57 results about "Pseudo data" patented technology

The invention belongs to the field of federated learning technology application, and particularly relates to a knowledge distillation-based self-adaptive personalized federated learning method, which comprises the following steps that: a client generates pseudo data according to local data through a variational automatic encoder, and performs data enhancement on the pseudo data; carrying out intensive training on a teacher model deployed in the server by utilizing the enhanced data; the student model deployed at the client generates a soft prediction label based on an input sample and uploads the soft prediction label to the server, and the server dynamically weights the soft prediction label of each client based on the confidence of the soft prediction label; performing further iterative training by using the dynamically weighted soft label to generate a personalized label, and issuing the personalized label to a corresponding client; the client constructs distillation loss according to the received personalized label to train a student model; according to the method, the federal model performance can be improved in a federal learning scene of data isomerism and model isomerism, the personalized learning effect of the client is enhanced, and the personalized demand of the client is met.

A method, computer system, and a computer program product for data-free knowledge amalgamation are provided. Multiple pre-trained teacher machine learning models are obtained. Each is trained on a respective different set of training data. Pseudo-data samples that mimic original training data of the teacher models are generated. A block-wise amalgamation with a self-regulative strategy to integrate knowledge from the multiple teacher models is implemented by inputting the pseudo-data samples into the teacher models and into a student machine learning model. The implementing also includes aligning intermediate representations of the student model with a unified representation capturing relevant features from the teacher models.

The embodiment of the invention provides an encryption service and key authorization method for preventing encrypted data from going out of a domain. The method comprises the following steps: a data owner applies for a key from an encryption service, manages key authorization, and authorizes a corresponding key ID to a data user; during encryption, the data user generates random pseudo data; the encryption service carries out encryption by using the authorization key according to random pseudo data given by a data user and returns a first encryption result; and the data user carries out calculation according to the first encryption result and the plaintext data to generate a corresponding ciphertext and a message identification code. During decryption, the encryption service performs encryption calculation on partial data and random pseudo data of the first encryption result by using the authorization key and returns a second encryption result; the data user performs calculation according to the second encryption result and the ciphertext data to generate an initial plaintext; if the initial plaintext passes the data verification, outputting the plaintext data, otherwise, decrypting mistakenly.

The invention relates to the technical field of artificial intelligence, discloses a hierarchical adaptive freezing fine tuning method and device for a multi-modal large model, and particularly discloses the method comprising the following steps: training a pre-trained multi-modal large model according to a total data set subset to obtain a field expert model; dividing the domain expert model into a plurality of structural layers, and distributing global unfreezing parameter quotas for each structural layer according to contribution degrees; and in each structural layer, performing adaptive importance sorting on each parameter, screening out parameters consistent with the corresponding global unfreezing parameter quota quantity according to an importance sorting sequence to generate a binary mask, and freezing other parameters. And finally, performing sparse training on the domain expert model according to the binary mask. According to the method, the dependence of traditional data playback on old task data or pseudo data generation is overcome, task boundary presetting or manual intervention is not needed, and efficient, lightweight and privacy-safe continuous learning is realized while the generalization ability of the pre-training model is protected.

The invention discloses an integrated circuit for encryption, encryption equipment and an encryption method, and relates to the technical field of encryption, the integrated circuit updates a pseudo state vector according to a pseudo data set when an encryption module does not receive a target data set, so that power consumption still exists when the encryption module is in a waiting state; and the power consumption is not the power consumption of the encryption algorithm and is equivalent to interference on the power consumption of the encryption module, so that the encryption algorithm operated by the encryption module cannot be deduced based on methods such as DPA. Therefore, the exposure risk of the encryption algorithm is reduced, the security of the encryption algorithm is improved, and the problem that the security of the encryption algorithm in some technologies is reduced is solved.

The invention belongs to the technical field of natural language processing, and discloses a domain perception simplified data set construction method for sensitive information identification. According to the method, a structure template is constructed through field seed driving and is used for synthesis and generation of large-scale pseudo data, and logits and attention output by a teacher model are adopted as distillation signals; the expression stability and the boundary perception ability of the pseudo tag are improved by carrying out numerical cutting on logits and carrying out asymmetric reconstruction on attention distribution; the student model is trained in a multi-target joint optimization mode such as soft label fitting and attention alignment, and finally seed self-feedback updating is realized on the basis of high-confidence pseudo label screening, so that multiple rounds of iterative learning are completed. The method has the characteristics of light weight, low dependence and high migration, and is suitable for sensitive information automatic extraction and cross-domain entity recognition tasks.

The objective is to improve the accuracy of the generated abnormal signal data. [Solution] The signal management device 1 includes a generator that generates pseudo-data statistically similar to true data, with each of the sequences of intensity for each frequency component corresponding to the second data being treated as true data, and a discriminator that identifies whether the pseudo-data generated by the generator belongs to the class corresponding to true data or the class corresponding to pseudo-data. The device also includes a third learning unit configured to learn multiple generative models, each having an objective function set by the probability distributions of the first data and the second data determined from the ratios estimated by the first learning unit, and the second parameter estimated by the second learning unit.

When an application execution error occurs, an entry for the execution error is included in an error log. In instances where user data associated with the error should be preserved, a placeholder corresponding to the error is included in the error log. The placeholder contains pseudo data representing the user data, but is formatted in such a way to prevent the vendor from accessing the content of the data without additional authorization. When reviewing the log file, the vendor has visibility to the pseudo data contained in the placeholders, but is not able to discern the content of the user data from the pseudo data. In response to a request from the vendor to access instances of user data, the customer can selectively provide access to particular instances of user data that are deemed to not contain sensitive information.

The invention belongs to the technical field of equipment monitoring, and particularly relates to a bearing state recognition method and system based on pseudo data, a terminal and a storage medium, and the method comprises the steps: obtaining historical bearing vibration data, carrying out the data processing, and constructing an initial data set; inputting the initial data set into the constructed TimeVQ-VAE model, generating pseudo data of the bearing state, and constructing an enhanced training set based on the pseudo data; training a BiIndRNN (Bidirectional Independent Recurrent Neural Network) model by utilizing the enhanced training set; and adopting a particle swarm optimization (PSO) algorithm to obtain an optimal hyper-parameter combination of the BiIndRNN model. According to the method, the TimeVQ-VAE model is introduced, so that high-quality pseudo data (the similarity index u is greater than or equal to 0.9) highly similar to real data can be generated, model training difficulty caused by scarcity of fault samples and imbalance of the number of various types of samples in an industrial field is effectively relieved, and dependence on a large amount of real labeled data is reduced.

A game-based Mongolian neural machine translation method converts problematic factors that affect translation decoding in Mongolian neural machine translation into interference noise in the form of pseudo data. A game model based on a generative adversarial network is constructed based on a noise interference strategy. The model consists of a generator G, an anti-interference device AJ, and a discriminator D. The model generalizes to the interference noise in the three-module game, improving the translation performance of the game model against this noise. During game training, the REINFORCE strategy is used to optimize the parameters of G to obtain training rewards for original data and noise data. The original corpus and pseudo corpus containing interference noise are trained using an adversarial training mechanism. AJ is used to determine the training priority of the noise data based on the judgment of the training data, so that G and D have sufficient decoding capabilities for the added interference noise. Finally, the trained G is used as the translation model for Mongolian translation. This invention can improve the accuracy and efficiency of Mongolian translation.

A forward-direction model for predicting a processing result for a substrate from processing conditions pertaining to the substrate is generated on the basis of a plurality of actual datasets. Using the forward-direction model, a pseudo processing result for the substrate is predicted from individual pseudo processing conditions pertaining to a plurality of substrates, whereby a plurality of pseudo datasets are generated. A reverse-direction model for predicting a processing condition pertaining to the substrate from a processing result for the substrate is generated on the basis of at least some of the plurality of pseudo datasets.

The objective is to improve the accuracy of the generated anomalous traffic data. [Solution] The communication management device 1 includes a generator 131 that generates pseudo-data statistically similar to true data, using each of the series of communication volume observed for each time period corresponding to the second data as true data, and a discriminator 132 that identifies whether the pseudo-data generated by the generator 131 belongs to the class corresponding to true data or the class corresponding to pseudo-data. The device also includes a third learning unit 13 configured to learn multiple generative models, each having an objective function set with the probability distribution of the second data and the probability distribution of the first data determined from the ratio estimated by the first learning unit 11, and the second parameter estimated by the second learning unit 12.

The present application provides a kind of network security system based on virtual environment, and the pseudo database constructed with general honeypot has significant difference.In the network security system provided by the present application, the original data is classified according to the relevance (i.e., sensitivity) of the log record before the attacker, and different intensities of noise are introduced to different dimensions of data with different sensitivities in different ways, so that the different dimensions of data in a data string in the finally formed pseudo database have different camouflage effects, which not only ensures that the sensitive dimension part is completely hidden to ensure data security, but also ensures that the insensitive dimension part is more confusing to network attackers due to the weak noise.

The invention relates to a pseudo data writing method and device, a medium and electronic equipment, and belongs to the technical field of storage. The pseudo data writing method comprises the following steps: determining the priority of a superblock in which pseudo data is to be written; and based on the priority, writing the pseudo data into the superblock in which the pseudo data is to be written. By adopting the technical scheme, the pseudo data can be preferentially written into the super block with the high priority to be written with the pseudo data, so that the stable pseudo data writing throughput can be ensured, and the free super block (namely the super block which can be rewritten) can be quickly and efficiently generated.

The present invention discloses a method and system for protecting public data of an industrial system, including: sorting and selecting public data of the industrial system; training a pseudo-data generator based on the original public data set; training a data discriminator based on the original public data set and a generative adversarial network; designing a hierarchical protection method for public data for objects with different security levels based on a noise addition strategy, and implementing equal distribution of public data for objects with different security levels and subsequent content restoration for high-security-level objects based on this method; the present invention is simple to implement and has a low computational complexity, effectively reducing the computational resource overhead during the later use of the model through the pre-training of the model. The present invention realizes that while public data is disclosed, it can prevent untrusted third parties from obtaining higher-security-level information through means such as inference inversion, so that the key information hidden in the disclosed data is not leaked, and through the differential hierarchical protection method, it ensures the complete acquisition of the real data content by the trusted party.

The present invention is a pseudo-data creation system (S1) for creating pseudo-data (D1) pertaining to a pseudo buried pipe (T1), the pseudo-data creation system comprising: a reflection profile point calculation unit (P1) for calculating a reflection profile point (T4), which is the foot of a vertical line drawn from a pseudo search point (T3) to a profile line (T2); a reference reflection profile point calculation unit (P2) for calculating a reference reflection profile point (R4), which is the foot of a vertical line drawn from a reference search point (R3) to a reference profile line (R2); a reference pseudo search point calculation unit (P3) for calculating a reference pseudo search point (T3'), which is a point where transformation for matching a minute line segment of the reference profile line (R2) at the reference reflection profile point (R4) to a minute line segment of the profile line (T2) at the reflection profile point (T4) is used in the pseudo search point (T3); and a buried pipe signal calculation unit (P4) for calculating a pseudo buried pipe signal at the reference pseudo search point (T3') and associating the pseudo buried pipe signal with the pseudo search point (T3) to calculate the pseudo-data (D1).

The invention relates to a Vietnamese dependency syntax analysis method and system based on multiple annotators and electronic equipment, and belongs to the field of natural language processing. According to the method, the parameters of the XLM RoBERTa model are finely adjusted by using the unlabeled data. Then training an initial Vietnamese dependency syntactic analysis model by using a UD tree library, inputting sentences into the model for analysis, and generating pseudo data containing noise; and inputting the pseudo data into a pre-designed prompt template, carrying out secondary annotation by utilizing a large language model DeepSeek, gradually correcting noise data through prompt learning, and outputting a high-quality annotation result. And finally, taking the pseudo data subjected to secondary annotation as an additional training corpus, combining with the original annotation data, and training a new Vietnam statement method analysis model. According to the method, the performance of the model in Vietnamese dependency syntax analysis tasks is remarkably improved by introducing pseudo data enhancement and a multi-tagger collaborative optimization mechanism.

A system for implementing data security using dynamic data fragmentation and storage is disclosed. The system accesses digital content and splits the digital content into a set of data shards according to a splitting rule. The splitting rule indicates that each data shard combined with a respective pseudo data shard should appear as the digital content. The system generates a pseudo data shard such that the first data shard together with the pseudo data shard appears as the digital content. The pseudo data shard comprises pseudo information that is a counterpart to the original information indicated in other data shards. The system communicates the first data shard and the pseudo data shard to a database.

The present invention relates to fields related to data security, and discloses an audio encryption transmission method and system based on DNA coding. By setting several fuse sequences that do not contain the original transmission data content, and randomly interspersing them between sequence segments that carry the transmission data content, the proportion of pseudo data is increased, so that the difficulty of decryption can be increased. In the fuse sequence, the sequence data records a preset fuse program segment, which is encrypted by a random interrupt key. When the decoder performs strong decryption, the corresponding interrupt key can be randomly matched, so that the fuse program segment is decrypted by the corresponding fuse sequence, and then the fuse program segment is triggered to destroy and eliminate the data in the current storage area. Not only can the decryption process be interrupted, but also the basic data and related decryption intermediate data can be cleared, hindering the decryption work to protect data security.

The application provides a data enhancement method for Chinese text proofreading, and relates to the technical field of artificial intelligence. The method judges the position and type of errors prone to occur in the correct source sentence through a sequence labeling model, makes up for the defects of current methods of randomly selecting error positions and error types, and makes the data closer to existing training data; in generating multi-word errors, syntax error data generated by using the model BERT is added, so that the semantic correlation of the generated error sentences is stronger; in the process of generating spelling errors, syntax error data generated by using the model BERT is added to simulate the situation of vocabulary selection errors in writing; at the same time, the spelling errors caused by pressing the wrong key when entering the text by using the keyboard in the real entry process are considered; the generated pseudo data contains common syntax error types, which can improve the robustness of the syntax correction model and the spelling correction model to a certain extent, and make the model learn more diverse and similar error sentence features to the real data.

The invention discloses a Mongolian-Chinese neural machine translation method fusing a pre-training model and pseudo data enhancement, the model adopts a shared encoder and bidirectional decoder architecture, the shared encoder is composed of a fusion encoding layer and a structure sensing layer, and the bidirectional decoders respectively correspond to Mongolian-Chinese and Chinese-Mongolian translation directions; the fusion coding layer is used for extracting semantic features by adopting MongolBERT and XLM-R-Mongolian respectively, and performing linear weighted fusion through a learnable weight so as to generate cross-language semantic representation; a lightweight syntactic information integration mechanism is introduced into the structure perception layer, syntactic distance is embedded into position coding, and the structure perception capability is enhanced in combination with a relative position attention mechanism. In the training stage, a staged multi-task learning strategy is adopted, and Mongolian-Chinese bilingual data generated by a practical pseudo-parallel corpus enhancement method is used for model optimization. Complementation is formed in the aspects of model structure, data expansion, multi-task optimization and the like, and an extensible, efficient and robust solution is provided for neural machine translation under the low-resource condition.

The information processing device 1X is an AI based device to support a decision making and includes an acquisition means 20X, a calculation means 22X, a selection means 23X, and a generation means 24X. The acquisition means 20X is configured to acquire a plurality of sets each including a sample and a label. The calculation means 22X is configured to calculate, for each of the plurality of sets, accuracy of the label based on a difference between the label and a predicted result of the label predicted from the sample paired with the label. The selection means 23X is configured to select a set to be used for generating pseudo data from the plurality of sets, based on the accuracy. The generation means 24X is configured to generate the pseudo data, based on the selected set.

The invention discloses a federated learning model protection method based on a super network, and relates to the technical field of federated learning and privacy protection. According to the invention, a super network is adopted to generate a global model and a local model for a server and each client; in order to better capture unique data distribution characteristics between a server and different clients, an embedded network is adopted to generate descriptors, and the descriptors are used as input of a super network; in addition, the server is lack of data, and a generator is trained by adopting a data-knowledge-free distillation method, so that a generated pseudo data set can reflect global data features. According to the method, the privacy of the model can be effectively protected while high performance is guaranteed; in addition, the method still retains the capability of generating a high-quality local model for the client.

This invention discloses a method for generating tabular data based on local context retrieval and fine-tuning, comprising: obtaining raw tabular data samples for machine learning training from a public benchmark database, performing preprocessing and data partitioning to obtain feature columns and label columns, and constructing a training set; fine-tuning the local calibration parameters of a TabPFN model based on the training set to obtain a parameter-fine-tuned TabPFN model; obtaining pseudo-data samples based on the training set by manually setting the target number of samples and target category labels; and iteratively updating the pseudo-data samples and the parameter-fine-tuned TabPFN model using an iterative sampling method based on dynamic local context to finally obtain a tabular dataset. This invention, through local calibration parameter fine-tuning and dynamic local context iterative sampling, can generate high-quality, diverse tabular datasets, effectively improving the realism and usability of the generated data.

The invention discloses a federal learning backdoor attack defense method and system based on SAGAN and fine tuning. The method comprises the following steps: S1, generating pseudo data through SAGAN; s2, cleaning pseudo data, and constructing a fine tuning data set; s3, finely adjusting the global model; s4, dynamically adjusting the weight of the regular term; s5, gradient calculation and parameter updating; and S6, iterative training is carried out. According to the method, the self-attention generative adversarial network is introduced to the server side to generate pseudo data, so that the problem caused by the fact that the server cannot directly access real data of the client side in federated learning backdoor defense is effectively solved, and a clean classification model can be constructed to perform data cleaning only by using a small amount of clean data; the dependency on real data is greatly reduced, and the privacy protection capability is enhanced.

The present application belongs to the field of man-machine conversation technology in natural language processing, and particularly relates to a knowledge-centered reply screening method and system in an open domain conversation field. Step 1, constructing required training data by using existing conversation data and encyclopedic knowledge data, and constructing a pseudo data algorithm; Step 2, performing three-stage training on a BERT model by using the training set constructed in step 1; Step 3, using the sum of scores of the three stages in step 2 as the final score during testing. The present application is used to solve the problems of low knowledge screening accuracy and inconsistency between knowledge screening results and screened replies existing in the prior art non-structured knowledge enhancement method.

To provide an information processing device capable of predicting what type of data content is combined with data that has not been provided for data combination.SOLUTION: An information processing device 1 includes: a generation unit 132 that, when either first data obtained from a first business operator or second data obtained from a second business operator is received as input, generates a data generation model having a plurality of latent variables, the data generation model for outputting pseudo-integrated data, the integrated data associating the input first data or second data with the pseudo data corresponding to non-input data, the data that has not been input, by generating values that can be taken by latent variables corresponding to the non-input data; and a transmission unit 133 that transmits the data generation model generated by the generation unit 132 to at least any of a first device 2 used by the first business operator and a second device 3 used by the second business operator.SELECTED DRAWING: Figure 2

The invention discloses a transmission system cross-sensor collaborative diagnosis method based on dynamic competitiveness balance. The method comprises the following steps: acquiring monitoring data of sensors in a transmission system and preprocessing the monitoring data; the preprocessed monitoring data is input as a sample set, a cross-sensor collaborative distillation network is constructed, and the cross-sensor collaborative distillation network comprises a teacher model and a student model; the teacher model and the student model execute standard judgment respectively, and a heterogeneous sensor dynamic competitiveness balancing mechanism is constructed after the standard is reached; iteratively training a student model through optimization of distillation loss and dynamic competitiveness balance mechanism regularization loss; based on the optimal model obtained by iterative training, inputting the monitoring data to be diagnosed into the model for diagnosis, and outputting a corresponding diagnosis result; based on a cross-sensor collaborative distillation network, association between multi-sensor fusion fault features and single-sensor local features is established, pseudo data does not need to be additionally generated in the whole process, and missing data can effectively participate in model training.

The invention relates to a cross-language dependency syntactic analysis method based on large model migration and self-optimization synthetic data enhancement, and belongs to the field of natural language processing. Firstly, on one hand, a fine-grained syntactic analysis instruction is designed to guide a large model to recognize cross-language syntactic generality and difference information so as to generate pseudo data based on the large model; on the other hand, a cross-language syntax parser is trained and is also used for generating pseudo data based on the cross-language syntax parser; secondly, designing an iterative self-optimization algorithm to enable the large model to fuse the advantages of the two types of pseudo data, so as to obtain high-quality synthetic data; according to the method, organic fusion of a large-model semantic understanding advantage and a traditional model structure analysis advantage is effectively realized, syntactic generality and difference of different languages are deeply mined, the accuracy of target low-resource language dependency analysis is remarkably improved, a high-quality synthetic syntactic tree bank is constructed, and the method is suitable for large-scale and large-scale analysis. And more convenience is provided for cross-language tasks.

The invention relates to the field of data encryption, and provides a supervision submission-oriented data encryption storage method and system, and the method comprises the steps: firstly obtaining to-be-encrypted data, and carrying out the preprocessing of the to-be-encrypted data, and obtaining a sensitive field; secondly, performing pseudo data filling and reconstruction on the sensitive field to obtain a pseudo data field and a reconstruction packet, and recording a mapping relation among the sensitive field, the pseudo data field and the reconstruction packet; generating a verification code through a digital signature algorithm and the integrity check bit, and embedding the verification code into the reconstruction packet; and finally, symmetrically encrypting the reconstructed packet embedded with the verification code, and storing the encrypted reconstructed packet into a system database. The system comprises a data preprocessing module, a pseudo data filling module, a reconstruction packet construction module, a verification code generation module and an encryption module, and can realize security encryption storage, source authentication, integrity verification and tamper-proof verification of data on the premise of ensuring that sensitive data is not directly exposed, thereby meeting the requirements of supervision departments on the security and compliance of submitted data.