convert a discrete-value attribute into a continuous-value attribute, and the method for converting a discrete-value attribute into a continuous-value attribute performed by the mixed-value attribute data processing module includes the following steps:
step S31, performing one-hot encoding on the discrete value attribute in a one-hot mode;
and step S32, constructing a deep neural network to convert the one-hot codes.
Preferably, in the training of the neural network involved in the training step S32, the loss function adopts a method of combining the mean square error and the information amount.
The invention also provides a transfer learning method, which comprises the following steps:
step S21, training a first deep neural network by using a data set with a class mark;
step S22, selecting a plurality of layers as a first base layer, and the rest layers are first target layers;
step S23, establishing a new deep neural network for the new data set without class marks or incomplete class marks;
step S24, the first basic layer is migrated to a new deep neural network to be used as a second basic layer, a new second target layer is trained, and the deep neural network is perfected;
and step S25, looping the steps S22 to S24, and trying to select different layers for the first base layer 21 for multiple times until a reasonable base layer number is obtained, and establishing the self-adaptive migration learning model.
Compared with the prior art, the invention has the beneficial effects that:
the intelligent health risk monitoring device establishes an efficient health risk assessment model capable of processing the hyperparameter and mixed value attribute data, and the model can well generalize newly-added data with different dimensions. The health risk assessment model is able to learn itself based on the constant acquisition of new data. The intelligent health risk monitoring device is not limited to the influence of data dimensions, the data effectiveness is high, and the learning efficiency of the transfer learning method is high.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.
Fig. 1 is a schematic structural diagram of an intelligent health risk monitoring device according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a deep neural network structure including a base layer and a target layer according to an embodiment of the present invention;
fig. 3 is a flowchart of a transfer learning method according to an embodiment of the present invention.
Reference numerals:
the system comprises a basic module 1, a migration learning module 2, a multi-level weighting parameter learning module 3, a health data updating module 4, a super-health parameter processing module 10, a mixed value attribute data processing module 11, a first basic layer 21, a first target layer 22, a multi-hidden-layer feature subspace module 31 and a nested mixed regular model module 32.
Detailed Description
The above and further features and advantages of the present invention are described in more detail below with reference to the accompanying drawings.
In the description of the present invention, it is to be understood that the terms "first", "second" and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implying any number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two unless specifically defined otherwise.
In the present invention, unless otherwise expressly stated or limited, the terms "mounted," "connected," "secured," and the like are to be construed broadly and can, for example, be fixedly connected, detachably connected, or integrally formed; may be mechanically coupled, may be electrically coupled or may be in communication with each other; they may be directly connected or indirectly connected through intervening media, or they may be connected internally or in any other suitable relationship, unless expressly stated otherwise. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
In the description of the specification, reference to the description of "one embodiment," "some embodiments," "an example," "a specific example," or "some examples" or the like means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention.
In the following description, for purposes of explanation and not limitation, specific details are set forth, such as particular system structures, techniques, etc. in order to provide a thorough understanding of the embodiments of the invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.
In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
In order to explain the technical means of the present invention, the following description will be given by way of specific examples.
Fig. 1 is a schematic structural diagram of an intelligent health risk monitoring device in an embodiment of the present invention; as shown in fig. 1, the intelligent health risk monitoring device provided in the embodiment of the present invention includes a base module 1, a migration learning module 2, a multi-level weighted parameter learning module 3, a health data updating module 4, a super-health parameter processing module 10, and a mixed value attribute data processing module 11. The migration learning module 2, the multi-level weighting parameter learning module 3, the health data updating module 4, the super-health parameter processing module 10 and the mixed value attribute data processing module 11 are respectively connected with the basic module 1.
A health risk assessment model can be generated in the health risk intelligent monitoring device.
The migration learning module 2 is used for data filling. The migration learning module 2 is preferably an adaptive migration learning module. The transfer learning module 2 is a transfer learning module that can operate in the absence of health parameters. Missing parameters in machine learning refer to clustering, grouping, pruning, or truncation of data in the coarse data due to lack of information. The loss of health data is often caused by failure of data acquisition and storage, as well as subjective errors or knowledge limitations during manual intervention.
In the learning process of the deep neural network, the base layer learns about generalized information (such as types, structures and the like of health data), and then utilizes the target layer to learn about different data sets in a targeted manner. The base layer has commonality for deep neural networks of different health data sets. The migration learning module 2 trains the structure and weight of the deep neural network basic layer by using the perfect health data set, and migrates to the deep neural network learning containing the missing data, thereby making up the missing data.
The migration learning method is adopted as a key technology of data filling, and high-precision data are provided for feature extraction and later-stage machine learning. In the process of identifying the sample by the human brain, basic judgment is made firstly, and then special classification is carried out. The deep neural network adopts the same mechanism, and the weights of all layers of the neural network are trained by identifying and recording the samples with the class marks, so that the unclassified objects are identified. Because the preliminary judgment processes of different samples are similar, the neural network (the basic layer) of the part has reusability.
As shown in fig. 2 and fig. 3, taking a deep neural network including a base layer and a target layer as an example, the migration learning method executed by the migration learning module 2 includes the following steps:
step S21, training a first deep neural network by using a data set with a class mark;
step S22, selecting several layers as the first basic layer 21, and the rest layers as the first target layer 22;
step S23, establishing a new deep neural network for the new data set without class marks or incomplete class marks;
step S24, the first base layer 21 is migrated to a new network to be used as a second base layer, a new second target layer is trained, and the deep neural network is perfected;
and step S25, looping step S22 to step S24, and trying to select different layers for the first base layer 21 multiple times to obtain a reasonable number of base layer layers, and building an adaptive migration learning model.
The transfer learning module 2 has the beneficial effects that:
the migration learning module 2 pre-trains a deep neural network through the acquired universal large data set, so that the network can fully learn shallow features. When a specific new data set is met, the pre-trained deep neural network is utilized to perform fine adjustment, so that the deep neural network can sufficiently learn high-level features on small data, and a better classification effect is achieved. The transfer learning module 2 can complete the missing health parameters and complete the subsequent machine learning operation by using a transfer learning method.
With the reduction of data acquisition cost and the maturity of field knowledge, ultra-large health data containing high-dimensional parameters are integrated as a new target for research, and challenge the traditional technology.
The multi-level weighting parameter learning module 3 is used for executing parameter learning of health data structures and variables containing high-dimensional parameters.
The intelligent health risk monitoring device with the multi-level weighting parameter learning module 3 can extract a small amount of effective classification parameters through the structure and the weight of a small amount of sample data learning parameters, can evaluate the health risks of people without traversing all parameters, and greatly reduces the calculation complexity.
Meanwhile, the intelligent health risk monitoring device constructs a self-adaptive multi-parameter health risk prediction model by learning information such as the hierarchical structure, grouping, weight, correlation and the like of parameters. Aiming at different accuracy requirements and software and hardware cost capabilities, a learning model with gradually increased accuracy is provided under the condition of selecting different major health (such as blood pressure) or minor health (high pressure or low pressure with blood pressure) parameters.
In the research development aiming at feature grouping, the statistical correlation between the healthy big data feature group and the group cannot be effectively mined, for example, whether Gaussian mixture distribution exists or not, whether positive correlation or negative correlation exists between the groups or not, and the like.
Therefore, a multi-hidden-layer feature subspace module 31 and a nested mixture regularization module 32 are provided in the multi-level weighting parameter learning module 3.
The multi-hidden-layer feature subspace module 31 establishes linear mapping by the feature subspace hidden layer method to obtain high-dimensional data
Can be grouped by
Projected into a low dimensional space. Through the framework, original data information is deconstructed, so that feature groups are optimized by using a discretized biological evolution algorithm, weights are optimized by using an analytical method, feature analysis with high precision is obtained through low calculated amount, and subsequent machine learning is carried out.
With respect to the nested mixture regularization module 32, there are different characteristics of the weights of features in different types of super-high dimensional data. For example, there are data with small differences in importance of features, and there are a large number of redundant variables and redundant features in the data. The regularization method provides an effective way for solving the above problems. The invention uses the interactive relation between the two learning processes of feature grouping and clustering to nest the mixed regular model into the unsupervised clustering process, and simultaneously clusters the data and learns the features.
The multi-level weighting parameter learning module 3 adjusts and increases the constructed learning model, for example, the overfitting degree between the models is reduced by introducing a regularization technology, so that the models have better generalization capability; and secondly, optimizing and reconstructing the initial features through a high-quality learning algorithm.
The health data update module 4 is used to perform updates and extensions of different types of health data. The intelligent health risk monitoring device has the self-learning capability according to continuous acquisition of new data so as to adapt to the real-time updating requirements of different types of health data.
The health data update module 4 needs to solve two problems:
firstly, how to use newly added data with the same dimensionality to complete the updating of the parameters of the existing health risk assessment model specifically comprises the calculation of the probability distribution of the newly added data, the consistency measurement of the probability distribution of the newly added data and the existing data, the fusion of the newly added data across platforms and regions, and the like;
secondly, how to update parameters of the existing intelligent health risk monitoring device by using newly added data with different dimensionalities, when the new health data has health parameters different from the existing data, the problem of fusion of a health risk assessment model based on newly added attribute training and a health risk assessment model based on existing attribute training is researched, and the method specifically comprises effectiveness measurement of the newly added health parameters and design of an efficient model fusion method.
The health data updating module 4 processes the following modes:
for newly added health data
Assume that the current data set is D old With a probability density function of p old The existing health risk assessment model of the intelligent health risk monitoring device is L old The health risk assessment model based on newly added data training is
Then relative dataset D new (relative data set D) new Can be expressed as:
the health risk assessment model of (a) may be expressed as:
wherein the operation sign
Represents an integration of the different evaluation models,
represent the weights of the base model, an
The key here is the determination of the weights. For longitudinally added data, first, new added health data is obtained by a kernel density estimation method
Of the probability density function
Passing through metric
And p old The similarity between them
The value of (a). For laterally increasing data, the weights are determined by measuring the amount of information in the data set
The current data set is D old And new addition ofHealth data of
The amount of information of (a) is calculated by substituting the entropy.
The health data updating module 4 can update different types of health data in real time.
The intelligent health risk monitoring device can construct an efficient health risk assessment model capable of processing the super-parameter and mixed value attribute data through the basic module 1, the super-health parameter processing module 10 and the mixed value attribute data processing module 11. The method solves the problem of the effectiveness of newly adding health data with different dimensions. Namely, how to effectively integrate the health risk assessment model trained based on the newly added health data with the existing health risk assessment model trained based on the historical health data.
The super-health parameter processing module 10 is configured to process the reduction of the super-health parameter, that is, construct a fast unsupervised feature learning model to extract a deep representation of the original super-health parameter. For processing reductions in the super-health parameters, the super-health parameter processing module 10 uses a deep neural network weight training method that does not rely on the gradient descent principle.
The deep neural network weight training method independent of the gradient descent principle comprises the following steps:
step S11, constructing a deep learning neural network containing K hidden layers in the form of the following formula:
wherein X is H (0) For the original super-healthy input matrix, { H } (K) Where K is 1, 2, L, K, X, w (K) Is the weight of the K layer of the deep neural network.
Step S12, determining w in a non-iterative manner (K) . The specific calculation can be expressed as:
W (K) ＝[H (K) ] T H (K-1) ；
wherein, V (K) For the random assignment matrix, [ H ] (K) ] T Represents H (K) The transposing of (1).
The mixed value attribute data processing module 11 is configured to complete reduction of the mixed value attribute of the health risk assessment model, where reduction of the mixed value attribute of the health risk assessment model requires conversion of a discrete value attribute into a continuous value attribute, and a method for converting the discrete value attribute into the continuous value attribute includes the following steps:
and step S32, constructing a deep neural network to convert the one-hot codes.
In the process of training the neural network, the loss function adopts a mode of combining the mean square error and the information quantity, so that the loss of the information quantity is minimum while the lowest error is obtained for the continuous value attribute after conversion.
And constructing a fast and efficient health risk assessment model based on the converted effective representation of the original super-health parameters, namely establishing the health risk assessment model of the individual.
The invention also provides an optimization problem solving method based on complementary constraint aiming at the non-convexity of the loss function starting from the logistic regression model.
For non-convex logistic regression loss functions, a complementary constrained optimization problem is constructed that is shaped as:
min f(x)
s.t.g(x)≥0,h(x)≥0,<g(x),h(x)>≥0；
the non-negative vertical constraint set has the advantages of equivalent optimization, good initialization, strong convergence, greedy monotony, module optimization and the like. The complementary constraint optimization problem is solved by adopting a smooth approximation method based on a minimized entropy function.
Health risk intelligent monitoring device's beneficial effect lies in:
the method can construct an efficient health risk assessment model capable of processing the hyperparameter and mixed value attribute data, and the model can well generalize newly-added data with different dimensions. The health risk assessment model can continuously acquire new data and has the self-learning ability.
Other contents of the basic module 1 belong to the prior art category, and are not described herein.
The above-mentioned embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the same; 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; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present invention, and are intended to be included within the scope of the present invention.
Claims (9)
1. An intelligent health risk monitoring device is characterized by comprising a basic module, a migration learning module, a multi-level weighted parameter learning module, a health data updating module, a super-health parameter processing module and a mixed value attribute data processing module, wherein the migration learning module, the multi-level weighted parameter learning module, the health data updating module, the super-health parameter processing module and the mixed value attribute data processing module are respectively connected with the basic module;
the intelligent health risk monitoring device can generate a health risk assessment model;
the transfer learning module is used for filling data;
the multi-level weighted parameter learning module is used for executing parameter learning of a health data structure containing high-dimensional parameters and variables;
the health data updating module is used for executing updating and expansion of different types of health data;
the super-health parameter processing module is used for processing reduction of the super-health parameter;
the mixed value attribute data processing module is used for completing reduction of the mixed value attribute of the health risk assessment model;
the super-health parameter processing module can execute a deep neural network weight training method independent of a gradient descent principle, and the deep neural network weight training method comprises the following steps:
step S11, constructing the following deep learning neural network containing K hidden layers:
wherein X is H (0) And is the original super-healthy input matrix, { H } (K) Where K is 1, 2, L, K, X, w (K) The weight of the K layer of the deep neural network;
step S12, determining w in a non-iterative manner (K) ，w (K) Expressed as:
W (K) ＝[H (K) ] T H (K-1)
wherein [ H ] (K) ] T Represents H (K) The transposing of (1).
2. The intelligent health risk monitoring device as claimed in claim 1, wherein a multi-hidden-layer feature subspace module is provided in the multi-level weighting parameter learning module, and the multi-hidden-layer feature subspace module is used for establishing linear mapping by a feature subspace hidden layer method, and projecting the grouping features of high-dimensional data into a low-dimensional space.
3. The intelligent health risk monitoring device as claimed in claim 1, wherein a nested mixture regularization module is provided in the multi-level weighted parameter learning module, and the nested mixture regularization module is configured to nest a mixture regularization model into an unsupervised clustering process and cluster data at the same time.
4. The intelligent health risk monitoring device as recited in claim 1, wherein the health data update module updates the relative dataset D new Health risk assessment model L new Expressed as:
wherein the operation sign
Represents an integration of the different evaluation models,
represent the weights of the base model, an
Representing a health risk assessment model trained on newly added data, L old Representing the original health risk assessment model.
5. The intelligent health risk monitoring device of claim 4, wherein the relative dataset D is new Expressed as:
wherein the content of the first and second substances,
for newly added health data, D old Is the original data set.
6. The intelligent health risk monitoring device of claim 5, wherein the weights of the base models
The determination method comprises the following steps:
when the newly added data is longitudinally added data, the newly added health data is obtained by a kernel density estimation method
Of the probability density function
Measure of the passing
And p old Similarity determination between them
Value of (a), p old For the original data set D old A probability density function of (a);
when the newly added data is the data which is transversely added, the weight of the base model is determined by the information amount of the data set
Value of (a), original data set D old And newly added health data
The amount of information of (a) is calculated by substituting entropy.
7. The intelligent health risk monitoring device according to any one of claims 1 to 6, wherein the mixed-value attribute data processing module is configured to convert a discrete-value attribute into a continuous-value attribute, and the method performed by the mixed-value attribute data processing module to convert the discrete-value attribute into the continuous-value attribute comprises the following steps:
step S31, performing one-hot encoding on the discrete value attribute in a one-hot mode;
and step S32, constructing a deep neural network to convert the one-hot codes.
8. The intelligent health risk monitoring device as claimed in claim 7, wherein the loss function employs a method of combining mean square error and information amount in the process of training the neural network involved in step S32.
9. A transfer learning method, which is used for being executed in a transfer learning module of the intelligent health risk monitoring device according to any one of claims 1 to 8; the transfer learning method comprises the following steps:
step S21, training a first deep neural network by using a data set with a class mark;
step S22, selecting a plurality of layers as a first basic layer, and the rest layers as first target layers;
step S23, establishing a new deep neural network for the new data set without class marks or incomplete class marks;
step S24, migrating the first base layer to a new deep neural network as a second base layer, training a new second target layer and perfecting the deep neural network;
and step S25, looping the steps S22 to S24, and trying to select different layers for the first base layer for multiple times until a reasonable base layer number is obtained, and establishing the self-adaptive migration learning model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910899722.8A CN110797119B (en) | 2019-09-23 | 2019-09-23 | Intelligent health risk monitoring device and transfer learning method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910899722.8A CN110797119B (en) | 2019-09-23 | 2019-09-23 | Intelligent health risk monitoring device and transfer learning method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110797119A CN110797119A (en) | 2020-02-14 |
| CN110797119B true CN110797119B (en) | 2022-09-20 |
Family
ID=69439650
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910899722.8A Active CN110797119B (en) | 2019-09-23 | 2019-09-23 | Intelligent health risk monitoring device and transfer learning method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110797119B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111493850A (en) * | 2020-04-13 | 2020-08-07 | 中国科学院深圳先进技术研究院 | Blood pressure measuring method and device based on deep neural network |
| CN113096796B (en) * | 2021-04-01 | 2022-09-02 | 四川大学华西医院 | Intelligent prediction system and method for cerebral hemorrhage hematoma expansion risk |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108389626A (en) * | 2018-02-09 | 2018-08-10 | 上海长江科技发展有限公司 | Cerebral apoplexy screening method based on artificial intelligence and system |
| CN109117864A (en) * | 2018-07-13 | 2019-01-01 | 华南理工大学 | Coronary heart disease risk prediction technique, model and system based on heterogeneous characteristic fusion |
| CN109119130A (en) * | 2018-07-11 | 2019-01-01 | 上海夏先机电科技发展有限公司 | A kind of big data based on cloud computing is health management system arranged and method |
| CN109558942A (en) * | 2018-11-20 | 2019-04-02 | 电子科技大学 | A kind of neural network moving method based on either shallow study |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11144825B2 (en) * | 2016-12-01 | 2021-10-12 | University Of Southern California | Interpretable deep learning framework for mining and predictive modeling of health care data |
-
2019
- 2019-09-23 CN CN201910899722.8A patent/CN110797119B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108389626A (en) * | 2018-02-09 | 2018-08-10 | 上海长江科技发展有限公司 | Cerebral apoplexy screening method based on artificial intelligence and system |
| CN109119130A (en) * | 2018-07-11 | 2019-01-01 | 上海夏先机电科技发展有限公司 | A kind of big data based on cloud computing is health management system arranged and method |
| CN109117864A (en) * | 2018-07-13 | 2019-01-01 | 华南理工大学 | Coronary heart disease risk prediction technique, model and system based on heterogeneous characteristic fusion |
| CN109558942A (en) * | 2018-11-20 | 2019-04-02 | 电子科技大学 | A kind of neural network moving method based on either shallow study |
Non-Patent Citations (2)
| Title |
|---|
| 学界|不通过梯度下降来训练神经网络：一个可扩展的ADMM途径;51黑科研;《WWW.51hei.com/bbs/dpj-50154-1.html》;20160511;第1-2页 * |
| 最后一期：如何更新LSTM模型？;邵洲;《www.jiqizhixin.com/articles/2019-04-01-8》;20190401;第3-5页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110797119A (en) | 2020-02-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106570597B (en) | Content popularity prediction method based on deep learning under SDN architecture | |
| CN109086805B (en) | Clustering method based on deep neural network and pairwise constraints | |
| US20200167659A1 (en) | Device and method for training neural network | |
| Černá et al. | Supervised classification of plant communities with artificial neural networks | |
| CN108805193B (en) | Electric power missing data filling method based on hybrid strategy | |
| CN110797119B (en) | Intelligent health risk monitoring device and transfer learning method | |
| CN106779064A (en) | Deep neural network self-training method based on data characteristics | |
| CN109214503B (en) | Power transmission and transformation project cost prediction method based on KPCA-LA-RBM | |
| CN108960488B (en) | Saturated load spatial distribution accurate prediction method based on deep learning and multi-source information fusion | |
| CN109934422A (en) | Neural network wind speed prediction method based on time series data analysis | |
| KR20210113336A (en) | Method for predicting target object properties based on machine learning and related devices | |
| CN116628510A (en) | Self-training iterative artificial intelligent model training method | |
| CN116029435A (en) | Environmental comfort early warning system is bred to live pig facility | |
| CN115526246A (en) | Self-supervision molecular classification method based on deep learning model | |
| CN110260914B (en) | Engineering safety monitoring system region division method based on measuring point space-time characteristics | |
| CN111091916A (en) | Data analysis processing method and system based on improved particle swarm optimization in artificial intelligence | |
| CN104850864A (en) | Unsupervised image recognition method based on convolutional neural network | |
| CN113887698A (en) | Overall knowledge distillation method and system based on graph neural network | |
| Yang | Construction of graduate behavior dynamic model based on dynamic decision tree algorithm | |
| CN116720743A (en) | Carbon emission measuring and calculating method based on data clustering and machine learning | |
| CN115661450A (en) | Category increment semantic segmentation method based on contrast knowledge distillation | |
| CN105512249A (en) | Noumenon coupling method based on compact evolution algorithm | |
| CN113360772B (en) | Interpretable recommendation model training method and device | |
| CN115293249A (en) | Power system typical scene probability prediction method based on dynamic time sequence prediction | |
| CN114943016A (en) | Cross-granularity joint training-based graph comparison representation learning method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |