31 results about "Similitude" patented technology

A method of calculating and encoding a score of semantic and spatial similarity between concepts of an ontology stored in hierarchically numbered trellis form, in which the score (NSSij) is calculated and encoded (1) for each concept (Ci) with respect to a central concept (Cj), taken two-by-two, by convergence relative to their common semantic characteristics respectively by separation according to their distance. The method is applicable to the consultation of ontologies representative of knowledge in the technical or non-technical domain.

The invention discloses a program similarity detection method and device, electronic equipment and a storage medium, and the method comprises the steps: extracting a source function and correspondingfunction basic information from a preset source program, and generating a disassembly instruction text; screening a preset target program based on the function basic information and the disassemblinginstruction text to obtain a potential similarity function, and determining a target source function; splitting the target source function into a plurality of source program fragments; splitting the potential similarity function into a plurality of target program fragments; respectively calculating the similarity between any source program fragment and the target program fragment to obtain a similarity score; determining the number of similar target program fragments of which the similarity score reaches a first preset threshold; determining the potential similarity function of which the number of the similar target program fragments reaches a second preset threshold value as a similarity function; and judging whether the source program and the target program have an association relationship or not according to the number of the similarity functions. The technical problems of low analysis speed and low analysis accuracy in the prior art are solved.

Adaptive verifiable training enables the creation of machine learning models robust with respect to multiple robustness criteria. In general, such training exploits inherent inter-class similarities within input data and enforces multiple robustness criteria based on this information. In particular, the approach exploits pairwise class similarity and improves the performance of a robust model by relaxing robustness constraints for similar classes and increasing robustness constraints for dissimilar classes. Between similar classes, looser robustness criteria (i.e., smaller ϵ) are enforced so as to minimize possible overlap when estimating the robustness region during verification. Between dissimilar classes, stricter robustness regions (i.e., larger ϵ) are enforced. If pairwise class relationships are not available initially, preferably they are generated by receiving a pre-trained classifier and then applying a clustering algorithm (e.g., agglomerative clustering) to generate them. Once pre-defined or computed pairwise relationships are available, several grouping methods are provided to create classifiers for multiple robustness criteria.

The invention discloses a software testing method integrated with fuzzy clustering, and aims at applying a fuzzy clustering method to software testing, clustering variants simulating real defects based on similarity, and improving the software defect detection efficiency as long as similar defects are detected once. The method comprises the following steps: firstly, generating variants based on a weak variation test criterion, and calculating the similarity between the variants by adopting a mathematical statistics method; then constructing a similarity matrix, sorting the variants based on the number of the variants similar to the variants, and then fuzzy clustering the variants based on the similarity among the variants, so that non-central variants are distributed to multiple clusters, and the efficiency of generating test data is improved. The proposed fuzzy clustering method is helpful for reducing the cost of variation test , has huge potential for improving the effectiveness and practicability of variation tests.

The invention particularly relates to a class case pushing method and device based on case description. The class case pushing method based on case description comprises the following steps: firstly,hierarchically encoding each paragraph of a decision by means of a deep learning model, forming features according to the context encoding of the paragraph, performing binary classification modeling through a multi-layer perceptron, generating a case description judgment model through training, and extracting case description of a to-be-processed decision by using the model; secondly, constructinga decision-making case description potential semantic space; and finally, mapping a query statement input by the user to a potential semantic space, calculating the similarity between the query statement and all case description vectors, sorting according to the similarity, and recommending a decision corresponding to the TopK case description to the user. According to class case pushing method and device, the influence of other types of elements except the case description is reduced, the class case pushing effect is improved, the case description can be accurately extracted from the decision, and the method and device are suitable for popularization and application.

A method (100) for managing an installed base of multiple artificial intelligence, AI, modules (2a-2d) comprising: obtaining (110), from multiple AI modules (2a-2d) and / or application domains (la- Id), pairs of input data (21) and corresponding output data (22), and / or pairs of training input data (21a) and corresponding reference data (23a), and / or configuration parameters (22a) relating to same or sufficiently similar operating situations according to a predetermined quantitative similarity criterion (3); and aggregating (120) the data (21, 23; 2 la, 23a), to form augmented training data (4) for training AI modules (2a-2d); and / or aggregating (130) the configuration parameters (22a), to form augmented configuration parameters (5) for configuring the internal processing chain (22) of AI modules (2a-2d); and / or determining (140), based at least in part on the data (21, 23), a quantitative indicator (6a-6d) for AI module (2a-2d) performance according to a predetermined quality criterion (6).

The present disclosure proposes a library function identification and detection method and system based on a convolutional autoencoder, including: extracting the instruction sequence corresponding to the binary function from each software to be tested, and then extracting the operation code in the instruction sequence; The continuous sequence of codes is counted to construct a co-occurrence matrix for each function, which is used as input to train the convolutional autoencoder model, and the convolutional autoencoder model is used to extract bottleneck features and complete model training; use training The final model encodes the library functions, and then performs similarity analysis on the library functions, and identifies the library function with the highest similarity coefficient as the final match. It has good versatility, and can directly mark various newly introduced matching objects, and obtain good results.

The invention discloses a tool feature recognition method based on deep learning, which is used to solve the technical problem of poor practicability of the existing non-standard special tool design method. The technical solution is to start with image vision, dig out the relationship between the machining features of the workpiece and the two-dimensional engineering drawing of the tool, and realize the reuse of the tool design scheme. First, based on the convolutional deep belief network generation model, the structure information and size information of the two-dimensional engineering drawing are fused, and the feature extraction of the high-dimensional tool image is performed to obtain the low-dimensional tool feature vector; then combined with the CDBN discriminant model and transfer learning, in the reuse On the basis of tool feature extraction network parameters, the relationship between processing features and tool feature vectors is mined; finally, through the two steps of tool classification based on Sofmax and similarity matching based on the cosine method of included angle, the tool feature vector to historical tool design scheme is realized. The mapping is good in practicability.

The invention discloses a similarity-based multi-example dictionary learning and classification method, comprising: obtaining a training set; for each positive candidate example selected from a positive bag set, repeatedly performing the following steps, cyclic iteration, training a classifier, Until the preset iteration stop condition is met: calculate the similarity weights of each example in the training set for the positive class and the negative class respectively; learn each example in the training set through dictionary learning to obtain the sparse coding of each example; learn a projection dictionary, The sparse coding of each example is re-represented; based on each example re-represented by projection and the similarity weight of each example, a classifier is trained; based on the trained classifier, the category of the target package is obtained. By applying the technical solution provided by the embodiment of the present invention, the category of the target package can be accurately determined, and the classification effect of the classifier is improved. The invention also discloses a multi-instance dictionary learning and classification device based on similarity, which has corresponding technical effects.

The invention discloses a multi-mode characteristic verification universal method based on a dynamic time bending algorithm, and belongs to the field of multi-mode optimization. The method takes a dynamic time bending algorithm as a theoretical support; the method specifically comprises the following steps: for a single-target problem, encoding and modeling an actual problem, then carrying out random walk sampling, calculating a dynamic time bending distance by using a sampling curve and a standard multimode test function, analyzing the similarity between the curves, and further judging the multimode characteristic of the actual problem; for a multi-target problem, after an actual problem is abstracted into a function curve, uniformly grouping a curve decision space, adopting a multi-target optimization algorithm for optimization to obtain Pareto leading edges corresponding to grouping areas, and calculating the dynamic time bending distance between multiple groups of leading edges; and analyzing the similarity between Pareto front edges of different areas, and judging the multi-mode characteristic of an actual problem in a decision space. The universal multi-mode characteristic verification method is high in universality, clear and rigorous in verification logic and good in reasonability.

The invention relates to semantic analysis, and provides an artificial intelligence-based method, device, equipment and medium for determining test scores. The present invention can accurately determine whether the user's answer is plagiarized through the Hamming distance of the standard vector and the user vector, and respectively analyzes the standard vector, the user vector and the test question vector through the preset encoder, and can make the test question feature vector, standard feature vector and Each element in the user feature vector integrates the contextual semantic information of the corresponding vocabulary. Through the similarity weighted operation of the standard feature vector and the user feature vector, the words with greater similarity between the user answer and the standard answer can be used in subsequent tasks. The semi-open-ended question can be accurately determined by analyzing the semantic similarity between the user's answer and the standard answer, and the matching degree between the user's answer and the test questions to be scored. In addition, the present invention also relates to blockchain technology, and the target score can be stored in the blockchain.

The invention provides a program change influence analysis method based on a code text and a calling relationship, which belongs to the field of software engineering and comprises the following steps:1, acquiring a method calling relationship of each method through third-party software; step 2, analyzing method text similarity of the initial change set and other method sets from the perspective of the code text; 3, analyzing the method call correlation of the initial change set and other method sets from the perspective of the method call relationship; and 4, calculating a comprehensive method dependence degree according to the method text similarity and the method calling correlation between the initial change set obtained in the step 2 and the method text similarity and the method calling correlation between the initial change set obtained in the step 3 and each method set, thereby obtaining a predicted affected method set. Compared with an existing word differentiation technology for experiment, the method provided by the invention has certain advantages in accuracy.

The present invention proposes a software partial plagiarism detection method based on dynamic instruction dependency graph birthmark, including: 1) using dynamic instrumentation to monitor the program at the instruction level to capture the instruction track of each function; 2) recording each function’s dynamic instruction trajectory, carry out data dependency and control dependency analysis, and construct a dynamic instruction dependency graph birthmark; 3) calculate the similarity between instruction dependency graph birthmarks, and realize the measurement of similarity between functions; 4) based on a given threshold, for the plaintiff Each function in the program builds a suspicious function table; 5) Extract the static function call graph of the program, and under the guidance of the call dependency, perform accurate one-to-one pairing of suspicious functions; 6) Based on the call dependency, assemble the matching function pair to generate Plagiarism evidence map, measuring the proportion of suspected plagiarism. The invention realizes the detection of partial plagiarism by constructing function-level birthmarks; the invention proposes the concept of plagiarism evidence map for the first time, which can greatly enhance the effectiveness of evidence.

The invention discloses a method for nesting and arranging 3D models based on model similarity, which comprises the following steps: step 1: voxelize the surface of the 3D model; step 2: voxelize the 3D model according to the similarity Combination classification, divided into similar model group and non-similar model; Step 3: Nest and arrange similar model group and non-similar model according to the order of yzx axis to get the placement area; Step 4: In the placement area, similar The model group and non-similar models are placed iteratively and nested in the order of the yxz axis, and the area is updated after each iteration until the iteration is completed. Firstly, according to the voxelized 3D model, make similar classification and combination of 3D models, then nest and place them in the yzx direction to get a placement area, and finally nest and place them in the yxz direction, iteratively adjust the xy direction size so that each The area of ​​the second placement is the smallest. Through the combination of similarities, the support material between the models is reduced, and the models are easy to remove after printing.

The invention relates to a test program plagiarism detection method based on test code segment similarity. This method is aimed at each method to be tested in the program to be tested. Firstly, the unique method identifier is calculated based on the class name, method name, and parameter sequence; secondly, a set of all test code fragments is extracted from the test program. The corresponding method; then, analyze the similarity between the test fragments to obtain the similarity analysis report, and thus calculate the similarity value between the fragments; finally, use the similarity value of the test fragment to calculate the overall similarity value of the test program, and use the overall test program The similarity value can more accurately judge the plagiarism between the test programs. The purpose of the present invention is to fill in the blank of test code similarity detection technology, solve the low precision of test code similarity analysis and the low efficiency of test code plagiarism detection mainly relying on manual operation, thereby improving the efficiency and accuracy of test code similarity detection .

The invention discloses a fault prediction method based on IAALO-SVM and similarity measurement, the steps are as follows: step 1: select key fault components; step 2: extract circuit system characteristic parameters; step 3: train support vector machine regression model; step 4 : Calculate the similarity distance; Step 5: Perform lifetime fusion. The present invention combines measured data with offline data, conducts in-depth research on the time series similarity measurement method, effectively avoids the single-step loop iteration method of adaptive antlion optimization support vector machine, and can achieve more accurate predictions in a longer period of time However, it still has the disadvantage that the farther the prediction time is from the initial training sample, the larger the prediction error is, and the larger the error of the calculated remaining service life is. At the same time, according to the characteristics of the sequence data to be tested, the dynamic time warping method is selected to mine the similarity between the test object and the offline data, and the offline data is fully utilized to predict the fault of the test object.