55 results about "Initial Seed" patented technology

The invention discloses a welding seam track extraction method and system based on RANSAC parameter fitting, and the method comprises the steps: S1, obtaining three-dimensional point cloud data, and obtaining the normal vector and local curvature of each point cloud data point; selecting an initial seed point from the three-dimensional point cloud data according to the local curvature of each point cloud data point; s2, establishing a feature similarity evaluation system, and adding the initial seed points and the corresponding similar adjacent points into the same plane; according to different planes where the point cloud data points are located, plane area coarse segmentation is carried out; and S3, a plurality of geometric fitting models are constructed for the welding track based on the RANSAC algorithm, the geometric fitting model with the highest matching degree is selected from the geometric fitting models, and the welding track in the welding area is extracted. According to the technical scheme, the welding seam track can be rapidly and effectively extracted, the universality of welding seam track recognition is improved, and the extraction error of the welding seam track is remarkably reduced.

A combined hyperparameter and proxy model tuning method is described. The method involves multiple search iterations. In each search iteration, candidate hyperparameters are considered. An initial (‘seed’) hyperparameter is determined, and used to train one or more first proxy models on a target dataset. From the first proxy model(s), one or more first synthetic datasets are sampled. A first evaluation model is fitted to each first synthetic dataset, for each candidate hyperparameter, enabling each candidate hyperparameter to be scored. Based on the respective scores assigned to the candidate hyperparameters, a candidate hyperparameter is selected and used to train one or more second proxy models on the target dataset

The invention relates to the technical field of data processing, and discloses a system kernel fuzzy test seed generation method and system based on a large language model. The method comprises the following steps: performing feature extraction on Linux kernel system calling on a Host host machine to obtain a feature data set; performing CFG analysis on the feature data set to obtain a system call sequence diagram and extracting a state transition table; constructing a test scene for the state transition table, and performing Syzkaller verification to obtain an effective scene library; generating an initial seed case set for the effective scene library, and executing a test in the QEMU to obtain result data; and performing coverage analysis on the result data to obtain a coverage rate report, and optimizing to generate a target test case set. According to the invention, the test coverage rate and the defect discovery capability are improved.

The invention discloses a business space layout method based on a centroid Voronoi diagram. The business space layout method comprises the following steps: step 1, determining boundaries of business space layout and dividing the boundaries into a plurality of levels; 2, randomly generating an initial seed point of a first level at the boundary of the commercial space, and optimizing the initial seed point; taking the boundary of the commercial space and the optimized seed point of the first hierarchy as the input of a centroid Voronoi diagram, dividing to obtain a plurality of regions of the first hierarchy, and obtaining a passable path in the first hierarchy and a refined and adjusted region; and step 3, for the Nth hierarchy, generating an initial seed according to the boundary of each region in the (N-1) th hierarchy, performing optimization, performing division based on the rectangular centroid Voronoi diagram to obtain a plurality of regions of the Nth hierarchy, and obtaining a passable path in the Nth hierarchy and a refined and adjusted region. The method is simple, easy to deploy in a computer, high in automation degree and high in controllability and flexibility.

The invention relates to the technical field of security calculation and privacy protection, in particular to a large model privacy reasoning method and device based on trusted hardware and electronic device.The method comprises the steps that before a large model reasoning task is executed, a client and a server initialize respective pseudo-random number generators based on information interaction so as to synchronize initial seeds, and the pseudo-random number generators are initialized; generating a random number based on the seed; in the reasoning execution process, homomorphic encryption is used for completing word embedding of large model reasoning, a reasoning instruction set is used for converting all operators into instruction combinations, privacy reasoning is carried out based on a privacy reasoning protocol and a random number generated by a pseudo-random number, and the random number is used for masking privacy data; and after the task is completed, obtaining an execution result and proof information provided by the server, performing verification, and refusing to receive the result if verification fails. Therefore, the problems of data leakage risk, difficulty in considering safety and efficiency and the like existing in related technologies are solved.

The invention discloses a weighted Voronoi graph vector generation method and device, equipment and a medium, and relates to the technical field of graph processing. The method comprises the steps of generating a first initial isoline set, a first initial to-be-divided region, a first target isoline set and a maximum weight point set based on a first initial seed point set; according to the maximum weight point set and the first initial to-be-divided region, performing weighted Voronoi division to obtain a maximum weight point region; when the current maximum weight point set does not include all points in the first initial seed point set, performing weighted Voronoi division according to the updated second initial isoline set, the second initial region to be divided and the second initial seed point set to obtain a new maximum weight point region; and outputting a vector diagram of the current maximum weight point area until the current maximum weight point set comprises all points in the first initial seed point set. And good vector generation efficiency and vector generation precision of the weighted Voronoi graph can be ensured at the same time.

The invention provides a large model assisted Fuzz test Harness generation method and system, and the method comprises the steps: migrating a library of a QNX platform to a Linux environment for operation, and carrying out the glue code packaging of interface difference and system call difference through the generality that the library of the QNX platform and the Linux environment share a POSIX standard, thereby achieving the automatic adaptation of key API and context constraint; then analyzing binary semantics by means of an intelligent Agent constructed by a large model, automatically generating a Harness skeleton, focusing potential dangerous points (such as memory management, character string processing functions and the like), generating an initial seed in combination with a static or dynamic analysis result, and finally executing a Fuzz test in a controllable execution environment. And high coverage and rapid vulnerability positioning of the multi-dimensional attack vector of the target library are realized. The technical problems that Harness writing highly depends on expert experience, a customized generation strategy for target binary semantics is lacked, and comprehensive understanding on the binary semantics and context constraints is lacked are solved.

The invention discloses a method and a system for detecting the inconsistency between Solidence compilers, and solves the technical problem that the existing method for detecting the inconsistency between the Solidence compilers lacks comprehensiveness in detection of the defect of the inconsistency between the Solidence compilers and other types of cross-compiler inconsistency between the Solidence compilers. The method comprises the following steps: acquiring initial seed programs, complete source codes, official difference documents and complete source codes of corresponding actuators of a plurality of Solidiity compilers; generating explicit and implicit code block data of each compiler and a target variation rule matched with the initial seed program through the large model, and repairing through the reference compiler and the large model to obtain an initial test program variant; then finely adjusting the large model, generating a target test program variant, identifying, compiling and executing an inconsistent result; and finally, in combination with the official difference document, generating an inconsistency detection report between Solidity compilers.

The invention relates to the technical field of tool path planning, and discloses a combined curved surface constant residual height fixed-axis machining tool path generation method and a storage medium, and the method comprises the steps: firstly, discretizing a combined curved surface into a triangular mesh, and calculating the height of a collision-free tool location point above a machining part through adaptive quadtree sampling; forming a net-shaped structure of a discrete three-dimensional configuration space; projecting the processing area to the space for cutting, and determining an effective processing area; taking the boundary of the area as an initial seed polygon, outwards expanding line segments of the boundary into a cylindrical surface in a limited range, intersecting the cylindrical surface with the configuration space to generate a next circle of equidistant cutter location points, connecting the equidistant cutter location points to form a new cutter path, and iterating the process until the processing area is filled. According to the method, tool path selfing is fundamentally eradicated, the method is completely independent of the normal direction of the curved surface, the method is not sensitive to the model quality, the interference-free and equal-residue-height high-quality tool path can be generated in a full-automatic and high-robustness mode, and the reliability, efficiency and quality of complex curved surface numerical control programming are greatly improved.

The embodiment of the invention provides a fuzz test seed sharing method and device and a medium, and belongs to the technical field of data processing, and the method specifically comprises the steps that a template meeting the specific format requirement of a fuzzer is written, and part of test cases are selected from existing test cases; operating the fuzzifier instance; step 3, generating a test case according to the template, adding the initial seeds into a seed queue, and selecting one seed from the initial seeds for variation to obtain the test case; 4, inputting the test case into the target program; step 5, if the current test case covers a new path, a new branch or triggers a new crash, adding the current test case to the shared seed pool; step 6, the generative fuzzer instance generates a template according to the obtained test case, and the variant fuzzer instance obtains the test case from the shared seed pool as a new seed; step 7, circulating the step 3 to the step 6; and evaluating the test effect. Through the scheme of the invention, the test efficiency and the program coverage rate are improved.

The present disclosure provides computer-implemented method for training a model using a central node and a plurality of client nodes, comprising, at one or more of the plurality of client nodes: receiving a set of model parameters and an initial seed from the central node; determining an update direction based on the initial seed; computing scalar reflecting a sign and a size of an approximation of a projection of a gradient of a loss function on the update direction, using the set of model parameters; sending the scalar of the approximated gradient projection to the central node; receiving a model update scalar from the central node; and updating the set of model parameters in the update direction based on the model update scalar.

The invention discloses an LLM-based fuzzy test initial seed generation method and system, a computer and a storage medium. Lightweight static analysis is performed on target software source codes, an expansion function call graph is constructed, two key features of hard coding constraint and security sensitive operation are extracted, and explicit tags are injected in situ into the source codes; a semantic aggregation strategy based on a recursive sub-function abstract is adopted, the sub-function source code injected with the label is converted into a semantic abstract from bottom to top through a large language model, and a global semantic abstract retaining key constraint information is generated through hierarchical cue word construction; constructing a seed generation task containing a thinking chain guide instruction, forcing the model to perform reasoning according to logic steps of constraint solution, abnormal construction and enhanced generation, and generating a targeted initial input sample (namely, an initial seed) for fuzzy testing; and performing static validity check on the grammar by utilizing a general format analyzer, and performing fault-tolerant processing by combining a multi-stage backspacing mechanism. Through the recursive abstract aggregation and thinking chain reasoning technology, deep semantic understanding of complex code logic and automatic generation of high-coverage-rate test cases are achieved.

The present invention discloses an X.509 certificate verification differential testing method and system based on a differential information transfer graph. Construction stage of the differential information transfer graph: Collect real DER format certificate chains captured from the network as initial seeds and store them in a corpus, and construct a differential information transfer graph based on this. Certificate generation stage: Traverse the differential information transfer graph, select differential information vertices and seed certificate chains according to the potential value, and generate test cases through fuzzification. Certificate verification stage: Send the test cases to all SSL / TLS protocol implementation programs to be tested, record the discovered verification differences, and update the corpus and the differential information transfer graph. The present invention makes full use of the differential information generated when different implementations verify the same certificate, and effectively solves the problems of insufficient diversity of test cases and insufficient effectiveness of certificate generation strategies existing in the existing work.

The present invention proposes a method and system for automated detection of program vulnerabilities under hardware TEE. The method places a target program, a fuzz testing engine, and a corpus in a TEE environment, preprocesses the target program, and obtains a number of initial seeds from the corpus. Through shared memory, the fuzz testing engine parent worker thread creates a number of corresponding execution processes, uses the execution processes to execute the preprocessed target program in parallel, uses the initial seeds as the initial input of the execution processes, and performs parallel fuzz testing in an iterative manner. In each iteration, the mutation strategy of the seed is dynamically adjusted until the termination condition is reached or an interrupt signal is received, so as to complete the automated detection of program vulnerabilities and output a fuzz testing report. The present invention gives full play to the advantages of fuzz detection, realizes a feedback-guided fuzz testing method in a real hardware trusted execution environment, automatically detects and analyzes application security vulnerabilities, and enhances the security and credibility of TEE.

The invention discloses a method for dispersedly generating private keys in an asymmetric algorithm, and relates to the technical field of cryptography and information security. The method comprises the following steps: acquiring entropy values of various system indexes, and extracting and expanding to obtain an initial seed; generating a complete private key based on the seed, and splitting the complete private key into a plurality of sub-private key segments by adopting threshold secret sharing; after integrity verification, the fragment registers a position and a reconstruction path by means of a virtual mapping structure, is distributed to distributed computing nodes through an encryption channel, and is confirmed and stored by a signature receipt. When a signature request arrives, a path combination meeting a threshold is selected according to a mapping structure to reconstruct a temporary private key, and verification and unrecoverable destruction are carried out in a trusted environment after completion; and if detecting that the node is abnormal, isolating and starting a standby path to replace the fragment. According to the method, integrated management of private key generation, scattered storage, dynamic reconstruction and safe destruction is realized, and the security and the anti-attack capability of an asymmetric algorithm are remarkably improved.

Aspects of the disclosure relate to an algorithmic- or machine learning-based iterative seed product recommendation system. A seed recommendation platform may analyze seed genetics data, environmental data, seed similarity parameters, fitness parameters, etc., to generate an initial seed product recommendation. Based on user feedback, the seed recommendation platform may iteratively update various modules used for generating seed product recommendation and provide revised seed product recommendations.

The application discloses a mirror structure topology optimization method and device based on thermal coupling and a mirror, and relates to the technical field of metal mirror research and development. The application provides a mirror structure optimization method driven by thermal coupling simulation, which can self-adaptively control seed density and complete constrained Voronoi partition in a mirror body design domain. The method comprises the following steps: creating an initial model of a mirror to be optimized by using a three-dimensional modeling software; assigning initial Voronoi seed points; matching a corresponding stress value closest to each initial Voronoi seed point; dividing the Voronoi seed data into a high-stress-area seed point set and a low-stress-area seed point set; merging after increasing the seed point density in the high-stress-area and decreasing the seed point density in the low-stress-area; and performing 3D Voronoi partition between a front mirror surface of the mirror and a support frame body. The optimization method is suitable for optimizing the structural parameters of an existing mirror.

The invention discloses a fuzzy testing and model detection combined hardware verification method and system, and belongs to the technical field of hardware testing. The method comprises the steps that multiple rounds of fuzzy testing guided by the coverage rate are executed on a processor based on an initial seed set, snapshots generated in the fuzzy testing process are recorded, value scoring is conducted on the snapshots to construct a snapshot pool, and the processor is designed and generated based on a processor to be verified; selecting an optimal snapshot from the snapshot pool, and initializing a processor design based on the optimal snapshot; under the condition that points which are not covered exist in the processor design, a new test seed is generated through bounding model detection; performing fuzzy testing on the processor design based on the new test seed to obtain a verification result of the optimal snapshot; and synthesizing verification results of a plurality of optimal snapshots in the optimal snapshot set to obtain a hardware verification result of the to-be-verified processor design. According to the method, the efficiency and the applicability of the BMC can be remarkably improved on the premise of not sacrificing the verification precision.

The invention belongs to the technical field of oil field data processing, and particularly relates to an instruction corpus automatic generation method based on a natural language large model. According to the instruction corpus automatic generation method based on the natural language large model, the corpus generation quality is improved, the data extraction difficulty is reduced, the manual intervention cost is reduced, a complete automatic process is achieved, good field adaptability and expansion generalization ability are achieved, and the method is suitable for popularization and application. Powerful support is provided for industrial application in the field of petrochemical engineering. The invention provides an instruction corpus automatic generation method based on a natural language large model. The instruction corpus automatic generation method at least comprises the following steps: step 1, carrying out preprocessing and feature extraction on an original document to obtain first edition corpus data; step 2, performing error correction on the first edition corpus data obtained in the step 1; 3, a corpus template is prepared, and initial seed tasks are generated in batches; and 4, customizing a pre-training model, and carrying out instruction assembly.

The invention discloses a fuzzy test case generation and self-repairing method and system based on logical thinking chain guidance. According to the method, interface and boundary knowledge is extracted from documents and formalized constraints, coverage and abnormal targets are tasked to generate candidate prompts according to a logical thinking chain, an optimal prompt p * is obtained through verification and scoring, and initial seeds are precipitated; constructing a training set by using a real defect sample and a structure-maintained approximate sample, and performing parameter fine adjustment on the model to enable the model to generate a required fuzzy test case under the guidance of p *; and uniformly compiling / executing use cases in a sandbox, measuring coverage and abnormity, improving failure samples according to hierarchical self-repairing and lightweight variation, and recharging the minimized high-value samples and constraints to prompt to form a self-evolution closed loop. Under the same computing power and time budget, the implementation rate and coverage increment of the use case are remarkably improved, the exception triggering diversity is enhanced, and the method has reproducible, auditing and cross-library / multi-back-end adaptation capabilities.

Embodiments of the invention disclose a laser array generation method and apparatus, and an electronic device. The method comprises the steps of obtaining an initial seed matrix; performing assignment processing on parameters of the initial seed matrix to generate a target seed matrix; and based on the manufacturing process parameters of the laser array, expanding the target seed matrix into the laser array. According to the mode, the mode of performing non-correlation optimization on the small-size seed matrix is adopted to expand and generate the large-size laser array, and through the mode, the operand can be effectively reduced, and the optimization efficiency of the laser array is improved.

The invention relates to an automatic neural architecture searching method based on a large language model, which comprises the following steps of: in an initialization stage, selecting a group of representative and diversified seed architectures to lay a foundation for subsequent operation; in the prompt evolution stage, crossing and variation are carried out on the initial seed architecture by utilizing a feed-shot prompt technology, so that a new candidate architecture is generated; in the training and evaluation stage, the generated candidate architecture is trained on a task training set, the performance of the candidate architecture is evaluated on a verification set, and an evaluation result is obtained; in the selection and prompt adjustment stage, the candidate architecture with the highest fitness is selected according to the evaluation result, and the prompt of the LLM is finely adjusted, so that the new architecture generation capability is further improved, the performance and design efficiency of the neural network architecture are improved, and the search efficiency is remarkably improved.

The invention relates to a stateful protocol grey box fuzzy test method based on heuristic variation scheduling, which comprises the following steps: S1, initializing a test environment, and establishing a high-speed communication channel based on a shared memory, S2, dividing an initial seed set into a plurality of source seed systems according to sources; s3, selecting a target test state based on an inferred protocol state machine, and selecting a message sequence capable of reaching the state from a corpus, S4, selecting an enabled mutation operator to perform havoc mutation on candidate subsequences according to a current generation offspring mutation strategy of a current source seed system; s5, the test case is sent to the server through the shared memory channel, and the execution state is monitored; s6, generating a new generation of filial generation strategy after the first generation of filial generation strategy of the current source seed system is evaluated; and S7, repeating the steps S3 to S6 until a test termination condition is met. According to the method, global exploration and local development can be effectively balanced, the diversity and pertinence of test cases are improved, and the coverage capability of protocol states and code paths and the discovery probability of deep vulnerabilities are remarkably enhanced.

This invention discloses a method and system for detecting inconsistencies between Solidity compilers, addressing the technical problem that existing Solidity compiler inconsistency detection methods lack comprehensiveness in detecting cross-compiler inconsistencies. The method includes obtaining initial seed programs, complete source code, official discrepancy documents, and corresponding executor complete source code for multiple Solidity compilers; then generating explicit and implicit code block data for each compiler and target mutation rules adapted to the initial seed programs using a large model; obtaining initial test program variants through benchmark compiler and large model repair; subsequently, fine-tuning the large model to generate target test program variants and identify inconsistent compilation and execution results; finally, combining the official discrepancy documents to generate a Solidity compiler inconsistency detection report.

The present application relates to the field of data processing technology, and discloses a system kernel fuzz test seed generation method and system based on a large language model. The method comprises: extracting features of Linux kernel system calls on a host host to obtain a feature data set; analyzing the feature data set based on CFG to obtain a system call sequence diagram and extracting a state transition table; constructing a test scenario for the state transition table and verifying it with Syzkaller to obtain a valid scenario library; generating an initial seed case set for the valid scenario library and executing the test in QEMU to obtain result data; performing coverage analysis on the result data to obtain a coverage report and optimizing the generation of a target test case set. The present application improves test coverage and defect discovery capabilities.

This invention provides a rapid analysis method for large-scale land space compliance review, comprising the following steps: receiving a compliance review task submitted by a user, the review task including the review scope geometry, target planning layer, and review rules; calculating the area of ​​the bounding rectangle (AreaEnvelope) and the area of ​​the geometric shape (Area) within the review scope geometry; if both AreaEnvelope and Area are present, it is determined to be a small-scale review, and the traditional single-step compliance analysis process is executed; otherwise, it is determined to be a large-scale review, triggering a parallel optimization process; performing balanced region division for large-scale reviews, including initial seed point generation, Voronoi diagram construction and pruning, and iterative optimization steps; performing parallel compliance analysis and generating a review report. This invention transforms a single long-duration calculation into multiple short-duration parallel calculations, reducing the review time from minutes / hours to seconds, meeting the needs of real-time review.

The application discloses a business space layout method based on a centroid Voronoi diagram, comprising the following steps: step 1, determining the boundary of the business space layout and dividing it into multiple levels; step 2, randomly generating initial seed points of the first level in the boundary of the business space and optimizing them; taking the boundary of the business space and the optimized seed points of the first level as inputs of the centroid Voronoi diagram, dividing multiple regions of the first level, and obtaining passable paths in the first level and the regions after refinement adjustment; step 3, for the Nth level, generating initial seeds from the boundary of each region in the (N-1)th level and optimizing them, dividing multiple regions of the Nth level based on the rectangular centroid Voronoi diagram, and obtaining passable paths in the Nth level and the regions after refinement adjustment. The method is simple, easy to deploy on a computer, highly automated, and has high controllability and flexibility.

The invention discloses an intelligent query method for structured data in the field of oil-gas exploration, which innovatively constructs initial seed data and a'thinking chain 'strategy expansion data set aiming at the intelligent query problem of the structured data in the field of oil-gas exploration. On the basis of a large natural language processing (NLP) model with a large amount of world knowledge, a low-rank adaptation algorithm (LoRA) is applied to carry out fine adjustment on the model, and meanwhile, by means of assistance of a vectorization knowledge base, query of data in the field of oil-gas exploration can be completed while the world knowledge is used. Compared with a traditional method, the method has remarkable advantages in the aspects of question-answering performance, accuracy and robustness, and a new and effective data question-answering scheme is provided for the field of oil-gas exploration.

The embodiment of the application provides a vulnerability detection method and a computing device. Test cases are generated according to a target optimization PASS of a compiler, which can ensure that the test cases are highly related to the characteristics of the target optimization PASS. Basic blocks of vulnerabilities are determined based on intermediate representation data obtained after the test cases are compiled, so that the position of potential vulnerabilities is located more accurately. The test cases are sliced according to the control flow characteristics of the basic blocks of vulnerabilities, and initial seeds generated based on this can have more targeted execution paths to reach the basic blocks of vulnerabilities, thereby significantly improving the directionality and efficiency of fuzz testing, increasing the vulnerability triggering probability and detection accuracy. By fusing the target distance and the optimization path coverage, high-value seeds are selected for mutation, and efficient and accurate detection of the Rust compiler optimization vulnerabilities is realized.