337 results about "White box" patented technology

The invention belongs to the technical field of bearing residual service life detection, and discloses a rolling bearing residual service life prediction method and system based on a convolution white box. According to the method, time domain and frequency domain feature extraction, noise reduction processing and health state and degradation state division are carried out on an original vibration signal, and deep neural network training is carried out in combination with a Weibull-MSE loss function, so that prediction of RUL conforms to the actual degradation process of a bearing; and synchronous extraction of long-time dependence information and local degradation characteristics of the bearing signal is completed through a convolution CRATE network architecture fusing an expansion causal convolution attention mechanism DCA and a multi-scale convolution MSC. According to the method, the accuracy of RUL prediction is improved in combination with health state evaluation, the local feature extraction capability is enhanced, the local modeling capability of the model is improved through multi-scale information fusion, and the interpretability and the prediction credibility are improved through a CRATE structure.

The invention discloses a large model end-to-end distillation deployment method and device for low-computing-power equipment, equipment and a medium, and relates to the field of artificial intelligence, and the method comprises the steps: in a computing cluster, determining a first target data set and a target student model, and deploying a distillation model training framework; deploying a preset large language model to a computing cluster by using a first preset large model reasoning framework, and determining the deployed preset large language model as a teacher model; if the distillation model training framework is a black box knowledge distillation framework, determining a second target data set based on the teacher model and the first target data set, and distilling the target student model by using the second target data set to obtain a distillation model; if the distillation model training framework is a white-box knowledge distillation framework, distilling the target student model based on a teacher model to obtain a distillation model; and deploying the distillation model to target equipment based on a second preset large model reasoning framework. Therefore, the efficiency of accessing the large language model can be improved.

The invention discloses a network flow confrontation defense method based on heuristic ant colony optimization, and belongs to the technical field of network security and privacy protection. According to the method, a self-adaptive disturbance generation strategy is designed according to spatial and temporal features of a network traffic trajectory, a disturbance injection rule meeting real-time constraint and direction consistency is established, an integrated substitution model is constructed, a feature fusion mechanism of spatial convolution, time sequence circulation and noise reduction coding is used, and a traffic fingerprint identification model with strong generalization ability is generated. An improved ant colony optimization algorithm is adopted to improve the disturbance search capability, and a batch processing method and a universal patch scoring mechanism are integrated to improve the defense universality of samples among different network flows. The method can be directly applied to a black box scene of real-time communication, gradient information or white box access permission of a target model is not needed, the attack recognition suppression rate exceeding 90% is achieved when only 20% of bandwidth overhead is increased, defense universality and deployment feasibility are remarkably improved, and meanwhile disturbance overhead is reduced.

The invention relates to a trajectory prediction model robustness enhancement method based on dynamic subspace projection decomposition. Comprising the following steps: firstly, extracting hidden layer semantic features containing historical tracks and map topology through a multi-modal feature encoder; secondly, constructing a dynamic routing mechanism based on scene self-adaption, and calculating projection weights of input features on a plurality of expert subspaces; then, executing truncation projection operation based on orthogonal decomposition, retaining core semantics located in a low-dimensional space, and filtering out adversarial disturbance located in an orthogonal complementary space; and finally, introducing a feature consistency constraint training mechanism, taking the reconstructed features of the clean sample as anchor points, and compulsively aligning the purified features of the confrontation sample with the anchor points. Compared with the prior art, the method has the advantages that the robustness of the model in white box gradient attack, black box query attack and physical semantic deception scenes is remarkably improved through feature purification of a physical level and structured consistency constraint, and the prediction reliability of the automatic driving system is ensured.

Typical autonomous systems implement black-box models for tasks such as motion detection and triaging failure events, and as a result are unable to provide an explanation for its input features. An explainable framework may utilize one or more explainable white-box architectures. Explainable models allow for a new set of capabilities in industrial, commercial, and non-commercial applications, such as behavioral prediction and boundary settings, and therefore may provide additional safety mechanisms to be a part of the control loop of automated machinery, apparatus, and systems. An embodiment may provide a practical solution for the safe operation of automated machinery and systems based on the anticipation and prediction of consequences. The ability to guarantee a safe mode of operation in an autonomous system which may include machinery and robots which interact with human beings is a major unresolved problem which may be solved by an exemplary explainable framework.

The invention discloses an intelligent power grid dispatching instruction optimization system based on semantic analysis, relates to the technical field of power grid dispatching instruction optimization, realizes the conversion of dispatching instruction generation from black box output to white box reasoning, constructs a new intelligent dispatching normal form with deep integration of knowledge driving and data driving, and improves the intelligent dispatching efficiency. The automation level, decision reliability and operation safety of power grid dispatching are remarkably improved, and the method has outstanding innovativeness and wide engineering application prospects.

The embodiment of the invention provides a method for determining a generative model. The method includes embedding a white box watermark and a black box watermark in a generative model. Black box watermarks are first embedded into probability density functions of data abstraction in different layers of the generative model. The method also includes embedding the white box watermarks into different layers for generative model output after the embedding of the black box watermarks is completed. Model data is generated by the generative model based on predetermined trigger data. The predetermined trigger data comprises a predetermined trigger text or a predetermined trigger image. An identity associated with the generative model is determined based on the model data. By using the method, white box and black box attacks can be resisted by embedding two complementary and independent watermarks, so that double-layer protection is provided for the generative model.

The invention provides a client identity authentication security evaluation method and system combined with adversarial sample simulation, and relates to the technical field of information security. According to the invention, end-cloud integrated evaluation is carried out under a white-box and black-box threat model for multi-mode identification objects of human face, voiceprint, fingerprint and behavior characteristics. The method comprises the steps of establishing reference performance through legal samples, generating transferable adversarial input, performing secure injection on a communication layer, calculating a robustness index, performing adaptive optimization, performing differential evaluation on protection effectiveness under protection configuration starting and non-starting, and performing risk grading and reinforcement according to a result. And finally, outputting a reproducible security and compliance conclusion through differential privacy and federal evaluation.

The invention provides a domain self-adaption-based generative attack and defense confrontation method. Directional interference on a black-box intelligent system is realized by constructing a generative confrontation network framework. The method takes a domain self-adaption theory in transfer learning as a core, builds a dynamic mapping relation between a white box model group and a black box target system, designs a dual-channel generative adversarial network, jointly supervises sample authenticity and cross-model transfer consistency through cross entropy loss and KL divergence constraint, builds a dynamic boundary feedback system, and achieves the dynamic boundary feedback. And a disturbance generation strategy of the generator is adjusted according to the real-time attack effect, and a self-adaptive attack and defense game closed loop is formed. Compared with a traditional black box attack method, the method has the advantages that migration robustness of an adversarial sample in a model heterogeneous scene is remarkably improved, stable attack efficiency can still be maintained when a defense system has protection mechanisms such as adversarial training and input reconstruction, and the method can be applied to the fields such as penetration testing of an intelligent security system and the like. And a technical support is provided for vulnerability evaluation of the intelligent system in a complex environment.

The invention discloses a performance prediction method for an intelligent computing system, and the method achieves the efficient and explainable performance prediction through mixed analysis modeling. The method mainly comprises the steps of performing tracking and programmed synthesis on a target deep learning model, and generating a final operator graph containing a complete calculation and communication operation sequence; a dynamic overlap analysis model based on available parallelism is adopted, independent calculation and memory access efficiency is predicted in combination with a machine learning method, and GPU operator execution delay is accurately estimated; based on a configurable network topology, applying a queuing theory network model, and carrying out white box type modeling on queuing and transmission delay of communication operation in each node in a network path; and the calculation and communication prediction delays of the operators are aggregated according to the scheduling characteristics of the distributed strategy to obtain the final system end-to-end delay. The method provided by the invention can ensure the precision while remarkably shortening the prediction time, and has the advantages of high efficiency, high interpretability and the like.

The present disclosure relates to a system and method for automated verification of a battery management system for a battery pack, an energy storage device, etc., and the problem to be solved is to enable automated verification for items that should be determined by modifying internal variable information or verifying results with the internal variable information during white box verification of the internal architecture logic of BMS firmware of an ESS. There is provided a system for automated verification of a BMS, including a device for automated verification configured to create a test case sequence for testing firmware operation items of the BMS, perform a simulation on the BMS, and model operation items of the BMS, a BMS controller that is connected to the device for automated verification and controls the BMS to operate by applying modeled data for the operation items to the BMS, and a BMS internal variable acquirer that acquires internal variable information of the BMS during execution of the simulation.

The invention discloses an iterative satellite internet confrontation method based on momentum, and mainly solves the problem of low success rate of black box migration attack in the existing method. According to the method, a momentum item is introduced in an iterative optimization process of adversarial sample generation and is used for recording a historical direction of gradient updating. According to the momentum iteration method, the gradient updating direction in the adversarial sample generation process can be more stable, meanwhile, the adversarial sample is prevented from falling into a local extreme point of an optimization problem, the mutual restriction of the white box attack effect and the migration capability of the adversarial sample generated by the existing method is effectively relieved, and the black box migration attack success rate is improved.

The invention discloses a blasting vibration and dust control-oriented multi-objective collaborative optimization parameter setting method, which comprises the following steps of: training a neural network model capable of predicting a complete vibration time travel curve and a three-dimensional dust concentration field on the basis of a training sample containing geological conditions, blasting parameters and an actual blasting effect; the model is used for carrying out multi-target optimization on a target function containing four indexes of vibration, dust, cost and regulation compliance through a preset evolutionary algorithm, a Pareto optimal solution set composed of a plurality of non-dominated parameter schemes is obtained, the schemes in the optimal solution set are analyzed through a preset explanation algorithm, and the optimal solution set is obtained. The influence and contribution degree of each blasting parameter on a prediction result are quantified, so that a black box model is converted into an interpretable white box decision; according to the method, the problems that a traditional method is difficult to balance multiple conflict targets and the decision-making process is opaque are solved, and an optimal blasting parameter scheme considering safety, environmental protection and cost effectiveness is provided for blasting operation.

The invention relates to a terminal software comprehensive analysis method and system based on artificial intelligence, and the method comprises the steps: obtaining the hardware configuration information of a terminal device and the performance index data of terminal software, carrying out the feature extraction, carrying out the clustering analysis in combination with a historical sample, and generating a software parameter configuration suggestion through a reinforcement learning tuning model, and further performing security evaluation through a white box constraint rule and a black box performance prediction model, and finally generating a software optimization scheme and monitoring performance change to optimize the model. According to the method, the internal structure and the potential mode of the obtained data can be mined, intelligent adjustment and optimization are realized in different hardware environments, meanwhile, dynamic optimization of safety and performance is considered, and the operation efficiency and the stability of terminal software are improved.

The invention provides an adversarial sample generation method and system based on a dynamic pseudo tag, and the method comprises the steps: calculating the classification confidence distribution of a sample x based on a white box model, and determining a misjudgment category with the highest confidence as a pseudo tag; calculating a synthetic gradient of a real label and the pseudo label loss function; based on the obtained synthetic gradient, obtaining an adversarial sample xadv; and updating the sample x by using the obtained adversarial sample xadd, repeating the steps, iterating for a preset number of times T, and outputting an adversarial sample xadv. According to the method, the mobility and the generation efficiency of the adversarial sample are improved.

The invention discloses an AI-driven full-scene intelligent test platform which comprises the following steps: performing static scanning on a target code under the triggering of a code submission event through a code static scanning module to generate a scanning result, and then inputting the scanning result into a pre-trained first AI model to obtain code quality evaluation data. And generating a white box test case through a white box case generation module. In addition, a black box test case is generated through a black box case generation module. An automatic script generation module inputs the black box test case into a pre-trained fourth AI model to generate an automatic test script; and finally, running the white box test case and the automatic test script to obtain an original test report. And inputting the original test report into a pre-trained fifth AI model through a result analysis module, and generating defect distribution data and test suggestion data. The full-scene test coverage of the software life cycle is realized, the test of each link of the software is comprehensively and systematically covered, and the test efficiency is improved.

The invention relates to the field of deep neural network security, in particular to a black box attack method based on gradient alignment and anti-distillation. Comprising the following steps: step 1, designing a sample disturbance generator with a maximized gradient difference; 2, designing a gradient alignment anti-distillation modeling framework for the substitution model; 3, alternately training the substitution model and the disturbance generator; and 4, completing the attack of the black box model based on a white box attack method and the substitution model. Experimental results on different data sets and models show that compared with a traditional static distillation method, a dynamic distillation method based on Gaussian distribution random disturbance and a dynamic distillation method based on adversarial disturbance, the method has the advantages that the adversarial success rate is remarkably improved, and certain innovativeness and universality are achieved.

The invention provides an intention-driven semantic evolution mechanism based on a tensor semantic field and an application system thereof, and relates to the technical field of artificial intelligence, semantic networks and cognitive computing. According to the method, five types of semantic primitives including data, information, knowledge, intelligence and intention are expressed as high-order tensor nodes, a semantic tensor field network is constructed, and dynamic evolution and intention driving of semantics are achieved. The system comprises a multi-scale semantic aggregation mechanism, an intention weight diffusion algorithm, a semantic tensor evolution operator and a white box interpretation interface, supports full-link semantic processing from original data to high-level wisdom to intention constraint, and overcomes the defects in semantic representation and evolution, intention fusion and system interpretability in the prior art. And the generative AI system has stronger intention perception, semantic self-optimization and process transparency capabilities, and is suitable for applications such as a semantic perception large model platform, an AI cognitive map system and an interpretable language generator.

The invention relates to the technical field of chips, and provides an interrupt module verification method and device, a medium and a program product, the method comprises the following steps: when it is monitored that an interrupt module generates an external observable behavior, comparing monitored behavior information with a reference value of the external observable behavior, and verifying the correctness of the external observable behavior; performing state calculation on the interrupt request input to the interrupt module, and determining an internal processing state of the interrupt request; an internal register state of the interrupt module is verified based on an internal processing state of the interrupt request. According to the invention, through a mode of combining black box verification of external observable behaviors with white box verification based on an internal processing state, comprehensive, deep and efficient function verification of a complex interrupt module is realized.

The invention discloses a black box attack optimization method based on gradient prior information, and relates to the technical field of artificial intelligence model security, and the method comprises the steps: in a white box model scene, obtaining a target picture feature and an adversarial sample picture feature based on a target text and a clean picture, generating a migration confrontation sample picture and gradient prior information based on a random fast gradient descent method; calculating a similarity degree estimated value of gradient prior information and a real gradient of a black box model loss function according to a finite difference method, and calculating a synthesis proportion and a disturbance vector; inputting the migration adversarial sample picture into a black box model to obtain a picture description text, performing similarity comparison on a description text feature code and a target text feature code to obtain a loss function, updating disturbance in the migration adversarial sample picture through a stochastic gradient approximation estimation method based on the loss function, and outputting a final adversarial sample. And implementing a black box query attack. According to the method, the attack effect, the attack effect stability and the attack method generality can be improved.

The invention relates to a scene self-adaptive white-box prompt injection attack system. The system comprises a self-adaptive context prompt generation module, an enhanced GCG algorithm-based confrontation suffix generation module and a simulator-based intelligent agent test module with a body. The adaptive context prompt generation module constructs specific prompt templates for different scenes, and automatically identifies and encodes unique semantic features of different application scenes. And the adversarial suffix generation module based on the enhanced GCG algorithm calculates the coordinate gradient of each token position, provides direction guidance for subsequent token replacement based on a double-component loss function, and adopts a weighted combination loss function to realize system optimization. The simulator-based intelligent agent test module carries out semantic understanding and intention recognition through a target large language model, converts a natural language into a control instruction, recognizes different types of commands, and maps the analyzed commands to a specific API for calling.

The invention discloses an attribute-based encryption method and system based on support of multi-authority accountability and outsourcing calculation, and the method specifically comprises the steps: introducing a multi-attribute authority mechanism to be responsible for key distribution and management, reducing the dependence on a single mechanism, and improving the safety and reliability of the system; the cloud server locally updates the access strategy and the ciphertext according to the minimum change increment set uploaded by the data owner; a white-box tracking technology is adopted to bind a user identity with an attribute key, so that a system can locate a specific identity of a malicious user and a specific identity of a malicious attribute authority through a leaked key; intensive computing tasks in the encryption and decryption processes are outsourced to a cloud server, and meanwhile it is ensured that user privacy is not leaked in the outsourcing process; for the untrusted problem of the cloud server, the calculation result is verified by using the smart contract in the block chain, and it is ensured that the verification process is open and transparent and the result is trusted; and an attribute version mark is introduced, so that efficient revocation of old version attributes is realized, and the safety is further improved.

The invention discloses a new energy grid-connected inverter parameter identification method fusing time-frequency domain dynamic characteristics, and the identification process is divided into three stages: a time domain identification stage, a frequency domain identification stage and a time-frequency domain characteristic verification stage. According to the method, a new energy grid-connected inverter impedance characteristic curve database is established by combining time domain and frequency domain operation characteristic curves of a new energy grid-connected inverter'white box model 'of a known control strategy, and the operation characteristic curve in the impedance characteristic curve database is compared with the operation characteristic curve of a new energy grid-connected inverter'black box model'. And the control mode conforming to the operation characteristics of the new energy grid-connected inverter is identified and obtained. Compared with an existing new energy inverter parameter identification method, the new energy inverter parameter identification method has the advantages that the modeling process of the new energy inverter is simplified, and the new energy inverter parameter identification efficiency is greatly improved; and the input of a new control strategy is supported, and a guarantee is provided for the identification range and precision of the impedance characteristic curve database.

The invention discloses an adversarial sample generation method and system based on noise erasure and logit loss optimization. The method comprises the steps of 1, performing data enhancement on an original picture in a given data set; 2, inputting the enhanced picture into a preset target detection model to obtain an erasure gradient value; and 3, updating the original picture in the given data set according to the erasure gradient value and the momentum, and repeating the steps for multiple iterations to obtain a final adversarial sample. According to the method, the higher attack success rate is achieved under the white box condition and the black box condition, and the significant progress is achieved in the aspect of improving the efficiency and robustness of adversarial sample generation.

The invention discloses a target detection confrontation sample generation method based on a space-frequency domain information mask. According to the method, on the basis of space loss and frequency domain loss of a backbone network or an RPN confidence coefficient branch, the intermediate feature map gradient of the backbone network / RPN is directly utilized to guide generation of countermeasure disturbance, the technical limitation that an existing method depends on back propagation of a detection head is broken through, and effective attacks under the condition that the detection head cannot be accessed are achieved. In order to generate a transferable adversarial sample when a detection head of the dual-stage target detection model is inaccessible, the invention designs a space-frequency domain information mask method, and by randomly masking partial information in a space domain and a frequency domain, adversarial disturbance is prevented from over-fitting the white-box dual-stage target detection model, so that the transferable adversarial sample is generated. And the mobility of the target detection confrontation sample is further enhanced, and corresponding loss functions are respectively designed for the RPN feature layer and the backbone network feature layer.

The invention discloses a confrontation sample defense method driven by a structural pixel retention strategy, and relates to the field of deep learning security. According to the method, a structural pixel retention strategy is provided, and by combining dual mechanisms of pixel-level input transformation and spatial feature transformation, the recognition performance of an original clean sample is not obviously influenced while the adversarial disturbance is effectively inhibited. According to the method, a fast gradient symbol method (FGSM), a projection gradient descent method (PGD) and Carliniamp can be effectively resisted on a CIFAR-10 data set; wagner (Camp; w) and the like, and the average recognition accuracy on a clean sample is hardly influenced. According to the method, the black box attack and the white box attack can also be resisted on the CIFAR-100 data set and the Mini-Image Net data set; and an adversarial sample subjected to white-box attack can be perceived by human eyes. The method has the advantages of being high in calculation efficiency, easy to integrate to an existing deep learning model and the like, and the adversarial robustness is improved.

The invention discloses a black box directional attack countermeasure method based on a general interaction mode, and belongs to the field of attack countermeasure. According to the method, the PGD-based gradient method is adopted to update the adversarial sample in the directional adversarial attack under the condition of the white box irrelevant to the prompt, so that the adversarial sample meets the l-infinity norm limitation, and the directional adversarial attack PATA under the condition of the white box irrelevant to the prompt is realized. According to the method, a PATA algorithm is constructed to aim at an attacked model in a general interaction mode, the attacked model generates a target mask to achieve interference, transferability of an adversarial sample among different models is achieved by adding a new regularization loss function, and defense and robustness of the model can be pointedly improved and enhanced. According to the method, a single competitive sample is used, but the competitive sample is changed in each iteration, so that the calculation overhead of the attack method is reduced, the black box directional attack countermeasure method with high calculation efficiency can be realized, and the dependence on the sample is lower.

The invention, which belongs to the technical field of product appearance inspection, discloses an all-dimensional inspection system for the appearance of an automobile pipeline finished product, and the system comprises four core modules, namely a full-view image acquisition work station, an image processing and interpretation engine, a result feedback and execution mechanism and a data management and tracing interface. Wherein the image acquisition work station realizes multi-angle dead-corner-free image acquisition of a pipeline through cooperative work of a multi-axis positioning clamp and a fixed camera array; the image processing engine adopts a white-box algorithm library based on physical imaging characteristics and combines a graphical configuration interface to realize efficient and reliable defect detection; and the data management interface is used for binding and tracing the whole process of the inspection data, and reversely guiding the quality information to the previous process through the closed-loop feedback module. According to the invention, manual inspection is converted into automatic and intelligent inspection, the detection efficiency and accuracy are obviously improved, the labor cost is reduced, and reliable quality guarantee is provided for automobile pipeline manufacturing.

The invention relates to the technical field of data processing, and provides a business processing method, device and equipment under a new and old system parallel architecture, and the method comprises the steps: receiving a processing request; shunting the processing request to a target system for processing according to shunting configuration; the target system comprises at least one of a new system and an old system; when the new system fails to process the processing request, identifying the failure category of the processing failure according to a white box model; when the white box model fails to identify the failure category, identifying the failure category of the processing failure according to a black box model; and re-distributing the processing request according to the identified failure category decision. According to the embodiment of the invention, the processing burden and invalid processing overhead of the old system can be reduced under the parallel architecture of the new and old systems.

The application belongs to the technical field of hydraulic systems, and discloses a helicopter hydraulic system joint simulation method based on multi-discipline, wherein step one is to acquire hydraulic system demand; step two is to analyze the hydraulic system demand to obtain a hydraulic system use case diagram; step three is to perform function analysis to obtain an activity diagram, a sequence diagram and a black box state diagram; step four is to perform system design synthesis based on the function analysis to obtain a white box state diagram and build a hydraulic system function logic model based on Rhapsody; step five is to build a hydraulic system physical performance model based on AMESim based on the hydraulic system physical function model based on Rhapsody; the cabin display panel of the hydraulic system is simulated in AMESim based on the sequence diagram; step six is to build a main tail rotor dynamics three-dimensional visual model based on CAT IA; step seven is to combine the hydraulic system function logic model, the complete hydraulic system physical prototype model and the main tail rotor dynamics three-dimensional visual model; and step eight is to set simulation conditions and perform joint simulation.