CN104992188B - A kind of distributed Handwritten Digit Recognition method based on the analysis of t hybrid cytokines - Google Patents
A kind of distributed Handwritten Digit Recognition method based on the analysis of t hybrid cytokines Download PDFInfo
- Publication number
- CN104992188B CN104992188B CN201510415750.XA CN201510415750A CN104992188B CN 104992188 B CN104992188 B CN 104992188B CN 201510415750 A CN201510415750 A CN 201510415750A CN 104992188 B CN104992188 B CN 104992188B
- Authority
- CN
- China
- Prior art keywords
- node
- data
- tmfa
- iteration
- training
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Abstract
The invention discloses a kind of distributed Handwritten Digit Recognition method of t hybrid cytokines analysis, this method carries out feature extraction first at each node to the handwritten numeral collected, then to being initialized for the characteristic corresponding to trained each numeral, then each node calculates local statistic based on the training data of itself and is broadcast to its neighbor node, at the same time, each node is according to the received local statistic from all neighbor nodes, calculate joint statistic, and the parameters in the analysis of t hybrid cytokines are estimated based on the joint statistic, complete distributed training process.In the Distributed identification stage, the data of test can input any node, calculate its log-likelihood on the corresponding tMFA of each trained numeral, using the corresponding numeral of max log likelihood value as recognition result.There is higher robustness to the outlier in data using tMFA, using distributed training and identification method, avoid the periods of network disruption brought by Centroid.
Description
Technical field
The present invention relates to a kind of distributed Handwritten Digit Recognition method based on the analysis of t hybrid cytokines, belong to data and
Row distributed approach and the technical field of application.
Background technology
At present, Handwritten Digit Recognition is an extremely challenging problem in pattern-recognition category, is mainly studied
It is to enable a computer to the numeral that independently identification human hand is write.Handwritten Digit Recognition has application at many aspects, such as：For big
The various aspects such as scale data statistics, finance, the tax, finance and sorting mail are widely used.Have at present a variety of
Method realizes Handwritten Digit Recognition.But when data volume is larger, single computer can not be handled fast and effectively, because
This needs handwritten numeral data being divided into some, is respectively stored on different computers, between designing a calculating machine
Collaboration method, it is achieved thereby that the distributed treatment of data.
In pattern-recognition, machine learning field, Handwritten Digit Recognition belongs to the category of supervised learning.In the training stage,
Based on markd sample, foundation and training can describe the model of the distribution of handwritten numeral characteristic, in ensuing knowledge
The other stage, by numerical characteristic data to be identified and it is trained represent all kinds of characteristic models and be compared, most closed so as to find
Suitable model, as final recognition result.In the scene for needing to carry out distributed treatment, in the training stage, how to pass through
Cooperation between computer node so that each computer node can make full use of the data of other each computer nodes, so that
Estimate consistent model；In cognitive phase, if handwritten numeral data to be identified can be inputted any one node, all may be used
It is very crucial to obtain correct recognition result.Method proposed by the present invention can efficiently solve the above problem, obtain fine
Recognition correct rate.
The content of the invention
Present invention aims at solve the defects of prior art, it is proposed that one kind in sensor network based on mixing because
The distributed clustering method of sub- analysis model.
The technical scheme adopted by the invention to solve the technical problem is that：A kind of distribution based on the analysis of t hybrid cytokines
Handwritten Digit Recognition method, this method comprises the following steps：
Step 1：The collection and feature extraction of data；
Equipped with M platforms computer/calculate node (i.e.：Node), network is formed, m-th of node is by being attached thereto
Handwriting pad is collected from numeral 0~9, and the initial data of totally 10 classes, handwriting pad record on each character writing track automatically
The two-dimensional coordinate position of each point, the coordinate of 8 points is taken in the first-class compartment of terrain in trackAs each initial data institute
Corresponding characteristic s, 16 is tieed up totally.In order to represent convenient, if the digital d for gathering at node m and being obtained by feature extraction
Training dataset isWhereinRepresent at node m, for trained handwritten numeral
N-th of characteristic of d, dimension p,For the training data number of digital d.
Described with t hybrid cytokines analysis model (tMFA)Distribution, it is noted that institute
Have and modeled at node on the public same tMFA of training data of digital d.TMFA is the hybrid guided mode that a component number is I
Type；For each dataIt can be expressed as：
With probability πi(i=1 ..., I),
Wherein, μiFor the mean value vector of i-th of blending constituent, dimension p；For withCorresponding lower dimensional space
In the factor, its dimension is q (q ＜ ＜ p), obey obey t distributionThe value of q is according to p in particular problem
Size chosen, generally take the arbitrary integer between q=p/5~p/3；AiFor i-th of blending constituent (p × q) because
Sub- loading matrix；Error varianceObey t distributionsWherein DiFor the diagonal matrix of (p × p), νiFor
The free degree of i blending constituent；The weight π of each blending constituentiMeetThe parameter sets Θ of so tMFA
For { πi,Ai,μi,Di,νi}I=1 ..., I.Note that for all nodes, parameters value in its tMFA parameter sets to be estimated
It is identical.Need exist for explanation is the integration that t distributions can be launched into Gaussian Profile and Gamma distributions：
Wherein,Be withCorresponding integration hidden variable.
In addition, the data transmission range of each node is set to Dis, it is all to be less than Dis with its distance for present node m
Node be its neighbor node, the neighbor node set expression of node m is Rm.Illustrated in Fig. 1 in some network each node it
Between relation, wherein computer icon representation node, if thering is side to be connected between two nodes, then it represents that can be between two nodes
Communicate, transmit information.Dotted line frame in Fig. 1 represents the R of the m of nodem.In the present invention, network topology is determined in advance,
Only need to ensure that at least there are a path that is direct or being reached through multi-hop between any two node.
Step 2：Distribution training, willFor distribution training, every class numeral is obtained
TMFA parameter sets corresponding to d
After the tMFA of network topology and description data distribution is established, then start distributed training.Here with digital d's
Exemplified by training, training process is as shown in Fig. 2, it is comprised the following steps that：
Step 2-1：Initialization；Fraction I is mixed into setting tMFA.Here I determines the complexity of tMFA models,
I can take the arbitrary integer in 3~8, take I=5 to obtain preferable performance in Handwritten Digit Recognition.According to I and data
Dimension p set the initial value of parameter in MFA.Wherein, at each nodeSelected at random in the data collected from the node
Take；WithIn the generation from the standardized normal distribution N (0,1) of each element；This group of parameter takes the arbitrary integer between 1~5.In addition, each node l (l=1 ..., M) by its
The data amount check collectedIt is broadcast to its neighbor node.When some node m receives what its all neighbor nodes broadcast came
After data amount check, which calculates weight clm：
The implication of the weight is each neighbor node l (l ∈ R for weighing node mm) information transmitted every time is in node m
The importance at place.After the completion of initialization, iteration count iter=1, starts iterative process.
Step 2-2：Calculate local statistic and broadcast；The information of neighbor node is not required in this step.In each node l
Place, the data collected based on itG is calculated firsti, Ωi,With
WhereinThe parameter value obtained afterwards is completed for preceding an iteration (i.e.：First
It is the initial value of parameter during iteration Represent nth data at node lBelong to i-th
A class is (i.e.：Blending constituent) probability,ForDesired value.
There is the result of calculation of above-mentioned variable, node calculates local statistic (LS)It is as follows：
Finally, the local statistic LS that each node l will be calculatedlIts neighbor node is given in broadcast diffusion, as shown in Figure 1.
Step 2-3：Calculate joint statistic；When node m (m=1 ..., M) is received from its all neighbor node l (l ∈
Rm), node m calculates joint statistic
Step 2-4：Estimate each parameter in model；The CS that node m (m=1 ..., M) is calculated according to previous stepm, estimate
Count out each parameter Θ={ πi,Ai,μi,Di,νi}I=1 ..., I, wherein, { πi,μi}I=1 ..., IEstimation procedure it is as follows：
For { Ai,Di}I=1 ..., IEstimation, process is as follows：
In addition, for { νi}I=1 ..., I, obtained by solving following equation：
Wherein ψ () is the digamma functions of standard, and specific solve uses Newton method.
Step 2-5：Judge whether to restrain；Node m (m=1 ..., M) calculates the log-likelihood under current iteration iter：
IfThen algorithmic statement, stops iteration；Otherwise step 2- is performed
2, start next iteration (iter=iter+1；).Wherein Θ represents the parameter value that current iteration estimates, ΘoldRepresent upper one
The parameter value estimated in secondary iteration.That is, the log-likelihood of adjacent iteration twice is less than threshold epsilon, algorithmic statement.ε takes 10-5~
10-6In arbitrary value.Significantly, since each node is parallel data processing in network, thus it is all can not possibly be one
Restrained at the same time in secondary iteration.For example, when node l has restrained and node m not yet restrains, then node l does not retransmit LSl,
No longer receive the information of neighbor node transmission.The LS that node m is then sent with the received node l of last timelUpdate its CSm.Do not receive
The node held back continues iteration, until all nodes are all restrained in network.
After above-mentioned steps 2-1~step 2-5, the correspondence model tMFA that is obtained by the training data of handwritten numeral d
(i.e.：Parameter Θ when being restrained with training is represented).Repeat the above steps 10 times, so as to obtain 10 corresponding tMFA of numeral
Model, in order to represent convenient, and is distinguish between, usesRepresent
The corresponding tMFA models of digital d.Distribution training is completed.
Step 3：Distributed identification；When any computer in network collects the new handwritten numeral for being used for identification,
Its character pair is obtained by step (1) first, is expressed as s', is then calculated on Θ(d)The log-likelihood of (d=0,1 ..., 9)
logp(s'|Θ(d)) (d=0,1 ..., 9)：
Recognition result d' using the corresponding sequence number of max log likelihood value as s'：
The idiographic flow of the Distributed identification method of the present invention is as shown in Figure 2.
Beneficial effect：
1. the t hybrid cytokines analysis employed in the present invention, has higher robustness to outlier present in data,
And high dimensional data can preferably be described, so that model corresponding with data is preferably obtained, so as to can also obtain more preferable
Training and recognition performance.
2. the distributed training process based on t hybrid cytokine analysis models employed in the present invention so that in network
Each computer node can make full use of information included in the data that other computer nodes collect, so that at training
More accurate model.
3. the distributed training process based on t hybrid cytokine analysis models employed in the present invention, in computer node
In cooperating process, exchange local statistic rather than directly transmit initial data, since the quantity and dimension of local statistic are remote
Less than data, therefore on the one hand this mode saves the expense of communication, on the other hand, hidden in the data that are conducive to adequately protect
Personal letter ceases, and improves the security performance of the system using this method.
4. the Distributed identification process based on t hybrid cytokine analysis models employed in the present invention, can be in a network
Any one computer node at gather new data, identical recognition result can be obtained.
Brief description of the drawings
Fig. 1 is the neighbor node collection R of the nodes m of the present inventionm, and showing for local statistic is received and dispatched between node
It is intended to.
Fig. 2 is the flow chart of the distributed Handwritten Digit Recognition method of the present invention based on the analysis of t hybrid cytokines.
Fig. 3 is the method and centralization tMFA of the present invention, the corresponding confusion matrix of no cooperation tMFA method recognition results
(i.e.：Confusion Matrix) hinton schematic diagrames.
Fig. 4 be the present invention method and centralization tMFA, the average and variance of the recognition correct rate of no cooperation tMFA methods
Schematic diagram.
Embodiment
The invention is described in further detail with reference to Figure of description.
In order to which a kind of distributed Handwritten Digit Recognition side based on the analysis of t hybrid cytokines of the present invention is better described
Method.We are described with specific application example.
(1) collection and feature extraction of data：If a total of 44 people, everyone each digital handwriting 25 times, is total up to 25*
10*44=11000 initial data.20 computer/nodes (M=20) altogether in network, the neighbor node of each node
Number is 3, there is direct or multihop path intercommunication between any two node.By the initial data of the wherein handwritten numeral of 30 people
(250*30=7500) are used for distributed training, are averaged and are divided into 20 parts, are assigned randomly in 20 nodes.For each
Initial data, equally spaced takes the coordinate of 8 points on its trackAs the characteristic corresponding to the initial data
According to s, totally 16 tie up.In order to represent convenient, if the training dataset of the digital d obtained at node m by feature extraction isWhereinRepresent at node m, n-th of feature for trained handwritten numeral d
Data, dimension p=16,For the training data number of digital d.
(2) distributed training：After the completion of step (1), start distributed training.Here by taking the training of digital d as an example, use
TMFA carrys out the distribution of Modelling feature dataTraining process is as shown in Fig. 2, it is comprised the following steps that：
(2-1) is initialized：Fraction I=5 is mixed into setting tMFA.Set according to I and the dimension p of data in MFA
The initial value of parameter.Wherein, at each nodeFrom
Randomly selected in the data that the node collects；WithIn each element from
Generation in standardized normal distribution N (0,1)；In addition, each node l (l=1 ..., M) by its
The data amount check collectedIt is broadcast to its neighbor node.The number of coming is broadcasted when some node m receives its all neighbor nodes
After number, which calculates weight clm：
The implication of the weight is each neighbor node l (l ∈ R for weighing node mm) information transmitted every time is in node m
The importance at place.After the completion of initialization, iteration count iter=1, starts iterative process.
(2-2) calculates local statistic and broadcasts：The information of neighbor node is not required in this step.At each node l, base
In the data that it is collectedIntermediate variable g is calculated firsti, Ωi,With
WhereinParameter value (the iteration first for completing to obtain afterwards for last iteration
When for parameter initial value Represent nth data at node lBelong to i-th of class
The probability of (blending constituent),ForDesired value.
There is the result of calculation of above-mentioned intermediate variable, node calculates local statistic (LS)It is as follows：
Finally, the local statistic LS that each node l will be calculatedlIts neighbor node is given in broadcast diffusion, as shown in Figure 1.
(2-3) calculates joint statistic：When node m (m=1 ..., M) is received from its all neighbor node l (l ∈
Rm), node m calculates joint statistic
Each parameter in (2-4) estimation model：The CS that node m (m=1 ..., M) is calculated according to previous stepm, estimate
Θ={ πi,Ai,μi,Di,νi}I=1 ..., I, wherein, { πi,μi}I=1 ..., IEstimation procedure it is as follows：
For { Ai,Di}I=1 ..., IEstimation, process is as follows：
In addition, for { νi}I=1 ..., I, obtained by solving following equation：
Wherein ψ () is the digamma functions of standard, generally with the above-mentioned equation of Newton method solution.
(2-5) judges whether to restrain：Node m (m=1 ..., M) calculates the log-likelihood under current iteration iter：
IfThen algorithmic statement, stops iteration；Otherwise step (2- is performed
2) next iteration (iter=iter+1, is started；).Wherein Θ represents the parameter value that current iteration estimates, ΘoldIn expression
The parameter value estimated in an iteration.That is, the log-likelihood of adjacent iteration twice is less than threshold epsilon, algorithmic statement.ε takes 10-5~
10-6In arbitrary value.Significantly, since each node is parallel data processing in network, thus it is all can not possibly be one
Restrained at the same time in secondary iteration.For example, when node l has restrained and node m not yet restrains, then node l does not retransmit LSl,
No longer receive the information of neighbor node transmission.The LS that node m is then sent with the received node l of last timelUpdate its CSm.Do not receive
The node held back continues iteration, until all nodes are all restrained in network.
After above-mentioned steps (2-1)~(2-5), the correspondence model tMFA that is obtained by the training data of handwritten numeral d
(parameter Θ when being restrained with training is represented).Repeat the above steps 10 times, so as to obtain the corresponding tMFA moulds of 10 numerals
Type, in order to represent convenient, and is distinguish between, usesRepresent number
The corresponding tMFA models of word d.Distribution training is completed.
(3) Distributed identification：When any computer in network collects the new handwritten numeral for being used for identification, first
Its character pair is obtained by step (1), s' is expressed as, then calculates on Θ(d)The log-likelihood logp of (d=0,1 ..., 9)
(s'|Θ(d)) (d=0,1 ..., 9)：
Recognition result d' using the corresponding sequence number of max log likelihood value as s'：
The Distributed identification flow of the present invention is as shown in Figure 2.
Performance evaluation：
Due to digital realistic value to be identified it is known that recognition methods according to the present invention and its true value will be used to be compared
Compared with obtaining recognition correct rate (i.e.：The quantity for the handwritten numeral that recognition correct rate=all nodes correctly identify/(20*3500)),
So as to evaluate and weigh out the validity of method according to the present invention and accuracy.For base more proposed by the present invention
In the distributed Handwritten Digit Recognition method (referred to as distribution tMFA) of tMFA and the performance of other methods, here and based on tMFA
Centralized Handwritten Digit Recognition method (referred to as centralization tMFA), the handwritten numeral knowledge without cooperation between each node based on tMFA
Other method is compared (referred to as without cooperation tMFA).It should be noted that in centralized tMFA, all nodes need will be original
Data are transferred to some Centroid, are completed training using traditional MFA by Centroid and are identified, then result is returned again
It is transmitted to each node, this mode in practice seldom, first, transmission raw data communication expense is very big, once there is packet loss
Or data packet damage, very big on last recognition performance influence, two are detrimental to the secret protection in data, and network security may
Sorrow.The object here is in order to which whether the recognition methods of distribution tMFA more proposed by the present invention can reach centralized tMFA
Same performance.Recognition result is represented with qualitative and quantitative two ways respectively.In the qualitative representation of result, using obscuring
The hinton figures of matrix, as shown in Figure 3.In the figure, the recognition result of each list registration word 0~9 and each row represent numeral 0~
9 true value.Blockage on leading diagonal represents situation about correctly identifying, the size of blockage shows more greatly the correct identification
Numeral is more, and occurs blockage in other positions and show there is a situation where wrong identification.It is it can be seen from this figure that centralized
Distributed tMFA (only providing node 3 therein as space is limited, other node results are identical) performance of tMFA and the present invention
Preferably, and without cooperation tMFA poor-performings.In the quantificational expression of result, using the average and variance two indices of discrimination,
As shown in Figure 4.In the figure, the recognition correct rate of the recognition correct rate for the distributed tMFA that the present invention designs and centralization tMFA
Average it is essentially identical, and without cooperation tMFA it is poor, the variance of the recognition correct rate of distribution tMFA is also much smaller than nothing in addition
Cooperate tMFA.Therefore, method using the present invention overcomes the shortcomings that Handwritten Digit Recognition of traditional centralized tMFA, realizes
Distributed identification and there is good performance.
The claimed scope of the present invention is not limited only to the description of present embodiment.
Claims (2)
- A kind of 1. distributed Handwritten Digit Recognition method based on the analysis of t hybrid cytokines, it is characterised in that the described method includes such as Lower step：Step 1：The collection and feature extraction of data：Equipped with M platform computers, i.e.,：Node, forms a network, the topology of network Be determined in advance, if its meet to exist between any two node directly or multi-hop transmission and intercommunication path, node m Neighbor node set expression be Rm；M-th of node, m=1 ..., M collect handwritten numeral by the handwriting pad being attached thereto 0~9, the initial data of totally 10 classes, handwriting pad record the two-dimensional coordinate position of each point on each character writing track automatically, The coordinate of 8 points is taken in the first-class compartment of terrain in trackAs the characteristic s corresponding to each initial data, totally 16 Dimension；If the training dataset for the digital d for gathering at node m and being obtained by feature extraction is WhereinRepresent at node m, for n-th of characteristic of trained handwritten numeral d, dimension p=16,For digital d Training data number；With a public t hybrid cytokines analysis tMFA come describe in all nodes with the relevant characteristic data sets of digital dDistribution；The parameter sets of tMFA are { πi,Ai,μi,Di,νi}I=1 ..., I, wherein I is to be mixed into fraction, πiFor the weight of i-th of blending constituent, AiFor the Factor load-matrix of (p × q) of i-th of blending constituent, q is the low-dimensional factor Dimension, takes the arbitrary integer between q=p/5~p/3, μiMean value vector, D are tieed up for the p of i-th of blending constituentiFor i-th of mixing The covariance matrix of (p × p) of the error of component；νiFor the free degree of i-th of blending constituent；Step 2：Distribution training, willFor distribution training, it is right to obtain every class numeral d institutes The tMFA parameter sets answeredStep 3：Distributed identification, it is first when any one node in network collects the new handwritten numeral for being used for identification Its character pair first is obtained by above-mentioned steps 1, is expressed as s', then calculates s' on Θ(d), the logarithm of d=0,1 ..., 9 is seemingly Right logp (s'| Θ(d)), d=0,1 ..., 9：Recognition result d' using the corresponding sequence number of max log likelihood value as s'：
- 2. a kind of distributed Handwritten Digit Recognition method based on the analysis of t hybrid cytokines according to claim 1, its feature It is, the step 2 includes the following steps：Step 2-1：Initialization；Fraction I is mixed into setting tMFA, the initial of each parameter in MFA is set according to I, p and q ValueWherein, at each node Randomly selected in the data collected from the node；WithIn each element from Generation in standardized normal distribution N (0,1)；This group of parameter takes the arbitrary integer between 1~5；In addition, The data amount check that each node l, l=1 ..., M are collectedIt is broadcast to its neighbor node；When some node m is received After the data amount check that its all neighbor nodes broadcast comes, which calculates weight clm：After the completion of initialization, iteration count iter=1, starts iterative process；Step 2-2：Calculate local statistic and broadcast；At each node l, the data that are collected based on itCalculate first Go out intermediate variable gi, Ωi,WithWhereinThe parameter value for completing to obtain afterwards for last iteration, i.e.,：First during iteration For the initial value of parameter Represent nth data at node lBelong to i-th of class, i.e.,： The probability of blending constituent,For the hidden variable in tMFADesired value；There is the result of calculation of above-mentioned intermediate variable, node calculates local statistic, i.e. LS,Including：Finally, the local statistic LS that each node l will be calculatedlIts neighbor node is given in broadcast diffusion；Step 2-3：Calculate joint statistic；When node m, m=1 ..., M are received from its all neighbor node l (l ∈ Rm) LSlAfterwards, node m calculates joint statisticStep 2-4：Estimate each parameter in model；The CS that node m, m=1 ..., M are calculated according to previous stepm, estimate Θ ={ πi,Ai,μi,Di,νi}I=1 ..., I, wherein, { πi,μi}I=1 ..., IEstimation procedure include：For { Ai,Di}I=1 ..., IEstimation, process includes：In addition, for { νi}I=1 ..., I, obtained by solving following equation, including：Wherein ψ () is the digamma functions of standard, and Newton method is used when specifically solving；Step 2-5：Judge whether to restrain；Node m (m=1 ..., M) calculates the log-likelihood under current iteration：IfThen algorithmic statement, stops iteration；Otherwise step 2-2 is performed, is started Next iteration (iter=iter+1)；Wherein Θ represents the parameter value that current iteration estimates, ΘoldRepresent last iteration The parameter value of middle estimation；That is, the log-likelihood of adjacent iteration twice is less than threshold epsilon, algorithmic statement；ε takes 10-5~10-6In Arbitrary value；It is all to be restrained at the same time in an iteration since each node is parallel data processing in network；Work as section Point l has restrained and when node m not yet restrains, then node l does not retransmit LSl, also no longer receive the information that neighbor node transmits； The LS that node m is then sent with the received node l of last timelUpdate its CSm；Not converged node continues iteration, until in network All nodes are all restrained；After above-mentioned steps 2-1~step 2-5, the correspondence model tMFA that is obtained by the training data of handwritten numeral d, i.e.,： Parameter Θ when being restrained with training is represented；Repeat the above steps 10 times, so that the corresponding tMFA models of 10 numerals are obtained, In order to represent convenient, and it is distinguish between, usesRepresent numeral d Corresponding tMFA models, so far, distribution training are completed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510415750.XA CN104992188B (en) | 2015-07-15 | 2015-07-15 | A kind of distributed Handwritten Digit Recognition method based on the analysis of t hybrid cytokines |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510415750.XA CN104992188B (en) | 2015-07-15 | 2015-07-15 | A kind of distributed Handwritten Digit Recognition method based on the analysis of t hybrid cytokines |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104992188A CN104992188A (en) | 2015-10-21 |
| CN104992188B true CN104992188B (en) | 2018-04-20 |
Family
ID=54304001
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510415750.XA Expired - Fee Related CN104992188B (en) | 2015-07-15 | 2015-07-15 | A kind of distributed Handwritten Digit Recognition method based on the analysis of t hybrid cytokines |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104992188B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1604121A (en) * | 2003-09-29 | 2005-04-06 | 阿尔卡特公司 | Method, system, client, server for distributed handwriting recognition |
| EP2515257A1 (en) * | 2009-12-15 | 2012-10-24 | Fujitsu Frontech Limited | Character recognition method, character recognition device, and character recognition program |
-
2015
- 2015-07-15 CN CN201510415750.XA patent/CN104992188B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1604121A (en) * | 2003-09-29 | 2005-04-06 | 阿尔卡特公司 | Method, system, client, server for distributed handwriting recognition |
| EP2515257A1 (en) * | 2009-12-15 | 2012-10-24 | Fujitsu Frontech Limited | Character recognition method, character recognition device, and character recognition program |
Non-Patent Citations (2)
| Title |
|---|
| An efficient ECM algorithm for maximum likelihood estimation in mixtures of t-factor analyzers;Wang W L等;《ACM》;20130430;全文 * |
| 基于统计和结构特征的手写数字识别研究;双小川等;《计算机工程与设计》;20120430;第33卷(第4期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104992188A (en) | 2015-10-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103945533B (en) | Wireless real time position localization methods based on big data | |
| CN109743683B (en) | Method for determining position of mobile phone user by adopting deep learning fusion network model | |
| US20210334604A1 (en) | Facial recognition method and apparatus | |
| CN109492076B (en) | Community question-answer website answer credible evaluation method based on network | |
| CN104992191A (en) | Image classification method based on deep learning feature and maximum confidence path | |
| CN110880019A (en) | Method for adaptively training target domain classification model through unsupervised domain | |
| CN109543939A (en) | A kind of method of green building productions certification risk evaluation model building | |
| CN109829072A (en) | Construct atlas calculation and relevant apparatus | |
| US11893773B2 (en) | Finger vein comparison method, computer equipment, and storage medium | |
| CN115577858B (en) | Block chain-based carbon emission prediction method and device and electronic equipment | |
| CN110457677A (en) | Entity-relationship recognition method and device, storage medium, computer equipment | |
| CN107682344A (en) | A kind of ID collection of illustrative plates method for building up based on DPI data interconnection net identifications | |
| CN108564040A (en) | A kind of fingerprint activity test method based on depth convolution feature | |
| CN106503631A (en) | A kind of population analysis method and computer equipment | |
| CN108183956B (en) | Method for extracting key path of propagation network | |
| CN110046941A (en) | A kind of face identification method, system and electronic equipment and storage medium | |
| CN109783805A (en) | A kind of network community user recognition methods and device | |
| CN108229692B (en) | Machine learning identification method based on dual contrast learning | |
| CN104992188B (en) | A kind of distributed Handwritten Digit Recognition method based on the analysis of t hybrid cytokines | |
| CN109361529A (en) | A kind of terminal type recognition methods, device, electronic equipment and storage medium | |
| CN104463896A (en) | Image corner point detection method and system based on kernel similar region distribution characteristics | |
| CN109101984A (en) | A kind of image-recognizing method and device based on convolutional neural networks | |
| CN111861038A (en) | College entrance examination score prediction method and system based on machine learning algorithm | |
| CN111539390A (en) | Small target image identification method, equipment and system based on Yolov3 | |
| CN110443323A (en) | Appearance appraisal procedure based on shot and long term memory network and face key point |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20201201 Address after: Room 214, building D5, No. 9, Kechuang Avenue, Zhongshan Science and Technology Park, Jiangbei new district, Nanjing, Jiangsu Province Patentee after: Nanjing Tian Gu Information Technology Co.,Ltd. Address before: Yuen Road Qixia District of Nanjing City, Jiangsu Province, No. 9 210023 Patentee before: NANJING University OF POSTS AND TELECOMMUNICATIONS Effective date of registration: 20201201 Address after: Gulou District of Nanjing City, Jiangsu Province, Beijing Road No. 20 210024 Patentee after: STATE GRID JIANGSU ELECTRIC POWER Co.,Ltd. INFORMATION & TELECOMMUNICATION BRANCH Address before: Room 214, building D5, No. 9, Kechuang Avenue, Zhongshan Science and Technology Park, Jiangbei new district, Nanjing, Jiangsu Province Patentee before: Nanjing Tian Gu Information Technology Co.,Ltd. |
|
| TR01 | Transfer of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180420 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |