CN110263919A - A kind of reverse transmittance nerve network training method based on medical treatment & health data safety - Google Patents
A kind of reverse transmittance nerve network training method based on medical treatment & health data safety Download PDFInfo
- Publication number
- CN110263919A CN110263919A CN201910535719.8A CN201910535719A CN110263919A CN 110263919 A CN110263919 A CN 110263919A CN 201910535719 A CN201910535719 A CN 201910535719A CN 110263919 A CN110263919 A CN 110263919A
- Authority
- CN
- China
- Prior art keywords
- data
- medical treatment
- researcher
- error
- method based
- 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
Classifications
-
- 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
- G06F21/602—Providing cryptographic facilities or services
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioethics (AREA)
- Computer Hardware Design (AREA)
- Computer Security & Cryptography (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The present invention discloses a kind of reverse transmittance nerve network training method based on medical treatment & health data safety, two stage separation of forward-propagating and backpropagation, be each responsible on different devices by two sides: forward-propagating process is responsible for by the data side of medical treatment & health data providing, and back-propagation process is executed by neural network researcher.Data side carries out after waiting pretreatment in batches sample data, researcher provides current feedforward network to data side, present lot data are inputted feedforward network by data side, then the error for obtaining certain batch returns to the error to researcher after audit, after researcher's backpropagation updates network, new current network is supplied to data side again, is then recycled, until error is sufficiently small or completes the entirety training of specific times.The problem of present invention efficiently solves both the training of reverse transmittance nerve network and the guarantee of medical treatment & health data safety contradictions.
Description
Technical field
The present invention relates to medical data security technology areas more particularly to a kind of based on the reversed of medical treatment & health data safety
Propagation Neural Network training method.
Background technique
Artificial neural network passes through the company of neuron as a kind of machine learning method effective in application practice
The simulation human brain that connects and cooperate solves the problems, such as.In order to obtain concrete application model in turn from structural model, need with a large amount of sample
Data are to neural network adjusting parameter.In supervised learning situation, backpropagation is a kind of common neural network training method,
I.e. positive to calculate error, error is in the reversed gradient decline of network.
With the rise of artificial neural network, the neural network of backpropagation is widely used in medical treatment & health field, such as
Image procossing, intelligent diagnostics etc., and produce huge social and economic benefit.As the method for data driven type, nerve net
Network is needed using a large amount of medical data, and these medical datas can touch the privacy of patient.Past, medical treatment & health data-privacy
Protection get the brush-off and managerial confusion, although facilitating the acquisition of medical treatment & health data and the training of neural network,
Result in leakage and the abuse of medical treatment & health data.Current people increasingly pay close attention to the safety of medical data, at the same time, right
The demand of more intelligent more accurate medical treatment & health neural network is also more and more urgent.The neural network training method of mainstream is research
Personnel completely get required medical treatment & health data, are then applied to neural network.In the case, how researcher makes
With data, how former data handle and guarantee that safety is all unable to get effective supervision after use, need data safety and training
Seek generation conflict.How while allowing neural network using the training of medical treatment & health data, prevent it from revealing and abusing
Data are a problem to be solved.
Summary of the invention
The purpose of the present invention is to provide a kind of reverse transmittance nerve network training sides based on medical treatment & health data safety
Method.
The technical solution adopted by the present invention is that:
A kind of reverse transmittance nerve network training method based on medical treatment & health data safety comprising following steps:
Step (1) researcher informs that required data, preprocess method and batch parameter are studied by data side；
Data needed for step (2) data side combines simultaneously execute pretreatment operation according to preprocess method；
After step (3) researcher establishes neural network model and black box, data side is sent to together with encrypted network parameter；
The network parameter of encryption input black box is updated black box by step (4) data side, and by the lot data input black box into
Row forward-propagating process obtains error；
After step (5) data side audit error passes through, error and present lot parameter are returned into researcher；
Step (6) researcher judges whether to terminate training process according to error and present lot parameter；It is then, to terminate training；
Otherwise, researcher carries out back-propagation process according to error, updates network parameter and encryption is sent to data side, execute step
(4) further training.
Further, batch parameter includes batch size, whole frequency of training in step (1).
Further, whole frequency of training is no more than both sample number and sample characteristics quantity product 1/10th.
Further, every batch of sample number is not less than ten times of error dimension.
Further, in step (2) preprocess method include data screening, normalization and in batches.
Further, the error precision after normalization is no more than both batch sample number and sample characteristics quantity quantity product
Reciprocal ten times.
Further, the method for black box is that neural network model is encapsulated as to can be performed with the interface UI in step (3)
File, data side run executable file in the virtual machine completely cut off with outer net.
Further, step (3) is encrypted using the method for symmetric cryptography AES, and key one is grasped by researcher,
Two are encapsulated in executable file.
The invention adopts the above technical scheme, and reverse transmittance nerve network training method is total to by researcher and two side of data side
With responsible.It can be seen that researcher need not can not obtain medical treatment & health initial data when neural metwork training.Neural metwork training
When, data side need not can not obtain the structure and parameter of neural network.Medical treatment & health data are always situated in data side, researcher
Have to error information.Error information is integrally obtained by same lot data, individual information of having erased.Error information is by pre-processing
Data obtain afterwards, non-primary data.Present invention efficiently solves the training of reverse transmittance nerve network and medical treatment & health data
The problem of both guarantees of safety contradiction.
Detailed description of the invention
The present invention is described in further details below in conjunction with the drawings and specific embodiments；
Fig. 1 is a kind of process signal of reverse transmittance nerve network training method based on medical treatment & health data safety of the present invention
Figure.
Specific embodiment
As shown in Figure 1, a kind of reverse transmittance nerve network training method based on medical treatment & health data safety of the present invention,
The following steps are included:
Step (1) researcher informs that required data, preprocess method and batch parameter are studied by data side；Batch parameter includes batch
Quantity, whole frequency of training
Data needed for step (2) data side combines simultaneously execute pretreatment operation according to preprocess method；Preprocess method includes number
According to screening, normalization and in batches.
After step (3) researcher establishes neural network model and black box, data are sent to together with the network parameter of encryption
Side；
Specifically, the method for black box is that neural network model is encapsulated as the executable file with the interface UI in step (3), number
Executable file is run in the virtual machine completely cut off with outer net according to side.Further, it is carried out using the method for symmetric cryptography AES
Encryption, key one is grasped by researcher, secondly being encapsulated in executable file.
Network model after black box has the following characteristics that (1) details is not obtained by data side；(2) input parameter can be distinguished
And data；(3) encryption parameter of input is decrypted；(4) network parameter is updated using the parameter after decryption；(5) using defeated
The data entered carry out propagated forward output error.
The network parameter input black box of encryption is updated black box by step (4) data side, and lot data input is black
Case carries out forward-propagating process, obtains error；
After step (5) data side audit error passes through, error and present lot parameter are returned into researcher；
Step (6) researcher judges whether to terminate training process according to error and present lot parameter；It is then, to terminate training；
Otherwise, researcher carries out back-propagation process according to error, updates network parameter and encryption is sent to data side, execute step
(4) further training.
Two stage separation of forward-propagating and backpropagation in technical solution of the present invention, on different devices by two sides
Be each responsible for: forward-propagating process is responsible for by medical treatment & health data providing (hereinafter referred to as data side), and data side has following
Business: (1) data are arranged；(2) black box is operated.And back-propagation process is held by neural network researcher (hereinafter referred to as researcher)
Row, researcher have following task: (1) describing data requirements；(2) neural network black box is established；(3) back-propagation process is executed.
After data side carries out equal pretreatment in batches to sample data, researcher provides current feedforward network to data side,
Data side by present lot data input feedforward network, then obtain the error of certain batch, by audit after, return the error to
New current network is supplied to data side again, then recycled by researcher after researcher's backpropagation updates network, until
Error is sufficiently small or completes the entirety training of specific times.
In back-propagation process, is used in sample data the forward-propagating stage, i.e., handles sample using current network
Notebook data simultaneously obtains batch error, and back-propagation process then only uses the error and calculated.Under normal circumstances, after obtaining data
The two stages are all responsible for progress by neural network researcher.
Researcher can only obtain batch error every time, and cannot obtain initial data, as a result, this invention ensures that original number
According to being retained under the control of data side without being leaked.At the same time, the feedforward network that data side obtains also is a black box,
It only needs also audit error, ensure that the structure and parameter of neural network is not learnt by data side.In addition, in order to prevent
Data are revealed from error, and technical method of the invention has following points for attention, and (1) entirety frequency of training should should not excessively be no more than
Both sample number and sample characteristics quantity product 1/10th；(2) every batch of sample number, which is not answered, very few should be not less than error dimension
Ten times；(3) error precision is not answered excessively high, and the error precision after normalization is no more than batch sample number and sample characteristics quantity
Reciprocal ten times of the two quantity product.
Reverse transmittance nerve network training method of the invention is collectively responsible for by researcher and two side of data side.It can be seen that nerve
When network training, researcher need not can not obtain medical treatment & health initial data.When neural metwork training, data side need not yet
The structure and parameter of neural network can be obtained.Medical treatment & health data have to error information always situated in data side, researcher.Accidentally
Difference data is integrally obtained by same lot data, individual information of having erased.Error information is obtained by data after pre-processing, non-primary
Data.
Present invention efficiently solves both the training of reverse transmittance nerve network and the guarantee of medical treatment & health data safety
The problem of contradiction.The present invention is different from multi-party cooperative research Zhong Ge partner functional equivalence while as researcher and data
Side, cloud platform is provided as auxiliary to be calculated power support and summarizes the situation of service, application service provider of the invention and data center
In two sides cooperation, the two function is not reciprocity, and application service provider is only used as researcher, and data center is only used as data side, the present invention
Two sides cooperation in training pattern result as the trade secret of application service provider be need for confidentiality, ground rather than multi-party cooperative
Training pattern in situation is studied carefully the result is that shared.
Claims (8)
1. a kind of reverse transmittance nerve network training method based on medical treatment & health data safety, it is characterised in that: it include with
Lower step:
Step (1) researcher informs that required data, preprocess method and batch parameter are studied by data side；
Data needed for step (2) data side combines simultaneously execute pretreatment operation according to preprocess method；
After step (3) researcher establishes neural network model and black box, data side is sent to together with the network parameter of encryption；
The network parameter of encryption input black box is updated black box by step (4) data side, and by the lot data input black box into
Row forward-propagating process obtains error；
After step (5) data side audit error passes through, error and present lot parameter are returned into researcher；
Step (6) researcher judges whether to terminate training process according to error and present lot parameter；It is then, to terminate training；
Otherwise, researcher carries out back-propagation process according to error, updates network parameter and encryption is sent to data side, execute step
(4) further training.
2. a kind of reverse transmittance nerve network training method based on medical treatment & health data safety according to claim 1,
It is characterized by: batch parameter includes batch size, whole frequency of training in step (1).
3. a kind of reverse transmittance nerve network training method based on medical treatment & health data safety according to claim 2,
It is characterized by: whole frequency of training is no more than both sample number and sample characteristics quantity product 1/10th.
4. a kind of reverse transmittance nerve network training method based on medical treatment & health data safety according to claim 2,
It is characterized by: every batch of sample number is not less than ten times of error dimension.
5. a kind of reverse transmittance nerve network training method based on medical treatment & health data safety according to claim 1,
It is characterized by: in step (2) preprocess method include data screening, normalization and in batches.
6. a kind of reverse transmittance nerve network training method based on medical treatment & health data safety according to claim 5,
It is characterized by: the error precision after normalization is no more than the inverse of both batch sample number and sample characteristics quantity quantity product
Ten times.
7. a kind of reverse transmittance nerve network training method based on medical treatment & health data safety according to claim 1,
It is characterized by: the method for black box is that neural network model is encapsulated as the executable file with the interface UI in step (3), number
Executable file is run in the virtual machine completely cut off with outer net according to side.
8. a kind of reverse transmittance nerve network training method based on medical treatment & health data safety according to claim 1,
It is characterized by: step (3) is encrypted using the method for symmetric cryptography AES, key one is grasped by researcher, secondly encapsulation
In executable file.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910535719.8A CN110263919A (en) | 2019-06-20 | 2019-06-20 | A kind of reverse transmittance nerve network training method based on medical treatment & health data safety |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910535719.8A CN110263919A (en) | 2019-06-20 | 2019-06-20 | A kind of reverse transmittance nerve network training method based on medical treatment & health data safety |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110263919A true CN110263919A (en) | 2019-09-20 |
Family
ID=67919768
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910535719.8A Pending CN110263919A (en) | 2019-06-20 | 2019-06-20 | A kind of reverse transmittance nerve network training method based on medical treatment & health data safety |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110263919A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160350648A1 (en) * | 2014-11-07 | 2016-12-01 | Microsoft Technology Licensing, Llc. | Neural networks for encrypted data |
| CN108520181A (en) * | 2018-03-26 | 2018-09-11 | 联想(北京)有限公司 | data model training method and device |
| CN108920981A (en) * | 2018-07-06 | 2018-11-30 | 成都大象分形智能科技有限公司 | It is related to the neural network model encryption protection system and method for data iterative cryptographic |
| CN109033854A (en) * | 2018-07-17 | 2018-12-18 | 阿里巴巴集团控股有限公司 | Prediction technique and device based on model |
| CN109325584A (en) * | 2018-08-10 | 2019-02-12 | 深圳前海微众银行股份有限公司 | Federation's modeling method, equipment and readable storage medium storing program for executing neural network based |
| CN109684855A (en) * | 2018-12-17 | 2019-04-26 | 电子科技大学 | A kind of combined depth learning training method based on secret protection technology |
| CN109685501A (en) * | 2018-12-04 | 2019-04-26 | 暨南大学 | Based on secret protection deep learning platform construction method auditable under block chain incentive mechanism |
-
2019
- 2019-06-20 CN CN201910535719.8A patent/CN110263919A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160350648A1 (en) * | 2014-11-07 | 2016-12-01 | Microsoft Technology Licensing, Llc. | Neural networks for encrypted data |
| CN108520181A (en) * | 2018-03-26 | 2018-09-11 | 联想(北京)有限公司 | data model training method and device |
| CN108920981A (en) * | 2018-07-06 | 2018-11-30 | 成都大象分形智能科技有限公司 | It is related to the neural network model encryption protection system and method for data iterative cryptographic |
| CN109033854A (en) * | 2018-07-17 | 2018-12-18 | 阿里巴巴集团控股有限公司 | Prediction technique and device based on model |
| CN109325584A (en) * | 2018-08-10 | 2019-02-12 | 深圳前海微众银行股份有限公司 | Federation's modeling method, equipment and readable storage medium storing program for executing neural network based |
| CN109685501A (en) * | 2018-12-04 | 2019-04-26 | 暨南大学 | Based on secret protection deep learning platform construction method auditable under block chain incentive mechanism |
| CN109684855A (en) * | 2018-12-17 | 2019-04-26 | 电子科技大学 | A kind of combined depth learning training method based on secret protection technology |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Naidu et al. | A Robust medical data sharing application with a collaborative hypothesis and Elastic net regression model | |
| US11947680B2 (en) | Model parameter training method, terminal, and system based on federation learning, and medium | |
| Bali et al. | Cognizant multitasking in multiobjective multifactorial evolution: MO-MFEA-II | |
| CN109284313A (en) | Federal modeling method, equipment and readable storage medium storing program for executing based on semi-supervised learning | |
| Abbass | An evolutionary artificial neural networks approach for breast cancer diagnosis | |
| US20190199733A1 (en) | Artificial intelligence data processing system and method | |
| Huizinga et al. | Evolving neural networks that are both modular and regular: Hyperneat plus the connection cost technique | |
| Cai et al. | A many-objective optimization based federal deep generation model for enhancing data processing capability in IoT | |
| Chen et al. | A quantum treatment of public goods economics | |
| Khanum et al. | Two new improved variants of grey wolf optimizer for unconstrained optimization | |
| Bauer et al. | Nonlinear growth: an origin of hub organization in complex networks | |
| CN114372871A (en) | Method and device for determining credit score value, electronic device and storage medium | |
| Ma | Application of artificial intelligence in computer network technology | |
| CN114168988A (en) | Method for aggregating federated learning models and electronic device | |
| Ali et al. | Blockchain for IoT-based cyber-physical systems (CPS): applications and challenges | |
| Pio-Lopez et al. | The scaling of goals from cellular to anatomical homeostasis: an evolutionary simulation, experiment and analysis | |
| Salam et al. | Implementation of Artificial Intelligence in Governance: Potentials and Challenges | |
| CN118247098A (en) | Virtual digital person-driven personalized online training system | |
| Yamamura et al. | Geometric landscape annealing as an optimization principle underlying the coherent ising machine | |
| CN111680798A (en) | Joint learning model system and method, apparatus, and computer-readable storage medium | |
| Pentland et al. | Human-AI decision systems | |
| Mouret et al. | Importing the computational neuroscience toolbox into neuro-evolution-application to basal ganglia | |
| CN110263919A (en) | A kind of reverse transmittance nerve network training method based on medical treatment & health data safety | |
| Mohana et al. | Heuristics for privacy preserving data mining: An evaluation | |
| US20220067621A1 (en) | Artificial intelligence data processing system and method |
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 | ||
| CB03 | Change of inventor or designer information |
Inventor after: Hai Rong Lu Inventor after: Zhou Rongchen Inventor after: Li Yan Inventor before: Hai Rong Lu Inventor before: Zhou Rongchen |
|
| CB03 | Change of inventor or designer information | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190920 |
|
| WD01 | Invention patent application deemed withdrawn after publication |