CN104103060B - The method for expressing and equipment of dictionary in a kind of sparse model - Google Patents
The method for expressing and equipment of dictionary in a kind of sparse model Download PDFInfo
- Publication number
- CN104103060B CN104103060B CN201310115751.3A CN201310115751A CN104103060B CN 104103060 B CN104103060 B CN 104103060B CN 201310115751 A CN201310115751 A CN 201310115751A CN 104103060 B CN104103060 B CN 104103060B
- Authority
- CN
- China
- Prior art keywords
- sample
- dictionary
- sparse
- signal
- discrete
- 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
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000012545 processing Methods 0.000 claims abstract description 15
- 230000008859 change Effects 0.000 claims description 23
- 238000010606 normalization Methods 0.000 claims description 12
- 238000012216 screening Methods 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 238000006467 substitution reaction Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 description 7
- 230000013016 learning Effects 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000009394 selective breeding Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000011437 continuous method Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000011664 signaling Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Abstract
The method for expressing and equipment of dictionary, are related to field of signal processing in a kind of sparse model provided in an embodiment of the present invention, reduce reconstructed error by using the solution procedure of discrete dictionary, and obtain clear and definite sample class information by solution procedure.This method includes：Sparse model is set up according to the signal acquisition sample of signal of reception, and by the data distribution characteristics of the sample of signal；The sparse coding of the signal is obtained by calculating the sparse model according to the sample of signal；The sparse coding iteration is entered into the sparse model calculating and obtains discrete word allusion quotation, the sample set of at least one sample of signal is obtained by discrete dictionary processing described in loop iteration, until terminating the loop iteration when meeting predetermined Rule of judgment, and at least one described sample of signal is subjected to the new signal of statistics composition；Output is by counting the new signal that the sample set is obtained.Embodiments of the invention are applied to Digital Signal Processing and image processing techniques.
Description
Technical field
The present invention relates to a kind of method for expressing of dictionary in field of signal processing, more particularly to sparse model and equipment.
Background technology
In signalling technique development, classical signal processing method is faced with asking for substantial amounts of data storage and data transfer
Topic, for the storage of processing mass data and the problem of data transfer, with the proposition of compressive sensing theory with far below it is traditional how
The mode of Qwest's sample rate is handled signal, wherein, compressed sensing is obtained on a suitable excessively complete primitive collection
Take the theory with reconstruction signal.Here input signal only needs several primitives just can be by accurate reconstruction, so reconstruction signal exists
The mistake of complete dictionary can be generated by signal set in the hope of sparse coding, by trying to achieve sparse coding by crossing on complete primitive collection
Journey is sparse modeling.The sparse model includes the sparse coding and the excessively complete dictionary generated by signal set.
Because the reconstructed error that reconstruction signal occurs in sparse model is big, in order to reduce reconstructed error, on dictionary learning
Prior art is broadly divided into two major classes：First, artificial selection sample set as redundancy big dictionary, wherein, artificial selection is big
High-quality sample of signal is measured as dictionary, so as to ensure that sparse coding has preferable classification capacity, but is due to artificial choosing
The mode selected it cannot be guaranteed that reduce reconstructed error, and excessive dictionary can cause too high computation complexity and storage to be born,
And due to effective sample set can not be selected so as to select substantial amounts of sample as dictionary, and then increase what is calculated and store
Burden；
Second, the dictionary of relative compact is solved using continuous method, wherein, the target of continuation method is that reduction reconstruct is missed
Difference, error matrix is obtained by solving, and then solving characteristic vector using matrix disassembling method reduces mistake as dictionary primitive
Difference, and then using reconstructed error as one in object function, using methods such as stochastic gradient descents, solution can reduce error
Dictionary primitive.But, because the primitive in continuous dictionary is not belonging to sample set, so interpretation difference does not have language yet
Justice；The sparse coding obtained on continuous dictionary does not have clear and definite classification information.
The content of the invention
Embodiments of the invention provide the method for expressing and equipment of dictionary in a kind of sparse model, by using discrete dictionary
Solution procedure reduction reconstructed error, and clear and definite sample class information is obtained by solution procedure.
To reach above-mentioned purpose, embodiments of the invention are adopted the following technical scheme that：
First aspect there is provided a kind of method for expressing of dictionary in sparse model, including：
Sparse mould is set up according to the signal acquisition sample of signal of reception, and by the data distribution characteristics of the sample of signal
Type；
The sparse coding of the signal is obtained by calculating the sparse model according to the sample of signal；
The sparse coding iteration is entered into the sparse model calculating and obtains discrete word allusion quotation, by discrete described in loop iteration
Dictionary processing obtains the sample set of at least one sample of signal, until being followed described in being terminated when meeting predetermined Rule of judgment
Ring iterative, and at least one described sample of signal is subjected to the new signal of statistics composition；
Output is by counting the new signal that the sample set is obtained.
In the first possible implementation, specifically included with reference to first aspect, the sparse model is embodied in：As { d1, d2..., dk∈ X when, wherein aiFor sparse coding,In j be
The a of sparse codingiSequence number, { d1, d2..., dkBe sample set X sample set, djFor the sample in the sample set
This, xiTo constitute the single sample of the sample set X, N is number of samples, and coefficient lambda is used to limit the dilute of the sparse coding
The degree of dredging, wherein, the sparse degree is the non-zero entry number in the sparse coding.
In second of possible implementation, with reference to the first possible realization side in first aspect or first aspect
Formula is specifically included, it is described according to the sample of signal by calculate the sparse model obtain the signal sparse coding it
Before, methods described also includes：
Setting circulation sum is T, and worst error value is eta, and the circulation sum T is the sparse model loop iteration meter
The cycle-index of calculation, the error amount that the worst error value eta is set when being and calculating reconstructed error by the sparse model
Threshold value, wherein, it is described to circulate the Rule of judgment that total or described worst error value calculates end for loop iteration；
By sample set { x1, x2... ..., xNNormalization, and K sample is arbitrarily selected as discrete dictionary, wherein, it is described
Sample set is normalized to the sample in the sample set turning to constant 1, the sample set { x by a square summation1, x2... ...,
xNNormalization be expressed asThe number of samples K is optional number of samples in the sample set.
In the third possible implementation, any possible realization included with reference to first aspect or first aspect
Mode is specifically included, the sparse coding for obtaining the signal by calculating the sparse model according to the sample of signal,
Including：
The discrete dictionary is set as known conditions, and the discrete dictionary is brought into the expression formula of the sparse model
Lasso-LARS algorithms are returned to obtain the sparse coding of N number of sample by minimum angular convolution according to the expression formula, wherein,
The sparse coding of N number of sample is the coding number of the multiple samples of correspondence of the sparse coding, and the sparse coding is
{a1, a2... ..., aN}。
In the 4th kind of possible implementation, any possible realization included with reference to first aspect or first aspect
Mode is specifically included, described the sparse coding iteration is entered into the sparse model to calculate to obtain discrete word allusion quotation, including：
The sparse coding is set as known conditions, and the sparse coding is substituted into the expression formula of the sparse model
Obtain calculating the expression formula of discrete dictionary, the expression formula of the discrete dictionary of calculating is as { d1, d2..., dk∈ X when,The sparse coding is A={ a1, a2... ..., aN, wherein,It is set as that the sparse model expression formula is substituted into after known conditions to be used to ask for the sparse coding
The normal form of the discrete dictionary is solved, the X is known sample set, and D is the discrete dictionary, { d1, d2..., dk∈ D are
Sample set in discrete dictionary；
As the dictionary primitive d in the sample setiThe expression formula of the discrete dictionary of calculating is substituted into successively, and is updated to K
During individual dictionary primitive, other dictionary primitives are set as known conditions, by the expression formula for calculating the discrete dictionary
Replace as E so that the expression formula of the discrete dictionary of calculating is converted to
And discrete dictionary is calculated, wherein X is known sample set, and A is the sparse coding, diFor known dictionary primitive, E-dkA is used
The discrete dictionary is obtained in being calculated by Screening Samples, to constitute new output signal,For X-DA
In dictionary primitive diExtend and obtain after substituting into successively.
In the 5th kind of possible implementation, specifically included with reference to the 4th kind of possible implementation, it is described when described
Dictionary primitive d in sample setiThe expression formula of the discrete dictionary of calculating is substituted into successively, and is updated to k-th dictionary primitive
When, including：
The sample set for quoting the k-th dictionary primitive is chosen according to the sparse coding, the sample set is
The change type that each sample in the sample set is substituted into the expression formula for calculating discrete dictionary calculates described
Error amount of the sample in the change type, and the error amount and current error value are compared, the current error value
For
If the error amount is more than the current error value, give up the error amount, and removed in the sample set
Remove the corresponding sample of the error amount；
Or,
If the error amount is less than the current error value, retain the corresponding error amount of the sample, and be updated to the
One current error value, and the corresponding sample of the error amount is removed in the sample set, circulate in the sample set
Sample bring into the change type and calculate error amount, until the sample set is empty set.
Second aspect there is provided a kind of electronic equipment, including：
Collecting unit, for the signal acquisition sample of signal according to reception, and passes through the data distribution of the sample of signal
Feature sets up sparse model；
Computing unit, the sample of signal for being gathered according to the collecting unit is obtained by calculating the sparse model
To the sparse coding of the signal；
The computing unit, is additionally operable to the sparse coding iteration entering the sparse model to calculate and obtains discrete word allusion quotation,
The sample set of at least one sample of signal is obtained by discrete dictionary processing described in loop iteration, until meeting predetermined
The loop iteration is terminated during Rule of judgment, and at least one described sample of signal is subjected to the new signal of statistics composition；
Transmitting element, for exporting by counting the new signal that the sample set is obtained.
In the first possible implementation, specifically included with reference to second aspect, the sparse model is embodied in：As { d1, d2..., dk∈ X when, wherein aiFor sparse coding,In j be
The a of sparse codingiSequence number, { d1, d2..., dkBe sample set X sample set, djFor the sample in the sample set
This, xiTo constitute the single sample of the sample set X, N is number of samples, and coefficient lambda is used to limit the dilute of the sparse coding
The degree of dredging, wherein, the sparse degree is the non-zero entry number in the sparse coding.
In second of possible implementation, specifically included with reference to second aspect or the first possible implementation,
The equipment also includes：
Setting unit, for obtaining the sparse of the signal by calculating the sparse model according to the sample of signal
Before coding, setting circulation sum is T, and worst error value is eta, and the circulation sum T is the sparse model loop iteration
The cycle-index of calculating, the error amount that the worst error value eta is set when being and calculating reconstructed error by the sparse model
Threshold value, wherein, the circulation is total or the worst error value is calculates the Rule of judgment that loop iteration terminates；
Unit is chosen, for by sample set { x1, x2... ..., xNNormalization, and arbitrarily select K sample to be used as discrete word
Allusion quotation, wherein, the sample set is normalized to the sample in the sample set turning to constant 1, the sample by a square summation
Collect { x1, x2... ..., xNNormalization be expressed asThe number of samples K be the sample set in arbitrarily select
The number of samples selected.
In the third possible implementation, any possible realization included with reference to second aspect or second aspect
Mode is specifically included, and the computing unit includes：
Computation subunit, for setting the discrete dictionary as known conditions, and the discrete dictionary is brought into described dilute
Dredge the expression formula of model
The computation subunit, is additionally operable to return Lasso-LARS algorithms to obtain N by minimum angular convolution according to the expression formula
The sparse coding of individual sample, wherein, the sparse coding of N number of sample is the coding of the multiple samples of correspondence of the sparse coding
Number, the sparse coding is { a1, a2... ..., aN}。
In the 4th kind of possible implementation, any possible realization included with reference to second aspect or second aspect
Mode is specifically included, the computing unit, in addition to：
Conversion subunit, for setting the sparse coding as known conditions, and sparse coding substitution is described dilute
Obtain calculating the expression formula of discrete dictionary in the expression formula for dredging model, the expression formula of the discrete dictionary of calculating is as { d1,
d2..., dk∈ X when,The sparse coding is A={ a1, a2... ..., aN, its
In,It is set as that the sparse model expression formula is substituted into after known conditions to be used for for the sparse coding
The normal form of the discrete dictionary is solved, the X is known sample set, and D is the discrete dictionary, { d1, d2..., dk}∈D
For the sample set in discrete dictionary；
The conversion subunit, is additionally operable to work as the dictionary primitive d in the sample setiIt is discrete that the calculating is substituted into successively
The expression formula of dictionary, and when being updated to k-th dictionary primitive, set other dictionary primitives as known conditions, will calculate it is described from
In the expression formula for dissipating dictionaryReplace as E so that the expression formula of the discrete dictionary of calculating is converted toAnd discrete dictionary is calculated, wherein X is known sample
This set, A is the sparse coding, diFor known dictionary primitive, E-dkA, which is used to calculate by Screening Samples, obtains described discrete
Dictionary, to constitute new output signal,It is X-DA in dictionary primitive diExtended after substituting into successively
Arrive.
In the 5th kind of possible implementation, specifically included with reference to the 4th kind of possible implementation, it is described to calculate single
Member, the sample set of the k-th dictionary primitive is quoted specifically for being chosen according to the sparse coding, and the sample set is
The change type that each sample in the sample set is substituted into the expression formula for calculating discrete dictionary calculates described
Error amount of the sample in the change type, and the error amount and current error value are compared, the current error value
For
If the error amount is more than the current error value, give up the error amount, and removed in the sample set
Remove the corresponding sample of the error amount；
Or,
If the error amount is less than the current error value, retain the corresponding error amount of the sample, and be updated to the
One current error value, and the corresponding sample of the error amount is removed in the sample set, circulate in the sample set
Sample bring into the change type and calculate error amount, until the sample set is empty set.
The method for expressing and equipment of dictionary, are obtained by calculating sparse model in sparse model provided in an embodiment of the present invention
Sparse coding, then obtains discrete dictionary by iterating to calculate according to the sparse coding, wherein the value model by limiting coefficient lambda
The sparse degree of sparse coding processed is contained, by learning discrete dictionary and then solving the problem of reconstructed error is big, and by asking
Discrete dictionary is solved to reduce amount of calculation and obtain the clear and definite classification information of sample by solving discrete dictionary.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the accompanying drawing used required in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, can be with
Other accompanying drawings are obtained according to these accompanying drawings.
Fig. 1 is the schematic flow sheet of the method for expressing of dictionary in a kind of sparse model provided in an embodiment of the present invention；
Fig. 2 is the schematic flow sheet of the method for expressing of dictionary in another sparse model provided in an embodiment of the present invention；
Fig. 3 is the technique effect schematic diagram of the method for expressing of dictionary in a kind of sparse model provided in an embodiment of the present invention；
Fig. 4 is the structural representation of a kind of electronic equipment provided in an embodiment of the present invention；
Fig. 5 is the structural representation of another electronic equipment provided in an embodiment of the present invention；
Fig. 6 is the structural representation of another electronic equipment provided in an embodiment of the present invention.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation is described, it is clear that described embodiment is only a part of embodiment of the invention, rather than whole embodiments.It is based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of creative work is not made
Embodiment, belongs to the scope of protection of the invention.
The method for expressing of dictionary in a kind of sparse model that the present invention is provided, shown in reference picture 1, specifically includes to flow down
Journey：
101st, electronic equipment is according to the signal acquisition sample of signal of reception, and the data distribution characteristics for passing through the sample of signal
Set up sparse model.
Wherein, sparse model is embodied in：As { d1, d2...,
dk∈ X when, wherein aiFor sparse coding,For the j powers of sparse coding, { d1, d2..., dkSub for sample set X sample
Collection, djFor the sample in sample set, xiTo constitute sample set X single sample, N is number of samples, and coefficient lambda is used to limit
The sparse degree of sparse coding, sparse degree is the non-zero entry number in sparse coding here.
102nd, electronic equipment obtains the sparse coding of signal according to the sample of signal by calculating sparse model.
Here electronic equipment is by using minimum absolute retract selection opertor Lasso (Least Absolute
Shrinkage and Selection Operator) Algorithm for Solving sparse coding, sample set { x is normalized first1, x2... ...,
xNSo thatAnd K sample is arbitrarily selected as initialization dictionary D, and by fixed dictionary D, will
It is transformed toLasso-LARS algorithms (Least is returned by minimum angular convolution again
Angle Regression) transform is solved, respectively obtain the sparse coding { a of N number of sample1, a2... ..., aN}。
In the expression formula of sparse model provided in an embodiment of the present invention | | | | it is normal form, what the embodiment of the present invention was mentioned
Normal form is a normal form and two normal forms.
103rd, electronic equipment by the sparse coding iteration enter sparse model calculate obtain discrete word allusion quotation, by loop iteration from
Dissipate dictionary processing and obtain the sample set of at least one sample of signal, changed until terminating circulation when meeting predetermined Rule of judgment
Generation, and at least one sample of signal is subjected to the new signal of statistics composition.
Wherein, the discrete dictionary is the sample set D in sample set X, and sample set D is dictionary primitive d herejComposition
's.
Here fixed sparse coding A={ a are passed through1, a2... ..., aN, A iteration is returnedObtain solving the change type of discrete dictionaryWherein
{d1, d2..., dk}∈X.By the way that dictionary primitive di is updated into k-th, and other dictionary primitives are fixed, by change typeReplace and be converted to again for E
Specifically, A is fixed sparse coding matrix, dictionary primitive d in EiIt is also fixed value, X is known sample set,
Therefore E can be learnt by calculating, here by updating dictionary primitive diTo dictionary primitive dkDiscrete dictionary is solved, that is, passes through dictionary
Dictionary primitive in the discrete dictionary that primitive correspondence sample set screening is solved so that solving the discrete dictionary drawn can retain
The data distribution characteristics of sample, and sparse coding can be obtained and there is clear and definite classification information.
104th, the new signal that electronic equipment output is obtained by statistical sample subset.
The method for expressing of dictionary in sparse model provided in an embodiment of the present invention, sparse volume is obtained by calculating sparse model
Code, then obtains discrete dictionary by iterating to calculate according to the sparse coding, wherein being controlled by the span for limiting coefficient lambda
The sparse degree of sparse coding, by learning discrete dictionary and then solving the problem of reconstructed error is big, and it is discrete by solving
Dictionary reduces amount of calculation and obtains the clear and definite classification information of sample by solving discrete dictionary.
Specifically, being illustrated below in conjunction with specific embodiment.
Can be on the basis of the embodiment shown in Fig. 1, shown in reference picture 2, The embodiment provides one kind is dilute
The method for expressing of dictionary in model is dredged, is mainly included：Electronic equipment obtains sparse coding by calculating sparse model, further according to dilute
Dredge coding iteration time sparse model expression formula calculating and obtain discrete word allusion quotation；It is that electronic equipment is obtained by calculating shown in reference picture 2
Sparse coding, then the process of discrete dictionary according to the sparse coding is calculated is obtained by iteration, comprise the following steps that：
201st, electronic equipment is according to the signal acquisition sample of signal of reception, and the data distribution characteristics for passing through the sample of signal
Set up sparse model.
Wherein, sparse model is embodied in：As { d1, d2...,
dk∈ X when, wherein aiFor sparse coding,For the j powers of sparse coding, { d1, d2..., dkSub for sample set X sample
Collection, djFor the sample in sample set, xiTo constitute sample set X single sample, N is number of samples, and coefficient lambda is used to limit
The sparse degree of sparse coding, sparse degree is the non-zero entry number in sparse coding here.
202nd, electronic equipment setting circulation sum is T, and worst error value is eta.
Wherein, circulation sum T is the cycle-index that sparse model loop iteration is calculated, and worst error value eta is logical
The threshold value that sparse model calculates the error amount set during reconstructed error is crossed, wherein, circulation sum or worst error value are tied to calculate
The decision condition of beam.
Here preset circulation sum and worst error value, for judge iterate to calculate discrete dictionary when, be repeatedly
Discrete dictionary chooses sample and provides criterion, and when the number of times of loop iteration reaches T times, calculating terminates；Or, when selection
When the error amount of sample is less than worst error value eta, calculating terminates.
203rd, electronic equipment is by sample set { x1, x2... ..., xNNormalization, and arbitrarily select K sample to be used as discrete word
Allusion quotation.
Wherein, the sample set is normalized to the sample in sample set turning to constant 1 by a square summation, the sample set
{x1, x2... ..., xNNormalization be expressed asNumber of samples K is optional sample in the sample set
Number.
Here the embodiment of the present invention calculates discrete dictionary and calculates sparse coding again using first fixing a discrete dictionary, then by
Sparse coding iteration returns former formula progress transformation calculations and obtains discrete dictionary.Therefore step 203 carry out calculating sparse coding before first
It is determined that initialization dictionary so that calculate sparse coding by fixed this variable of initialization dictionary.
204th, electronic equipment obtains the sparse coding of signal according to the sample of signal by calculating sparse model.
Here electronic equipment is by using minimum absolute retract selection opertor Lasso (Least Absolute
Shrinkage and Selection Operator) Algorithm for Solving sparse coding, sample set { x is normalized first1, x2... ...,
xNSo thatAnd K sample is arbitrarily selected as initialization dictionary D, and by setting dictionary D as known bar
Part, willIt is transformed toAgain by minimum angular convolution
Return Lasso-LARS algorithms (Least Angle Regression) to solve transform, respectively obtain the sparse coding of N number of sample
{a1, a2... ..., aN, wherein, the sparse coding of N number of sample is the coding number of the multiple samples of correspondence of the sparse coding,.
In the expression formula of sparse model provided in an embodiment of the present invention | | | | it is normal form, what the embodiment of the present invention was mentioned
Normal form is a normal form and two normal forms.
In solution procedure, the λ value of setting is bigger, and the nonzero term in sparse coding is fewer, and reconstructed error is also bigger.Therefore
The embodiment of the present invention provides the coefficient lambda set to be used to limit the sparse of the sparse coding as any value in the range of 0.1~0.25
Degree, wherein, when coefficient lambda is 0.15, the classification capacity of sparse model reaches peak value, specifically, coefficient lambda fluctuation generally exists
0.15 positive negative error is calculated between 0.05 obtains peak value, and sparse degree is the non-zero entry number in the sparse coding here.This
In can not only obtain suitable sparse coding and can also reduce reconstructed error, specifically when being calculated using Lasso-LARS algorithms
Can be the dictionary primitive distribution coefficient during sparse coding, the coefficient is generally smaller so that the sample error after reconstruct is located at the word
In the bisector of allusion quotation primitive and other dictionary primitives, cycle assignment process, it is known that the enough dictionary primitives of selection carry out reconstructed sample.
When the non-zero entry number of sparse coding is far smaller than the number of dictionary primitive, Lasso algorithms are stable and restrained
, now it can ensure the sparse degree of sparse coding by controlling λ value to be more than or equal to 0.01.
205th, electronic equipment by the sparse coding iteration enter sparse model calculate obtain discrete word allusion quotation, by loop iteration from
Dissipate dictionary processing and obtain the sample set of at least one sample of signal, changed until terminating circulation when meeting predetermined Rule of judgment
Generation, and at least one sample of signal is subjected to the new signal of statistics composition.
Wherein, the discrete dictionary is the sample set D in sample set X, and sample set D is dictionary primitive d herejComposition
's.
Here by setting sparse coding A={ a1, a2... ..., aNIt is known conditions, A iteration is returnedObtain solving the change type of discrete dictionary, as { d1, d2..., dk∈ X when,Wherein,It is set as sparse coding after known conditions
The normal form that sparse model expression formula is used to solve the discrete dictionary is substituted into, X is known sample set, and D is the discrete word
Allusion quotation, { d1, d2..., dk∈ D be discrete dictionary in sample set.By by dictionary primitive diK-th is updated to, and is set
Other dictionary primitives are known conditions, by change typeReplace and be converted to again for EAnd discrete dictionary is calculated, wherein X is known sample
This set, A is sparse coding, diFor dictionary primitive, E-dkA, which is used to calculate by Screening Samples, obtains discrete word allusion quotation, to constitute
New output signal,It is X-DA in dictionary primitive diExtend and obtain after substituting into successively.
Specifically, A is fixed sparse coding matrix, dictionary primitive d in EiIt is also fixed value, X is known sample set,
Therefore E can be learnt by calculating, here by updating dictionary primitive diTo dictionary primitive dkDiscrete dictionary is solved, that is, passes through dictionary
Dictionary primitive in the discrete dictionary that primitive correspondence sample set screening is solved so that solving the discrete dictionary drawn can retain
The data distribution characteristics of sample, and sparse coding can be obtained and there is clear and definite classification information.
Further, as the dictionary primitive d in sample setiThe expression formula for calculating discrete dictionary is substituted into successively, and is updated
During to k-th dictionary primitive, including：
205a, electronic equipment choose the sample set for quoting k-th dictionary primitive according to sparse coding.
Wherein, the sample set is
205b, electronic equipment calculate the change type that each sample in sample set substitutes into the expression formula for calculating discrete dictionary
Error amount of the sample in change type, and the error amount and current error value are compared.
Wherein, the current error value is expressed as ‖ E-dkA||2。
If 205c, error amount are more than current error value, give up error amount, and the removing error amount correspondence in sample set
Sample.
Or,
If 205d, error amount are less than current error value, retain the corresponding error amount of sample, and be updated to first and currently miss
Difference, and the corresponding sample of error amount is removed in sample set, circulate and bring the sample in the sample set in change type into
Calculation error value, until sample set is empty set.
Specifically, provided in an embodiment of the present invention is by loop iteration by using the discrete dictionary of Lasso algorithms calculating
Realize, i.e., after step 204 calculates sparse coding according to fixed discrete dictionary, further fix dilute by step 205
Dredge coding and calculate the dictionary primitive that discrete dictionary obtains constituting dictionary, step 204 is then carried out again fix new discrete dictionary to ask
Solve sparse coding and then new dictionary primitive is obtained by step 205 and so circulate, until obtaining that known sample collection can be corresponded to
Close X discrete dictionary.
206th, the new signal that electronic equipment output is obtained by statistical sample subset.
Here the embodiment of the present invention is provided in the method for calculating discrete dictionary, because the dictionary primitive in discrete dictionary comes from
Sample set X, therefore always exist and dictionary primitive d when sample is chosenkIdentical sample xj.According to Lasso algorithms, sample
X in this set XjD can be selectedk, therefore updating dkWhen, e sample can not be reduced if there is other samples, then xjIt can be chosen
For dictionary primitive, and e is constant；Opposite, e sample can be reduced if there is other samples, then new samples are saved, and e subtracts
It is small.Wherein, during dictionary updating, reconstructed error can keep constant or reduce.
Wherein, the method for expressing of dictionary can also pass through orthogonal matching pursuit in sparse model provided in an embodiment of the present invention
Algorithm (Orthogonal Matching Pursuit, OMP) algorithm is realized, because OMP algorithms are more suitable for orthogonal primitive dictionary,
And offer of the embodiment of the present invention realizes that the process object of the method for expressing of dictionary in sparse model was complete dictionary, therefore use
Lasso algorithms make obtained excessively complete dictionary more stable more accurate.
The method for expressing of dictionary in sparse model provided in an embodiment of the present invention, obtains dilute by calculating sparse model dictionary
Coding is dredged, and the sparse degree of sparse coding is controlled by limiting the span of coefficient lambda, it is sparse according to this by iterating to calculate
Coding obtains discrete dictionary, and calculate discrete dictionary portion by the selection to sample cause reconstructed error keep it is constant or
Reduce, so that overall reduces reconstructed error, and amount of calculation is reduced and by solving discrete word by solving discrete dictionary
Allusion quotation obtains the clear and definite classification information of sample.
Specifically, the method for expressing of dictionary is referring in particular to shown in Fig. 3 in sparse model provided in an embodiment of the present invention, with three
Exemplified by dimension sample, the discrete dictionary of one 3 × 10 is obtained from 100 three-dimensional data set learnings, left figure is sample set space
Part, right figure is the dictionary Primitive Element Distribution (being the crunode shown in right figure) of discrete dictionary, wherein, the primitive in the discrete dictionary is
Sample set.
Here the data distribution characteristics in sample set space are remained by the discrete dictionary calculated in sparse model, i.e., at place
Manage the feature that ensure that original signal after the signal received when showing the signal.
The method for expressing of dictionary in sparse model provided in an embodiment of the present invention, finds to lead to by the analysis to discrete dictionary
Cross learn discrete dictionary can be with the data distribution characteristics of stick signal sample, and due to the dictionary base do not obscured in discrete dictionary
Member so that the sparse coding that can be obtained on discrete dictionary has clear and definite classification information.
The embodiment of the present invention provides a kind of electronic equipment 3, and the electronic equipment is specially any electricity in signal processing system
Sub- equipment, such as computer, notebook computer are any in signal processing system can to realize the expression of dictionary in sparse model
The equipment of method, using can realize the method for expressing of dictionary in any of the above-described sparse model that embodiments of the invention are provided as
Standard, shown in reference picture 4, including：
Collecting unit 31, for the signal acquisition sample of signal according to reception, and the data distribution for passing through the sample of signal
Feature sets up sparse model；
Computing unit 32, the signal is obtained for the sample of signal that is gathered according to collecting unit by calculating sparse model
Sparse coding；
Computing unit 32, be additionally operable to by the sparse coding iteration enter sparse model calculate obtain discrete word allusion quotation, pass through circulation
The processing of iterative discrete dictionary obtains the sample set of at least one sample of signal, is followed until being terminated when meeting predetermined Rule of judgment
Ring iterative, and at least one sample of signal is subjected to the new signal of statistics composition；
Transmitting element 33, for exporting the new signal obtained by statistical sample subset.
Electronic equipment provided in an embodiment of the present invention, obtains sparse coding by calculating sparse model, then passes through iteration
Calculating obtains discrete dictionary according to the sparse coding, wherein controlling the sparse of sparse coding by the span for limiting coefficient lambda
Degree, calculating is reduced by learning discrete dictionary and then solving the problem of reconstructed error is big, and by solving discrete dictionary
Measure and obtain the clear and definite classification information of sample by solving discrete dictionary.
Further, sparse model is embodied in：As { d1,
d2..., dk∈ X when, wherein aiFor sparse coding,In j be sparse coding aiSequence number, { d1, d2..., dkIt is sample
This set X sample set, djFor the sample in sample set, x is composition sample set X single sample, and N is sample
Number, coefficient lambda is used for the sparse degree for limiting sparse coding, wherein, the sparse degree is the non-zero entry number in sparse coding.
Optionally, shown in reference picture 5, electronic equipment 3 also includes：
Setting unit 34, for before the sparse coding of signal is obtained by calculating sparse model according to sample of signal,
Setting circulation sum is T, and worst error value is eta, and circulation sum T is the cycle-index that sparse model loop iteration is calculated,
The threshold value of error amount that worst error value eta is set when being and calculating reconstructed error by sparse model, wherein, circulation sum or most
The decision condition that big error amount terminates for calculating；
Unit 35 is chosen, for by sample set { x1, x2... ..., xNNormalization, and K sample is arbitrarily selected as discrete
Dictionary, wherein, the sample set is normalized to the sample in sample set turning to constant 1 by a square summation, the sample set { x1,
x2... ..., xNNormalization be expressed asNumber of samples K is optional number of samples in sample set.
Optionally, shown in reference picture 6, computing unit 32 includes：
Computation subunit 321, for setting discrete dictionary as known conditions, and discrete dictionary is brought into the table of sparse model
Up to formula
Computation subunit 321, is additionally operable to return Lasso-LARS algorithms to obtain N number of sample by minimum angular convolution according to expression formula
Sparse coding, wherein, the sparse coding of N number of sample is the coding number of the multiple samples of correspondence of sparse coding, sparse coding
For { a1, a2... ..., aN}。
Further, shown in reference picture 6, computing unit 32, in addition to：
Conversion subunit 322, for setting sparse coding as known conditions, and by sparse coding substitute into sparse model table
Obtain calculating the expression formula of discrete dictionary up in formula, the expression formula for calculating discrete dictionary is as { d1, d2..., dk∈ X when,Sparse coding is A={ a1, a2... ..., aN, wherein,It is set as that sparse model expression formula is substituted into after known conditions to be used to solve discrete word for sparse coding
The normal form of allusion quotation, X is known sample set, and D is discrete dictionary, { d1, d2..., dk∈ D be discrete dictionary in sample set
Close；
Conversion subunit 322, is additionally operable to work as the dictionary primitive d in the sample setiThe table for calculating discrete dictionary is substituted into successively
Up to formula, and when being updated to k-th dictionary primitive, other dictionary primitives are set as known conditions, by the expression formula for calculating discrete dictionaryReplace as E so that the expression formula of the discrete dictionary of the calculating is converted to
And discrete dictionary is calculated, wherein X is known sample set, and A is sparse coding, diFor known dictionary primitive, E-dkA is used to lead to
Cross Screening Samples calculating and obtain the discrete dictionary, to constitute new output signal,It is X-DA in word
Allusion quotation primitive diExtend and obtain after substituting into successively.
Further, computing unit 32, sample of k-th dictionary primitive is quoted specifically for being chosen according to sparse coding
Collect, sample set is
The change type that each sample in sample set is substituted into the expression formula for calculating discrete dictionary calculates sample in change type
In error amount, and error amount is compared with current error value, the current error value is | | E-dkA||2；
If error amount is more than current error value, give up the error amount, and the error amount is removed in sample set and correspond to
Sample；
Or,
If error amount is less than current error value, retain the corresponding error amount of sample, and be updated to the first current error value,
And the corresponding sample of error amount is removed in the sample set, the sample in sample set is brought into calculate in change type and missed by circulation
Difference, until sample set is empty set.
Electronic equipment provided in an embodiment of the present invention, obtains sparse coding, and pass through limit by calculating sparse model dictionary
The span for determining coefficient lambda controls the sparse degree of sparse coding, and discrete word is obtained according to the sparse coding by iterating to calculate
Allusion quotation, and reconstructed error is kept constant or is reduced by the selection to sample in the discrete dictionary portion of calculating, so that overall
Reduce reconstructed error, and discrete dictionary reduces amount of calculation and to obtain sample clear and definite by solving discrete dictionary by solving
Classification information.
The foregoing is only a specific embodiment of the invention, but protection scope of the present invention is not limited thereto, any
Those familiar with the art the invention discloses technical scope in, change or replacement can be readily occurred in, should all be contained
Cover within protection scope of the present invention.Therefore, protection scope of the present invention described should be defined by scope of the claims.
Claims (8)
1. the method for expressing of dictionary in a kind of sparse model, it is characterised in that including：
Sparse model is set up according to the signal acquisition sample of signal of reception, and by the data distribution characteristics of the sample of signal；
The sparse coding of the signal is obtained by calculating the sparse model according to the sample of signal；
The sparse coding iteration is entered into the sparse model calculating and obtains discrete word allusion quotation, passes through discrete dictionary described in loop iteration
Processing obtains the sample set of at least one sample of signal, is changed until terminating the circulation when meeting predetermined Rule of judgment
Generation, and the sample set of at least one sample of signal is subjected to the new signal of statistics composition；
Export the new signal；
Wherein, the sparse model is embodied in：As { d1,d2,...,dk}
During ∈ X, wherein aiFor sparse coding,In j be sparse coding aiSequence number, { d1,d2,...,dkFor sample set X's
Sample set, djFor the sample in the sample set, xiTo constitute the single sample of the sample set X, N is number of samples,
Coefficient lambda is used for the sparse degree for limiting the sparse coding, wherein, the sparse degree is the non-zero entry in the sparse coding
Number；
It is described before the sparse coding for obtaining the signal by calculating the sparse model according to the sample of signal
Method also includes：
Setting circulation sum is T, and worst error value is eta, and the circulation sum T is what the sparse model loop iteration was calculated
Cycle-index, the threshold value for the error amount that the worst error value eta is set when being and calculating reconstructed error by the sparse model,
Wherein, it is described to circulate the Rule of judgment that total or described worst error value calculates end for loop iteration；
By sample set { x1,x2... ..., xNNormalization, and K sample is arbitrarily selected as discrete dictionary, wherein, the sample
Collection is normalized to the sample in the sample set turning to constant 1, the sample set { x by a square summation1,x2... ..., xN}
Normalization is expressed asThe number of samples K is optional number of samples in the sample set.
2. according to the method described in claim 1, it is characterised in that described described sparse by calculating according to the sample of signal
Model obtains the sparse coding of the signal, including：
The discrete dictionary is set as known conditions, and the discrete dictionary is brought into the expression formula of the sparse model
Lasso-LARS algorithms are returned to obtain the sparse coding of N number of sample by minimum angular convolution according to the expression formula, wherein, it is described
The sparse coding of N number of sample is the coding number of the multiple samples of correspondence of the sparse coding, and the sparse coding is { a1,
a2... ..., aN}。
3. according to any described method of claim 1~2, it is characterised in that it is described the sparse coding iteration is entered it is described
Sparse model calculates and obtains discrete word allusion quotation, including：
The sparse coding is set as known conditions, and will be obtained in the expression formula of the sparse coding substitution sparse model
The expression formula of discrete dictionary is calculated, the expression formula of the discrete dictionary of calculating is as { d1,d2,...,dk∈ X when,The sparse coding is A={ a1, a2... ..., aN, wherein,It is set as that the sparse model expression formula is substituted into after known conditions to be used to ask for the sparse coding
The normal form of the discrete dictionary is solved, the X is known sample set, and D is the discrete dictionary, { d1,d2,...,dk∈ D are
Sample set in discrete dictionary；
As the dictionary primitive d in the sample setiThe expression formula of the discrete dictionary of calculating is substituted into successively, and is updated to k-th
During dictionary primitive, other dictionary primitives are set as known conditions, by the expression formula for calculating the discrete dictionaryReplace as E so that the expression formula of the discrete dictionary of calculating is converted toAnd discrete dictionary is calculated, wherein X is known sample
This set, A is the sparse coding, diFor known dictionary primitive, the E-dkA, which is used to calculate by Screening Samples, obtains described
Discrete dictionary, it is described to constitute new output signalIt is X-DA in dictionary primitive diAfter substituting into successively
Extension is obtained.
4. method according to claim 3, it is characterised in that the dictionary primitive d worked as in the sample setiSuccessively
The expression formula for calculating discrete dictionary is substituted into, and when being updated to k-th dictionary primitive, including：
The sample set for quoting the k-th dictionary primitive is chosen according to the sparse coding, the sample set is
The change type that each sample in the sample set is substituted into the expression formula for calculating discrete dictionary calculates the sample
Error amount in the change type, and the error amount and current error value are compared, the current error value is | |
E-dkA||2；
If the error amount is more than the current error value, give up the error amount, and remove in the sample set institute
State the corresponding sample of error amount；
Or,
If the error amount is less than the current error value, retains the corresponding error amount of the sample, and be updated to first and work as
Preceding error amount, and the corresponding sample of the error amount is removed in the sample set, circulate the sample in the sample set
Originally bring into the change type and calculate error amount, until the sample set is empty set.
5. a kind of electronic equipment, it is characterised in that including：
Collecting unit, for the signal acquisition sample of signal according to reception, and passes through the data distribution characteristics of the sample of signal
Set up sparse model；
Computing unit, institute is obtained for the sample of signal that is gathered according to the collecting unit by calculating the sparse model
State the sparse coding of signal；
The computing unit, is additionally operable to the sparse coding iteration entering the sparse model to calculate and obtains discrete word allusion quotation, passes through
Discrete dictionary processing obtains the sample set of at least one sample of signal described in loop iteration, until meeting predetermined judgement
The loop iteration is terminated during condition, and the sample set of at least one sample of signal is subjected to statistics and is constituted newly
Signal；
Transmitting element, for exporting the new signal；
Wherein, the sparse model is embodied in：As { d1,d2,...,dk}
During ∈ X, wherein aiFor sparse coding,In j be sparse coding aiSequence number, { d1,d2,...,dkFor sample set X's
Sample set, djFor the sample in the sample set, xiTo constitute the single sample of the sample set X, N is number of samples,
Coefficient lambda is used for the sparse degree for limiting the sparse coding, wherein, the sparse degree is the non-zero entry in the sparse coding
Number；
The equipment also includes：
Setting unit, for obtaining the sparse coding of the signal by calculating the sparse model according to the sample of signal
Before, setting circulation sum is T, and worst error value is eta, and the circulation sum T calculates for the sparse model loop iteration
Cycle-index, the threshold of error amount that the worst error value eta is set when being and calculating reconstructed error by the sparse model
Value, wherein, it is described to circulate the Rule of judgment that total or described worst error value calculates end for loop iteration；
Unit is chosen, for by sample set { x1,x2... ..., xNNormalization, and K sample is arbitrarily selected as discrete dictionary,
Wherein, the sample set is normalized to the sample in the sample set turning to constant 1, the sample set by a square summation
{x1,x2... ..., xNNormalization be expressed asThe number of samples K be the sample set in arbitrarily select
Number of samples.
6. equipment according to claim 5, it is characterised in that the computing unit includes：
Computation subunit, the sparse mould is brought into for setting the discrete dictionary as known conditions, and by the discrete dictionary
The expression formula of type
The computation subunit, is additionally operable to return Lasso-LARS algorithms to obtain N number of sample by minimum angular convolution according to the expression formula
This sparse coding, wherein, the sparse coding of N number of sample is individual for the coding of the multiple samples of correspondence of the sparse coding
Number, the sparse coding is { a1, a2... ..., aN}。
7. according to any described equipment of claim 5-6, it is characterised in that the computing unit, in addition to：
Conversion subunit, the sparse mould is substituted into for setting the sparse coding as known conditions, and by the sparse coding
Obtain calculating the expression formula of discrete dictionary in the expression formula of type, the expression formula of the discrete dictionary of calculating is as { d1,d2,...,
dk∈ X when,The sparse coding is A={ a1, a2... ..., aN, wherein,It is set as that the sparse model expression formula is substituted into after known conditions to be used to ask for the sparse coding
The normal form of the discrete dictionary is solved, the X is known sample set, and D is the discrete dictionary, { d1,d2,...,dk∈ D are
Sample set in discrete dictionary；
The conversion subunit, is additionally operable to work as the dictionary primitive d in the sample setiThe discrete dictionary of calculating is substituted into successively
Expression formula, and when being updated to k-th dictionary primitive, set other dictionary primitives as known conditions, the discrete word will be calculated
In the expression formula of allusion quotationReplace as E so that the expression formula of the discrete dictionary of calculating is converted toAnd discrete dictionary is calculated, wherein X is known sample
This set, A is the sparse coding, diFor known dictionary primitive, E-dkA, which is used to calculate by Screening Samples, obtains described discrete
Dictionary, to constitute new output signal,It is X-DA in dictionary primitive diExtended after substituting into successively
Arrive.
8. equipment according to claim 7, it is characterised in that the computing unit, specifically for according to the sparse volume
Code chooses the sample set for quoting the k-th dictionary primitive, and the sample set is
The change type that each sample in the sample set is substituted into the expression formula for calculating discrete dictionary calculates the sample
Error amount in the change type, and the error amount and current error value are compared, the current error value is | |
E-dkA||2；
If the error amount is more than the current error value, give up the error amount, and remove in the sample set institute
State the corresponding sample of error amount；
Or,
If the error amount is less than the current error value, retains the corresponding error amount of the sample, and be updated to first and work as
Preceding error amount, and the corresponding sample of the error amount is removed in the sample set, circulate the sample in the sample set
Originally bring into the change type and calculate error amount, until the sample set is empty set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310115751.3A CN104103060B (en) | 2013-04-03 | 2013-04-03 | The method for expressing and equipment of dictionary in a kind of sparse model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310115751.3A CN104103060B (en) | 2013-04-03 | 2013-04-03 | The method for expressing and equipment of dictionary in a kind of sparse model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104103060A CN104103060A (en) | 2014-10-15 |
| CN104103060B true CN104103060B (en) | 2017-09-12 |
Family
ID=51671184
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310115751.3A Active CN104103060B (en) | 2013-04-03 | 2013-04-03 | The method for expressing and equipment of dictionary in a kind of sparse model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104103060B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113297879A (en) * | 2020-02-23 | 2021-08-24 | 深圳中科飞测科技股份有限公司 | Acquisition method of measurement model group, measurement method and related equipment |
| CN112203089B (en) * | 2020-12-03 | 2021-02-23 | 中国科学院自动化研究所 | Image compression method, system and device based on code rate control of sparse coding |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102354395A (en) * | 2011-09-22 | 2012-02-15 | 西北工业大学 | Sparse representation-based blind restoration method of broad image |
| CN102708576A (en) * | 2012-05-18 | 2012-10-03 | 西安电子科技大学 | Method for reconstructing partitioned images by compressive sensing on the basis of structural dictionaries |
-
2013
- 2013-04-03 CN CN201310115751.3A patent/CN104103060B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102354395A (en) * | 2011-09-22 | 2012-02-15 | 西北工业大学 | Sparse representation-based blind restoration method of broad image |
| CN102708576A (en) * | 2012-05-18 | 2012-10-03 | 西安电子科技大学 | Method for reconstructing partitioned images by compressive sensing on the basis of structural dictionaries |
Non-Patent Citations (3)
| Title |
|---|
| 主元分析中的稀疏性;向馗等;《电子学报》;20121231;第40卷(第12期);2525-2532 * |
| 基于压缩感知的分布式视频编码及其图像超分辨率重建研究;顾莹;《中国优秀硕士学位论文全文数据库 信息科技辑》;20120415;第2012年卷(第04期);I138-1949 * |
| 基于稀疏分解的心电信号特征波检测及心电数据压缩;王春光;《中国博士学位论文全文数据库 医药卫生科技辑》;20101015;第2010年卷(第10期);E080-4 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104103060A (en) | 2014-10-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106326288B (en) | Image search method and device | |
| CN110070178A (en) | A kind of convolutional neural networks computing device and method | |
| WO2018209932A1 (en) | Multi-quantization depth binary feature learning method and device | |
| CN110210618A (en) | The compression method that dynamic trimming deep neural network weight and weight are shared | |
| CN108764273A (en) | A kind of method, apparatus of data processing, terminal device and storage medium | |
| CN108090565A (en) | Accelerated method is trained in a kind of convolutional neural networks parallelization | |
| CN113159239B (en) | Method for processing graph data by quantum graph convolutional neural network | |
| CN111488986A (en) | Model compression method, image processing method and device | |
| CN106845640A (en) | It is heterogeneous in layer based on depth convolutional neural networks to pinpoint quantization method at equal intervals | |
| CN110033089A (en) | Deep neural network parameter optimization method and system based on Distributed fusion algorithm | |
| CN108416187A (en) | A kind of method and device of determining pruning threshold, model pruning method and device | |
| CN109740660A (en) | Image processing method and device | |
| CN114881237A (en) | Quantum calculation processing method and device and electronic equipment | |
| CN107395211A (en) | A kind of data processing method and device based on convolutional neural networks model | |
| CN110620586B (en) | Data compression method and device and terminal equipment | |
| CN103970719B (en) | Fitting method and fitting device | |
| CN104103060B (en) | The method for expressing and equipment of dictionary in a kind of sparse model | |
| WO2023124342A1 (en) | Low-cost automatic neural architecture search method for image classification | |
| Samragh et al. | Encodeep: Realizing bit-flexible encoding for deep neural networks | |
| CN110110402A (en) | The fast solution method of resource allocation system reachable state based on Binary Decision Diagrams | |
| Abbas et al. | Effective dimension of machine learning models | |
| CN115238568A (en) | Digital twin model construction method and device and terminal equipment | |
| CN108764280A (en) | A kind of medical data processing method and system based on symptom vector | |
| Ly et al. | A high-performance FPGA architecture for restricted Boltzmann machines | |
| Qi et al. | Learning low resource consumption cnn through pruning and quantization |
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: 20171214 Address after: 510640 Guangdong City, Tianhe District Province, No. five, road, public education building, unit 371-1, unit 2401 Patentee after: Guangdong Gaohang Intellectual Property Operation Co., Ltd. Address before: 518129 Bantian HUAWEI headquarters office building, Longgang District, Guangdong, Shenzhen Patentee before: Huawei Technologies Co., Ltd. |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20171219 Address after: 402260 Chongqing city Jiangjin Jijiang 90 Minsheng Road No. 1-7-2 Building 1 Patentee after: Chongqing GA doll Technology Co., Ltd. Address before: 510640 Guangdong City, Tianhe District Province, No. five, road, public education building, unit 371-1, unit 2401 Patentee before: Guangdong Gaohang Intellectual Property Operation Co., Ltd. |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20181220 Address after: 401120 Yubei District, Chongqing, Longxi street, Hong Jin Avenue 498, Jiale, Violet 1, 5- business. Patentee after: Chongqing Kai Tuo development in science and technology company limited Address before: 402260 Jiangjin 90 District, Chongqing City, No. 1, No. 1, No. Patentee before: Chongqing GA doll Technology Co., Ltd. |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20201110 Address after: No. ZC030, College Students Pioneer Park, Tongling City, Anhui Province, 244000 Patentee after: TONGLING HUIHENG ELECTRONIC TECHNOLOGY Co.,Ltd. Address before: 401120 Jiale Ziguang Building 5-Business, 498 Hongjin Avenue, Longxi Street, Yubei District, Chongqing Patentee before: CHONGQING KAITUO TECHNOLOGY DEVELOPMENT Co.,Ltd. |