CN114661940B - Method suitable for quickly acquiring voice countermeasure sample under black box attack - Google Patents
Method suitable for quickly acquiring voice countermeasure sample under black box attack Download PDFInfo
- Publication number
- CN114661940B CN114661940B CN202210106435.9A CN202210106435A CN114661940B CN 114661940 B CN114661940 B CN 114661940B CN 202210106435 A CN202210106435 A CN 202210106435A CN 114661940 B CN114661940 B CN 114661940B
- Authority
- CN
- China
- Prior art keywords
- curr
- attack
- area
- sample
- disturbance
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/60—Information retrieval; Database structures therefor; File system structures therefor of audio data
- G06F16/63—Querying
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/23—Updating
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a method suitable for fast obtaining voice countermeasure sample under black box attack, the method includes S1, adopting bipartite inquiry algorithm to determine decision boundary of original audio x, and iterating with sliding window method to select best attack area [ S: e ], S initial value is 0, e initial value is l, l is original audio length; s2, adding disturbance in a low-frequency area in the selected attack area [ S: e ], determining an updating step length by calculating a gradient direction, updating the disturbance, and obtaining a next iteration countermeasure sample by utilizing a binary query algorithm until the set sampling times are completed, so as to obtain a final countermeasure sample x. The method improves the efficiency of challenge sample generation.
Description
Technical Field
The invention relates to the field of voice processing, in particular to a method for quickly acquiring a voice countermeasure sample under a black box attack.
Background
Black boxes are an index and means of fairly evaluating the security of a model against sample attacks, which use a query model to generate a challenge sample. The existing black box attack main application objects are all models in the image field, and no corresponding work is performed in the voice field.
The audio data is a time sequence, is one-dimensional data, contains little information, is difficult to estimate accurate gradient information, and the image is two-dimensional data, has large information content, has stronger information dependence in space and can be used for more information; the typical sampling points per second of audio are 16000 or more, and a section of speaker audio usually has a time of 4s or more, which results in audio information having tens of thousands of data points in a single dimension, and images having only hundreds of sampling points in one dimension, compared with images, it is difficult for the audio to acquire an accurate update direction; the image is typically in the range of [0,1] after normalization, while the audio corresponds to typically [ -1,1].
For the above reasons, the voice has larger difference compared with the image, so that the black box attack needs more inquiry times when being applied to the audio field, the difficulty of generating an countermeasure sample is increased, and a fair judgment cannot be provided.
Disclosure of Invention
In view of the foregoing, it is an object of the present invention to provide a method for fast generating a challenge sample suitable for a black box attack, which can greatly reduce the number of queries of the attack.
In order to achieve the above purpose, the technical scheme of the invention is as follows: the method for quickly acquiring the voice countermeasure sample under the black box attack is characterized by comprising the following steps of: the method may include the steps of,
s1, determining a decision boundary of an original audio x by adopting a bipartite query algorithm, and iterating by matching with a sliding window method to select an optimal attack area [ S: e ], wherein the initial value of S is 0, the initial value of e is l, l is the length of the original audio,
s2, adding disturbance in a low-frequency area in the selected attack area [ S: e ], determining an updating step length by calculating a gradient direction, updating the disturbance, and obtaining a next iteration countermeasure sample by utilizing a binary query algorithm until the set sampling times are completed, so as to obtain a final countermeasure sample x.
Further, the S1 specifically includes,
s11, S takes a temporary value S curr For s curr Initializing the area with 0 to perform minimum function assignment to obtain a temporary end point e of the area curr =min (s+l α, l) and temporary audio x curr [s curr :e curr ]＝x t [s curr :e curr ]Wherein x is t For target audio, α represents the length ratio of the attack region to the original audio;
s12, performing dichotomy query according to the distance between the temporary audio and the original audio, and judging the dichotomy queryFractional query of residual size d relative to original audio curr ＝||B(x,x curr )-x|| 2 Whether or not less than d and temporary audio x curr Output f (x after model f curr ) Whether or not equal to t, where d represents the size of the minimum disturbance previously searched, and t represents the tag of the targeted speaker;
s13, when d curr < d and f (x) curr ) When t, the current optimal disturbance scale d=d is updated curr Updating attack area s=s curr ，e＝e curr And s curr Repeating S12 after taking w as a sliding step length and taking the value next time until S curr Executing S14 when the total number of the components is greater than or equal to l;
s14, updating the contracted attack area alpha by 0.9 x alpha, and ending the selection of the attack area when the current flag bit is negative, wherein the last attack area is taken as the optimal attack area.
Further, the S2 calculates the current gradient direction by adopting a Monte Carlo combined finite difference method, and the formula is that
Wherein H represents the number of samples, u h Representing a disturbance of the random sampling,representing the challenge sample at the ith iteration, e is a very small constant, the value 0.001, phi represents the indicator function for determining the gradient direction.
Further, the step S2 specifically comprises the following steps,
s21, setting an optimal attack area [ S: e ]]The low frequency region length of (c) is l' ≡β· (e-s), and the initial disturbance is against the sample
S22, in the process of [0,1]]Variable u with random sampling length of l h And at u h Post-filling l-l' length 0, using inverse discrete cosine transform IDCT variation to transform u h Conversion from the frequency domain back to the time domain u h ←IDCT(u h ) And nulling disturbances outside the attack area, i.e. u h [0:s)←0、u h (e:l]C, performing S23 when the sampling times are larger than H;
s23, through the formulaCalculating gradient and calculating step size xi by using grid search step length method i The disturbance is updated using the following formula>
S24, updating the countermeasure sampleObtaining a countercheck sample of the (i+1) th iteration until the value of i reaches the preset query times, and obtaining the countercheck sample by the last iteration as a final countercheck sample.
Further, the value of beta is 0.65.
Compared with the prior art, the invention has the advantages that:
the method comprises the steps of dividing the acquisition process of the countermeasure sample into two stages, adopting a binary inquiry algorithm and a sliding window algorithm to obtain an optimal attack area in the first stage, adopting gradient direction calculation and a binary inquiry algorithm on the basis of the optimal attack area selected in the first stage in the second stage, and obtaining the final countermeasure sample through iteration, so that the disturbance adding range is reduced, the attack inquiry times are greatly reduced, and the generation speed of the countermeasure sample is improved.
Drawings
Fig. 1 is a block diagram of the overall structure of a method for quickly acquiring a challenge sample under a black box attack.
Fig. 2 is a schematic structural diagram of an iterative algorithm corresponding to adding low-frequency disturbance.
Detailed Description
Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are illustrative only and are not to be construed as limiting the invention.
Fig. 1-2 show illustrations of a method of rapidly acquiring challenge samples, suitable for use in a black box attack, the method specifically comprising,
s1, determining a decision boundary of an original audio x by adopting a bipartite query algorithm, and iterating by matching with a sliding window method to select an optimal attack area [ S: e ], wherein the initial value of S is 0, the initial value of e is l, l is the length of the original audio,
s2, adding disturbance in a low-frequency area in the selected attack area [ S: e ], determining an updating step length by calculating a gradient direction, updating the disturbance, and obtaining a next iteration countermeasure sample by utilizing a binary query algorithm until the set sampling times are completed, so as to obtain a final countermeasure sample x.
The black box attack algorithm is applied to the field of speaker recognition, the query cost is high, the whole attack flow is shown in fig. 1, the attack flow is divided into two steps, the first step is to find the optimal local attack area so as to reduce the attack range and improve the query accuracy, and the second step is to add disturbance in the local low-frequency area of the audio by adopting the iterative query algorithm, because the voice energy of the speaker is gathered in the medium-low frequency area, and the attack can be effectively prompted in the low-frequency area. By this method, a high quality challenge sample can be produced more quickly.
For S1, it specifically includes,
s11, S takes a temporary value S curr For s curr Initializing the area with 0 to perform minimum function assignment to obtain a temporary end point e of the area curr =min (s+l α, l) and temporary audio x curr [s curr :e curr ]＝x t [s curr :e curr ]Wherein x is t For target audio, α represents the length ratio of the attack region to the original audio;
s12, performing a dichotomy query according to the distance between the temporary audio and the original audio, and judging the residual scale d between the dichotomy query and the original audio curr ＝||B(x,x curr )-x|| 2 Whether or not less than d and temporary audio x curr Output f (x after model f curr ) Whether or not equal to t, where d represents the size of the minimum disturbance previously searched, and t represents the tag of the targeted speaker;
s13, when d curr < d and f (x) curr ) When t, the current optimal disturbance scale d=d is updated curr Updating attack area s=s curr ，e＝e curr And s curr Repeating S12 after taking w as a sliding step length and taking the value next time until S curr Executing S14 when the total number of the components is greater than or equal to l;
s14, updating the contracted attack area alpha by 0.9 x alpha, and ending the selection of the attack area when the current flag bit is negative, wherein the last attack area is taken as the optimal attack area.
Since the timing signal of the audio is a one-dimensional signal, one second usually contains 16000 sampling values, which results in the audio sequence being too lengthy in a single dimension, and the original residual x is directly used t Taking x as the initial disturbance direction will make the initial disturbance be too large to reduce the optimization efficiency, so the invention adopts a continuously shrinking sliding window to search for an attack area with smaller disturbance, and the attack area is concretely shown in the following table frame, wherein B represents a binary query algorithm.
Algorithm B includes two inputs x 1 ，x 2 The goal of this algorithm is to find a linear dividing point gamma x between the two inputs 1 +(1-γ)x 2 Such that data around a division point, commonly referred to as the decision boundary of the model, is classified by the model into different results.
The above block diagram uses the distance of the decision boundary of the selected local region from the original audio as a selection criterion in order to make the resulting residual as small as possible. If a more suitable area is searched in a certain round of iteration, the algorithm displayed in the table frame reduces the size of the sliding window in the next round of sliding search so as to search for a smaller attack area, and if no more suitable area is searched in a certain round, the loop is jumped out, and the attack area obtained last time is taken as the optimal area.
The human speaking frequency is gathered at the middle and low frequencies, so that the voice information of the middle and low frequencies is more abundant, and therefore, the boundary attack algorithm is adopted to add low-frequency disturbance to the local area as an optimization process on the basis of local attack. The whole process is an iterative updating algorithm as shown in fig. 2, and each iteration is divided into three steps: including gradient direction estimation, step search, and binary search for decision boundaries.
In the step of gradient direction estimation, we estimate the current gradient direction using monte carlo plus finite difference estimation, as follows,
wherein H represents the number of samples, u h Representing a disturbance of the random sampling,representing the challenge sample at the ith iteration, e is a very small constant, the value 0.001, phi represents the indicator function for determining the gradient direction.
After inquiring the gradient direction, the step length updated along the gradient direction needs to be determined, and the step length xi is searched by using a grid searching step length method in the invention i Updating a disturbance formula after querying the step size as follows:
then go through the third stepA challenge sample was obtained for the i+1st iteration.
The corresponding block diagram algorithm is specifically as follows:
in short, S2 specifically comprises the following steps,
s21, setting an optimal attack area [ S: e ]]The low frequency region length of (c) is l' ≡β· (e-s), and the initial disturbance is against the sample
S22, in the process of [0,1]]Variable u with random sampling length of l h And at u h Post-filling l-l' length 0, using inverse discrete cosine transform IDCT variation to transform u h Conversion from the frequency domain back to the time domain u h ←IDCT(u h ) And nulling disturbances outside the attack area, i.e. u h [0:s)←0、u h (e:l]C, performing S23 when the sampling times are larger than H;
s23, through the formulaCalculating gradient and calculating step size xi by using grid search step length method i The disturbance is updated using the following formula>
S24, updating the countermeasure sampleObtaining a countercheck sample of the (i+1) th iteration until the value of i reaches the preset query times, and obtaining the countercheck sample by the last iteration as a final countercheck sample.
In this way, the obtaining process of the countermeasure sample is divided into two stages, the best attack area is obtained by adopting the binary inquiry algorithm and combining the sliding window algorithm in the first stage, the final countermeasure sample is obtained by combining the gradient direction calculation and the binary inquiry algorithm on the basis of the best attack area selected in the first stage and iterating, the scope of adding disturbance is reduced, the inquiry times of the attack are greatly reduced, the countermeasure sample is generated more quickly, a more fair and fair security evaluation means is provided for the speaker identification system,
the invention does not need to acquire information of a model or a training set based on an attack transferability assumption, and only attacks by accessing a prediction result of the model, so that details, privacy and the like of a user model are not involved in the evaluation process, the privacy of the user model is protected while the safety of the model is evaluated, the evaluation capability of the safety of the model is enhanced, and the problem of blank black box attack in the field of speaker identification is solved.
While embodiments of the invention have been shown and described, it will be understood by those skilled in the art that: many changes, modifications, substitutions and variations may be made to the embodiments without departing from the spirit and principles of the invention, the scope of which is defined by the claims and their equivalents.
Claims (2)
1. A method for quickly obtaining a voice challenge sample under a black box attack, the challenge sample being used for fair evaluation of model security, the method comprising: the method may include the steps of,
s1, determining a decision boundary of an original audio x by adopting a bipartite query algorithm, and iterating by matching with a sliding window method to select an optimal attack area [ S: e ], wherein the initial value of S is 0, the initial value of e is l, and l is the length of the original audio;
the step S1 specifically includes the steps of,
s11, S takes a temporary value S curr For s curr Initializing the area with 0 to perform minimum function assignment to obtain a temporary end point e of the area curr =min (s+l α, l) and temporary audio x curr [s curr :e curr ]＝x t [s curr :e curr ]Wherein x is t For target audio, α represents the length ratio of the attack region to the original audio;
s12, performing a dichotomy query according to the distance between the temporary audio and the original audio, and judging the residual scale d between the dichotomy query and the original audio curr ＝||B(x,x curr )-x|| 2 Whether or not less than d and temporary audio x curr Output f (x after model f curr ) Whether or not is equal to t, wherein B represents a bipartite query algorithm, d represents the size of the minimum disturbance searched before, and t represents the tag of the target speaker;
s13, when d curr < d and f (x) curr) When t, the current optimal disturbance scale d=d is updated curr Updating attack area s=s curr ，e＝e curr And s curr Repeating S12 after taking w as a sliding step length and taking the value next time until S curr Executing S14 when the total number of the components is greater than or equal to l;
s14, updating the contracted attack area alpha by 0.9 x alpha, and ending the selection of the attack area when the current flag bit is no, wherein the last attack area is used as the optimal attack area;
s2, adding disturbance in a low-frequency area in the selected attack area [ S: e ], determining an update step length by calculating a gradient direction, updating the disturbance, and acquiring a next iteration countermeasure sample by utilizing a binary query algorithm until the set sampling times are completed, so as to obtain a final countermeasure sample x;
the step S2 specifically comprises the following steps,
s21, setting an optimal attack area [ S: e ]]The low frequency region length of (c) is l' ≡β· (e-s), and the initial disturbance is against the sampleBeta is the low frequency area duty cycle;
s22, in the process of [0,1]]Variable u with random sampling length of l h And at u h Post-filling l-l' length 0, using inverse discrete cosine transform IDCT variation to transform u h Conversion from the frequency domain back to the time domain u h ←IDCT(u h ) And nulling disturbances outside the attack area, i.e. u h [0:s)←0、u h (e:l]C, performing S23 when the sampling times are larger than H;
s23, through the formulaThe gradient is calculated and the gradient is calculated,
wherein the method comprises the steps ofH represents the number of samples, u h Representing a disturbance of random sampling ∈>Representing the challenge sample at the ith iteration, e being a very small constant, the value 0.001, phi representing the indicator function for determining the gradient direction; and calculates the step size xi by using a grid search step size method i The disturbance is updated using the following formula>
S24, updating the countermeasure sampleObtaining a countercheck sample of the (i+1) th iteration until the value of i reaches the preset query times, and obtaining the countercheck sample by the last iteration as a final countercheck sample.
2. The method for quickly acquiring a voice challenge sample under a black box attack according to claim 1, wherein:
the value of beta is 0.65.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210106435.9A CN114661940B (en) | 2022-01-28 | 2022-01-28 | Method suitable for quickly acquiring voice countermeasure sample under black box attack |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210106435.9A CN114661940B (en) | 2022-01-28 | 2022-01-28 | Method suitable for quickly acquiring voice countermeasure sample under black box attack |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114661940A CN114661940A (en) | 2022-06-24 |
| CN114661940B true CN114661940B (en) | 2023-08-08 |
Family
ID=82026211
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210106435.9A Active CN114661940B (en) | 2022-01-28 | 2022-01-28 | Method suitable for quickly acquiring voice countermeasure sample under black box attack |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114661940B (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110321790A (en) * | 2019-05-21 | 2019-10-11 | 华为技术有限公司 | The detection method and electronic equipment of a kind of pair of resisting sample |
| CN110444208A (en) * | 2019-08-12 | 2019-11-12 | 浙江工业大学 | A kind of speech recognition attack defense method and device based on gradient estimation and CTC algorithm |
| CN110992934A (en) * | 2019-10-28 | 2020-04-10 | 浙江工业大学 | Defense method and defense device for black box attack model of voice recognition system |
| CN111507384A (en) * | 2020-04-03 | 2020-08-07 | 厦门大学 | Method for generating confrontation sample of black box depth model |
| CN111709435A (en) * | 2020-05-18 | 2020-09-25 | 杭州电子科技大学 | Countermeasure sample generation method based on discrete wavelet transform |
| CN112017669A (en) * | 2020-11-02 | 2020-12-01 | 鹏城实验室 | Voice countercheck sample detection method and device, terminal equipment and storage medium |
| CN112200257A (en) * | 2020-10-16 | 2021-01-08 | 支付宝(杭州)信息技术有限公司 | Method and device for generating confrontation sample |
| CN112216296A (en) * | 2020-09-25 | 2021-01-12 | 脸萌有限公司 | Audio anti-disturbance testing method and device and storage medium |
| CN113571067A (en) * | 2021-06-21 | 2021-10-29 | 浙江工业大学 | Voiceprint recognition countermeasure sample generation method based on boundary attack |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11222651B2 (en) * | 2019-06-14 | 2022-01-11 | Robert Bosch Gmbh | Automatic speech recognition system addressing perceptual-based adversarial audio attacks |
| US11768932B2 (en) * | 2019-06-28 | 2023-09-26 | Baidu Usa Llc | Systems and methods for fast training of more robust models against adversarial attacks |
-
2022
- 2022-01-28 CN CN202210106435.9A patent/CN114661940B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110321790A (en) * | 2019-05-21 | 2019-10-11 | 华为技术有限公司 | The detection method and electronic equipment of a kind of pair of resisting sample |
| CN110444208A (en) * | 2019-08-12 | 2019-11-12 | 浙江工业大学 | A kind of speech recognition attack defense method and device based on gradient estimation and CTC algorithm |
| CN110992934A (en) * | 2019-10-28 | 2020-04-10 | 浙江工业大学 | Defense method and defense device for black box attack model of voice recognition system |
| CN111507384A (en) * | 2020-04-03 | 2020-08-07 | 厦门大学 | Method for generating confrontation sample of black box depth model |
| CN111709435A (en) * | 2020-05-18 | 2020-09-25 | 杭州电子科技大学 | Countermeasure sample generation method based on discrete wavelet transform |
| CN112216296A (en) * | 2020-09-25 | 2021-01-12 | 脸萌有限公司 | Audio anti-disturbance testing method and device and storage medium |
| CN112200257A (en) * | 2020-10-16 | 2021-01-08 | 支付宝(杭州)信息技术有限公司 | Method and device for generating confrontation sample |
| CN112017669A (en) * | 2020-11-02 | 2020-12-01 | 鹏城实验室 | Voice countercheck sample detection method and device, terminal equipment and storage medium |
| CN113571067A (en) * | 2021-06-21 | 2021-10-29 | 浙江工业大学 | Voiceprint recognition countermeasure sample generation method based on boundary attack |
Non-Patent Citations (1)
| Title |
|---|
| Yash Sharma等.On the Effectiveness of Low Frequency Perturbations.《IJCAI'19: Proceedings of the 28th International Joint Conference on Artificial Intelligence》.2019,第3389–3396页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114661940A (en) | 2022-06-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ren et al. | Sinusoidal parameter estimation from signed measurements via majorization–minimization based RELAX | |
| CN108614992B (en) | Hyperspectral remote sensing image classification method and device and storage device | |
| CN111709435B (en) | Discrete wavelet transform-based countermeasure sample generation method | |
| JP6070956B2 (en) | Similarity detection apparatus and directivity vicinity detection method | |
| KR20040005895A (en) | Image retrieval using distance measure | |
| CN110297218B (en) | Method for detecting unknown modulation mode of radar signal based on generation countermeasure network | |
| CN112163145B (en) | Website retrieval method, device and equipment based on editing distance and cosine included angle | |
| CN112861066B (en) | Machine learning and FFT (fast Fourier transform) -based blind source separation information source number parallel estimation method | |
| CN117076941A (en) | Optical cable bird damage monitoring method, system, electronic equipment and readable storage medium | |
| CN117116290A (en) | Method and related equipment for positioning defects of numerical control machine tool parts based on multidimensional characteristics | |
| CN117033657A (en) | Information retrieval method and device | |
| CN114661940B (en) | Method suitable for quickly acquiring voice countermeasure sample under black box attack | |
| CN114911846A (en) | FAD and DTW-based hydrological time sequence similarity searching method | |
| CN112711032B (en) | Radar target detection method and system based on graph data and GCN | |
| JP3507083B2 (en) | Polynomial filters for high-order correlation and multi-input information integration | |
| CN117119377A (en) | Indoor fingerprint positioning method based on filtering transducer | |
| Grycuk et al. | A novel method for solar image retrieval based on the parzen kernel estimate of the function derivative and convolutional autoencoder | |
| CN113177078B (en) | Approximate query processing algorithm based on condition generation model | |
| Gao et al. | Aspects of 2D-adaptive Fourier decompositions | |
| JP4460277B2 (en) | Corresponding point search method of image, corresponding point search device, and corresponding point search program | |
| CN113610942B (en) | Pulse waveform segmentation method, system, equipment and storage medium | |
| CN112329589B (en) | Face recognition method based on high-dimensional feature algorithm | |
| CN117235137B (en) | Professional information query method and device based on vector database | |
| CN117877525B (en) | Audio retrieval method and device based on variable granularity characteristics | |
| Zhao et al. | Understanding and Improving the Intermediate Features of FCN in Semantic Segmentation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |