28 results about "Robustness testing" patented technology

The invention discloses a privacy protection and robustness test method and system for large model fine tuning, and belongs to the technical field of machine learning security. The method comprises the steps that a three-layer distributed architecture comprising an edge server, a cloud server and a plurality of edge clients is constructed, the edge clients distill local privacy data and cooperate with the edge server to train a global model, and a candidate detection sample set is formed; screening a sample set based on the potential feature deviation evaluation index, and sending the sample set to a cloud server for vulnerability detection to obtain an optimal backdoor detection candidate sample set; and multi-trigger parallel and progressive trigger sequence backdoor implantation is respectively used for scenes of single fine tuning and multiple fine tuning of the large model, an optimal backdoor detection candidate sample set is combined with a preset trigger to generate a backdoor test sample set, the backdoor test sample set is mixed with a clean data set, and then the robustness of the backdoor test sample set is tested through fine tuning of the large model. Large model fine tuning and robustness testing of privacy protection can be realized in a heterogeneous model cooperative training environment.

The invention relates to the technical field of reliability testing of semiconductor power devices, and discloses a short-circuit robustness testing method and system based on grid stress degradation of a silicon carbide device, and the method comprises the steps of pre-testing short circuit, grid stress application and post-testing short circuit are continuously executed under the same physical connection by configuring a coupling testing time sequence. The control unit cooperates with the driving unit to generate a static positive and negative bias and dynamic switch stress mode, and cooperates with the power loop auxiliary switch action to simulate different short circuit faults. A dynamic and static index correlation model is established to quantitatively evaluate robustness by collecting transient waveforms and measuring static parameters and calculating short-circuit energy integral, threshold voltage and drift distance of on resistance. According to the invention, the in-situ test of the aging and short-circuit characteristics of the device is realized, the contact resistance change and the loop parasitic parameter difference caused by the repeated disassembly and assembly of the device in the traditional discrete test are eliminated, and the change of the two short-circuit waveforms completely corresponds to the internal degradation of the device caused by the grid stress.

The invention discloses a robustness test method and device, which are used for improving the reliability of robustness test. The method comprises the steps that an adversarial sample is acquired, the adversarial sample is obtained by adding N disturbances to an original sample, N is a positive integer larger than 1, the original sample is sensing data or regulation control data, and the N disturbances comprise one key disturbance and N-1 non-key disturbances; the robustness of a processing model is tested using an adversarial sample, the processing model being used to identify the sample. According to the scheme, the robustness of the processing model is tested by constructing the adversarial sample with the multi-dimensional disturbance, the performance of the processing model in a real environment can be simulated, the actual use condition of the processing model is better met, and therefore the reliability of the robustness test can be improved.

The invention provides a VLA model safety testing method based on characteristic disturbance, and belongs to the technical field of VLA model safety testing. In order to solve the technical problems of strong perceptibility and insufficient characteristic disturbance of the existing model robustness test method in a multi-modal model, the adopted technical scheme is as follows: carrying out standardization processing on an input image to enable the input image to accord with an input format of a visual encoder; obtaining internal multi-layer value representation of the model on an undisturbed image as an attack reference target; generating confrontation disturbance by using a generative network, and enabling the final feature expression to generate obvious offset; obtaining the internal multilayer value representation of the model on the disturbance image, obtaining the internal representation of the disturbance image in the visual encoder, and evaluating the attack effect; constructing a confrontation loss function to maximize the difference between the original feature and the disturbance feature, introducing a disturbance regular term to limit the disturbance intensity, and training to obtain a test model; the method is applied to VLA model security testing.

This invention discloses a photovoltaic power prediction simulation platform based on fitting correction and digital twins, comprising modules such as a hybrid data source, scheme configuration and comparison, a digital twin simulation engine, scenario scripts, and robustness testing. The method of this invention constructs a forward simulation path through triple fitting correction (meteorological data correction, power conversion correction, and scheduling standard evaluation); and utilizes the results of the third evaluation correction to form a feedback signal, dynamically adjusting the feature weighting and power calibration parameters in the forward path to form a closed-loop simulation; the digital twin engine integrates long-cycle factors such as component aging and a sudden event injection mechanism to simulate a high-fidelity dynamic environment. This invention enables dynamic evaluation of the adaptive learning capability of prediction schemes, robustness and resilience testing under extreme conditions, and provides a fair comparison environment and refined error attribution analysis, significantly improving the efficiency and depth of photovoltaic power prediction system research and development and verification.

The invention relates to a service robustness test method and device, equipment, a storage medium and a program product. The method comprises the following steps: firstly, obtaining a directed dependency graph which is used for representing a dependency relationship among a plurality of services and an error injection weight of each service among the plurality of services, then, carrying out fault injection on the plurality of services according to the directed dependency graph, meanwhile, obtaining state monitoring data of the plurality of services, and finally, according to the state monitoring data, carrying out fault injection on the plurality of services. And determining a service robustness test result of each service in the plurality of services. By adopting the method, a set of complete service robustness test method can be realized, and meanwhile, fault injection is carried out on a plurality of services by injecting wrong weights without manual experience, so that the efficiency and the accuracy of the service robustness test are improved.

The invention discloses a robustness test method for a retrieval enhancement generative software vulnerability evaluation system, and belongs to the technical field of software engineering and information security. The method comprises the following steps: S1, for a vulnerability code to be evaluated, acquiring a related original example set by utilizing a retrieval module; s2, analyzing and generating candidate variants which keep semantic consistency based on the abstract syntax tree; s3, selecting a confrontation example set through a comprehensive scoring function and a beam search strategy; and S4, constructing a prompt context by utilizing an adversarial example to interfere with vulnerability severity evaluation of the large language model, and particularly inducing high-risk / key vulnerability degradation to be misjudged as medium-risk / low-risk. The testing method can effectively reveal potential security risks caused by a retrieval enhancement mechanism, and is used for quantitatively evaluating the robustness of a vulnerability evaluation system.

The application discloses a kind of based on the robustness test method, device, computer equipment and medium of ceph's storage system, belong to computer technology field.The robustness test method based on the ceph's storage system includes: mounting ceph storage system directory or volume on test press, calling storage test tool vdbench runs storage service test script, simulates business pressure;Set fault simulation operation related parameters in configuration file and parse configuration file;According to the analysis result of configuration file, simulate the corresponding fault test scene;Monitoring storage service test script running output log, and record current key information when detecting that exception occurs;And read current key information and analyze, and output robustness test summary result.The test method can realize the various fault scenes and associated tests of ceph storage system automatically, improve test efficiency and accuracy.

The application relates to the technical field of artificial intelligence security, and provides a method and device for testing the robustness of a calculation imaging model. The preprocessed data is segmented to obtain a target region mask, and the disturbance is limited to act on the target region. The difference between the reconstruction result of the calculation imaging model for the adversarial sample and the adversarial sample is calculated by using an MSE loss function, so that the reconstruction ability of the calculation imaging model for the adversarial sample is completely disabled, the attack target is more accurate, and the loss calculation is more suitable for the task scenario. The adversarial disturbance is optimized by using an ADAM optimizer, and the advantages of the momentum method and the adaptive learning rate are used to optimize the high-dimensional adversarial disturbance. The optimized adversarial disturbance is combined with the target region mask to generate an optimized adversarial sample, so as to calculate the robustness of the calculation imaging model. Through the processes of initializing the disturbance, constructing the adversarial sample and optimizing the disturbance, it is ensured that the generated adversarial sample is an adversarial sample with maximum attack strength, and the robustness of the model can be accurately measured.

The invention discloses a method, a system and a device for finely tuning a large language model in a closed-loop fine tuning environment. The method comprises the steps that firstly, fine tuning task information is preprocessed to obtain target data, then a closed-loop fine tuning environment is created according to a preset strategy, and noise disturbance is conducted on the target data through differential privacy in the environment to obtain disturbance data; performing preliminary fine tuning iteration based on the disturbance data, and obtaining associated information including check points, intermediate indexes and the like after a first preset condition is met; then refining and iterating by using a parameter efficient fine tuning algorithm and associated information, and inputting an adversarial sample to perform a robustness test after the number of times is reached; if the conditions are met, the refined and iterated model is a fine-tuned model; and if not, activating a rollback mechanism, recovering the model to a check point state, dynamically adjusting hyper-parameters, and refining and iterating again based on new information. According to the method, data privacy can be effectively guaranteed, it can be effectively guaranteed that the fine-tuned model has high credibility, and safety guarantee of the model fine-tuning process is achieved.

The invention relates to the technical field of artificial intelligence safety, and provides a test method and device for calculating robustness of an imaging model. The pre-processed data is segmented to obtain a target area mask, and disturbance is limited to only act on a target area. The MSE loss function is used for calculating the difference between the result of reconstructing the adversarial sample by the calculation imaging model and the adversarial sample, so that the reconstruction capability of the calculation imaging model on the adversarial sample is completely invalid, the target attack is more accurate, and the loss calculation is more fit for the task scene. The adversarial disturbance is optimized by an ADAM optimizer, and the high-dimensional adversarial disturbance is optimized by utilizing the advantages of a momentum method and an adaptive learning rate of the ADAM optimizer. And generating an optimized adversarial sample by combining the optimized adversarial disturbance with a target area mask, so as to calculate the robustness of the imaging model. Through the processes of disturbance initialization, adversarial sample construction and disturbance optimization, it is ensured that the generated adversarial sample is an adversarial sample with the maximum attack intensity, and the robustness of the model can be accurately measured.

This application provides a method and system for robustness testing based on dynamic heterogeneous networks. The method includes: acquiring target data based on a multi-level target smart grid; determining the target robustness based on the target data; and, if the target robustness is determined to be less than a preset robustness, performing a target operation to make the target robustness greater than or equal to the preset robustness. The target data includes: target connectivity, target fault tolerance, target attack resistance, and / or target mortality of the multi-level target dynamic heterogeneous network. This application can improve the accuracy of perception of the dynamic heterogeneous network state, thereby improving the accuracy of robustness testing of dynamic heterogeneous networks, and ultimately improving the robustness and survivability of dynamic heterogeneous networks.

This invention provides a robustness testing method and system for the kernel interface of an airborne real-time operating system. It employs a multi-dimensional equivalence class partitioning model for parameter classification, systematically enumerating six categories—legal values, boundary values, illegal values, null pointers, extreme values, and permission exceptions—based on two dimensions: data type and semantic role. This avoids the problem of overlooking exception categories in manual, experience-based design. Orthogonal lists are used for parameter combination, ensuring that any equivalence class combination between any two parameters is tested at least once. At the test execution level, four mechanisms are deployed: watchdog deadlock detection, memory sentinel out-of-bounds detection, kernel panic callback capture, and non-volatile progress marker restart detection, making result determination independent of the API return value itself. The structured test report generated by this invention includes coverage statistics, detailed abnormal test cases, and defect location information, serving as objective evidence of DO-178C robustness testing activities and meeting the airworthiness certification requirements for test sufficiency and traceability.

The invention discloses a credibility evaluation method and system for a text reasoning chain. The credibility evaluation method comprises the following steps: receiving a to-be-evaluated text reasoning chain; constructing a logic dependency graph based on the text reasoning chain; performing verification interaction based on the graph; an evaluation result is generated based on feedback, so that deep analysis of the internal structure of the inference chain is realized through structured logic dependence and multi-round verification interaction, and the problems of insufficient verification depth, dependence on an external knowledge base, lack of robustness test, poor interpretability and the like in the prior art are solved; the method has the advantages that the credibility evaluation which can perform depth, fine granularity, interpretability and robustness test capability on the text reasoning chain can be realized.

The present application discloses a robustness testing method and apparatus, used for improving the reliability of robustness testing. The method comprises: acquiring an adversarial sample, wherein the adversarial sample is obtained by adding N types of disturbance to an original sample, N is a positive integer greater than 1, the original sample is perception data or planning and control data, and the N types of disturbance comprise one type of key disturbance and N-1 types of non-key disturbance; and using the adversarial sample to test the robustness of a processing model, wherein the processing model is used for identifying the sample. According to the solution, the robustness of the processing model is tested by constructing the adversarial sample having multi-dimensional disturbance, so that the performance of the processing model in a real environment can be simulated, being more in line with the actual use situation of the processing model, thus improving the reliability of robustness testing.

The invention discloses an adversarial sample generation and robustness test system for safety-critical intelligent software. The system comprises a model and data set import module, a model adaptation module, an adversarial sample generation module, a test execution module and a robustness evaluation module, wherein the model and data set importing module is used for importing a tested model and data; the model adaptation module analyzes the structure and parameters of the input model; the adversarial sample generation module integrates various adversarial sample generation methods, and a user can select one or more methods according to specific requirements to generate different types of adversarial samples; and the test execution module is responsible for flow configuration, test execution and result display, and provides a visual interface and an evaluation assembly line configuration unit. According to the method, the reliability and the safety of the intelligent software in various safety-critical applications are improved, the risk caused by confronting sample attacks is reduced, and important support is provided for constructing a safer intelligent system.

The invention discloses an intelligent model robustness test method and system based on smooth noise injection, a medium and equipment. The method comprises the following steps: S1, training a classification model based on a noiseless clean data set; s2, performing category prediction on samples in the clean data set by using the trained classification model, and distinguishing simple samples from difficult samples through prediction difficulty based on sample prediction categories and real labels; s3, calculating the number of noise samples according to the number of the simple samples and a set noise rate, comparing the number of the noise samples with the number of the difficult samples, and entering S4 when a first comparison condition is met; s4, samples with the same number as the noise samples are selected from the difficult samples, sample labels of the samples are disturbed, noise samples are generated, and a noise data set is constructed according to the simple samples and the noise samples; and S5, performing an intelligent model robustness test by using the noise data set. According to the invention, the noise data set with any noise rate can be generated based on the noiseless clean data set and tested.

The invention relates to a target detection model robustness test method based on uncertainty, and belongs to the technical field of image processing. The method comprises the steps of seed variation, model prediction, uncertainty analysis, test case amplification and robustness test. Diversified data variation strategies such as affine transformation and target-level transformation of the image are introduced, and target information in the image is changed, so that diversity and aggressiveness of variation samples are enhanced. In addition, the probability of correct prediction of the target is measured through an uncertainty index measurement model. The test cases generated under the guidance of uncertainty have higher value for revealing model vulnerabilities or repairing model errors, potential defects in the aspect of robustness of the target detection model can be more efficiently mined, a large number of test cases can be automatically generated, and the model vulnerabilities can be efficiently found; and the robustness of the target detection system can be improved.

The application discloses a kind of spatiotemporal prediction model robustness test method, device, equipment and medium, by obtaining the traffic state and prediction label of the spatiotemporal traffic prediction model of the traffic network to be tested, establish the adversarial attack model of robustness test;According to the traffic state of the spatiotemporal traffic prediction model, the traffic label of the spatiotemporal traffic prediction model is estimated;According to the traffic label and the adversarial attack model, the significance of different nodes is calculated;According to the indicator function and the significance of different nodes, determine victim node;For victim node, generate local optimization after the adversarial sample of multiple-step iteration is carried out.It can produce dynamic adversarial sample to attack traffic spatiotemporal prediction model, only need to identify a small amount of victim node to test the robustness of traffic spatiotemporal prediction model, the adversarial sample generated has stronger attack, can fully test the robustness of traffic spatiotemporal prediction model.

The invention discloses a cloud switching data management method and system combined with federal learning, and relates to the technical field of data management. The method comprises the following steps: acquiring a cloud processing center and a plurality of edge nodes; obtaining a plurality of learning tasks corresponding to the plurality of edge nodes; carrying out shared feature extraction on the plurality of learning tasks, and establishing M groups of shared nodes with shared features; based on federated learning, performing task model training and adversarial sample generation according to the local data set of each edge node in the M groups of shared nodes to obtain M groups of edge lightweight models and M groups of adversarial sample data; sending the edge lightweight model to a cloud processing center, carrying out an aggregation robustness test on the edge lightweight model, carrying out aggregation updating according to a robustness test result, and generating M aggregation models; and issuing the M aggregation models to corresponding edge nodes in the M groups of shared nodes. The technical problem of low collaboration efficiency of the cloud and the edge node in the prior art is solved, and the technical effect of improving the collaboration efficiency of the federal learning task is achieved.

The application discloses an adversarial sample generation method and system combining intra-sample disturbance evolution and disturbance amplitude normalization, a device and a storage medium, and belongs to the technical field of artificial intelligence security, and comprises the following steps: 1, obtaining an input image and a corresponding label and initializing each hyperparameter and a disturbance tensor; 2, in each disturbance iteration, intra-disturbance evolution is performed based on a separated copy of the current disturbance; 3, performing uniform scaling processing on the gradient obtained by the intra-disturbance evolution; 4, updating the final disturbance tensor by using the scaled gradient; 5, repeating steps 2 to 4 until a set number of disturbance iterations is reached, and generating a final adversarial sample; by introducing intra-sample disturbance evolution and external disturbance amplitude normalization, the disturbance effect and stability of the adversarial sample are effectively improved, and the application has wide application value in safety evaluation and robustness testing of an artificial intelligence model.

The invention relates to the field of robot control and decision models, and discloses a robustness evaluation method of a visual language motion model in a changing scene. Aiming at the problems that an existing evaluation method is single in disturbance, cannot be physically realized and depends on white box information, the method converts object three-dimensional transformation, illumination variation and adversarial patches into a continuous parameter space, realizes worst scene search in a black box environment by adopting a CMA-ES algorithm, and constructs a simulation-evaluation integrated framework. And the model robustness system test is realized by unifying the task benchmark and the quantitative index. Gradient information is not needed, automatic execution can be achieved, and the result can be physically verified. The method is suitable for safety verification and robustness testing of various visual language action models, and provides technical support for reliable deployment of a robot system in the fields of manufacturing, medical treatment and service.

The application provides a communication network robustness test method based on a graph neural network and related equipment; the method comprises the following steps: inputting graph data of a communication network into a graph convolutional neural network, slicing the graph convolutional neural network, and determining a slice attribute matrix; determining the spatial range of the perturbed slice attribute matrix, classifying the perturbed nodes by using the sliced graph convolutional neural network, obtaining a perturbed classification result, and constructing a constraint condition by using the perturbed classification result and the perturbed node attribute matrix; constructing a linear boundary of each layer activation function to determine an output boundary, performing back propagation based on the output boundary of the layer, obtaining the output boundary of the sliced graph convolutional neural network, determining the maximum value of the perturbation budget according to whether the output of the node under the current perturbation budget meets the constraint condition, and judging whether the classification result of the node is reliable according to the maximum value of the perturbation budget.

The application provides a kind of robustness test and device of code agent, and it relates to artificial intelligence technical field.The method comprises: obtaining original instance from original test data set;The original instance includes code repository, problem description, test case and reference repair patch;According to the reference repair patch, determine target file from the corresponding original instance;Based on the target file, carry out repository layer disturbance to the original instance, obtain disturbance instance, to obtain robustness test data set according to the disturbance instance;Run the code agent on the robustness test data set to carry out robustness test, can comprehensively reflect the robustness of code agent when facing real code repository.

The invention relates to the technical field of vehicle testing, and provides a vehicle robustness testing method and device. The test method comprises the steps of obtaining service configuration information of a basic service provided by a bottom layer module of a vehicle controller, and determining a service object address range corresponding to the basic service according to the service configuration information; respectively generating test messages corresponding to the service object addresses in the service object address range under a preset test working condition; and sending each test message to a bottom layer module, and determining whether the bottom layer module can normally provide a basic service under a preset test working condition according to a reply message generated by the bottom layer module. According to the vehicle robustness test method, the coverage rate of the test address of the underlying module can be improved, and then the reliability of performing the robustness test on the underlying module is improved.

This invention provides an automotive software development method, system, computer, and readable storage medium. The method includes: acquiring historical automotive software application data in real time and determining the boundary conditions of the target automotive project to be developed based on the historical automotive software application data; formulating a target engineering scheme corresponding to the target automotive project based on the boundary conditions, and extracting various work indicators from the target engineering scheme, including engineering objectives, attribute objectives, and quality objectives; designing corresponding raw automotive software based on the engineering objectives, attribute objectives, and quality objectives, and performing robustness testing on the raw automotive software according to preset testing standards to obtain the corresponding target automotive software. Through the above method, corresponding automotive software can be designed according to real-time requirements, and corresponding robustness testing can be performed on the automotive software, thereby ensuring the stability of the automotive software in use.

The application discloses a text reasoning chain credibility evaluation method and system, which comprises receiving a text reasoning chain to be evaluated; constructing a logical dependency graph based on the text reasoning chain; performing verification interaction based on the graph; and generating an evaluation result based on feedback, so that deep analysis of the internal structure of the reasoning chain is realized through structured logical dependency and multi-round verification interaction, and problems such as insufficient verification depth, dependence on an external knowledge base, lack of robustness testing, poor interpretability and the like in the prior art are solved, and the credibility evaluation can be deep, fine-grained, interpretable and robust.

This invention relates to 3D point cloud intelligent perception security testing technology, specifically a method and system for adversarial robustness testing of LiDAR point cloud target detection models. The method includes the following steps: acquiring point cloud scene data to be evaluated; calling the victim point cloud target detection model to obtain a baseline detection output; establishing a constraint set consistent with the LiDAR sampling mechanism, and generating local perturbation patterns that satisfy the constraint set; optimizing the injection position of the local perturbation patterns as a continuous decision-making process, iteratively updating the injection parameters; injecting the local perturbation patterns into the point cloud scene to form adversarial test samples, obtaining the test output of the victim point cloud target detection model; calculating robustness indices and determining failure types based on the baseline detection output and test output, and generating a security assessment report. This invention does not require access to the victim model's structure, parameters, or gradient information; robustness evaluation can be completed solely based on the inference output.