Patents
Literature
14 results about "Misclassification error" patented technology
Efficacy Topic
Property
Owner
Technical Advancement
Application Domain
Technology Topic
Technology Field Word
Patent Country/Region
Patent Type
Patent Status
Application Year
Inventor
Code generation and repair method and device based on multi-round feedback of large language model
The invention discloses a code generation and restoration method and device based on multi-round feedback of a large language model, and the method comprises the following steps: 1, converting problem description into a structured tetrad (target, input, output and constraint) through a two-stage task demand deconstruction mechanism, and generating a pseudo-code frame according to the structured tetrad; 2, generating a plurality of candidate code schemes in parallel based on different implementation strategies; 3, selecting an optimal code implementation scheme through multi-dimensional evaluation; 4, generating an extended test case through boundary condition analysis and public test reasoning, and expanding a verification range; 5, classifying error types in combination with static analysis and dynamic execution information, and applying a special repair strategy; 6, providing detailed repair reasons and ideas by an interpretable repair mechanism; and 7, establishing a multi-round feedback repair verification iteration mechanism composed of testing, analysis, repair and verification. According to the method, the performance of an existing large language model in processing a complex programming task is effectively improved.
Owner:WUHAN UNIV +1
Method for establishing a defect detection model and electronic device
ActiveCN115730242BElectrical apparatusTesting/monitoring control systemsMisclassification errorEngineering
A method for establishing a defect detection model and an electronic device. A first classification model is established based on a training sample set, wherein the training sample set includes a plurality of training samples. The training samples are respectively input into the first classification model to obtain respective classification results of the training samples. A plurality of outlier samples with classification errors are obtained from the training samples based on the classification results. Part of the outlier samples with classification errors are deleted from the training samples, and the remaining training samples are used as an optimal sample set. A second classification model is established based on the optimal sample set, so that a defect detection is performed by the second classification model.
Owner:WISTRON CORP
Model training method for embodied intelligence perception, image classification method and device
This application relates to the field of artificial intelligence, providing a model training method, image classification method, and apparatus for embodied intelligent perception. The method includes: constructing a hierarchical input sequence containing global category labels, local category labels, and patch labels; inputting the hierarchical input sequence into a visual Transformer backbone network; combining local mask multi-head attention and Hungarian matching cross-image voting to determine foreground and background local category labels; performing region semantic self-supervised pre-training, triplet conditional feature interaction, and label-level feature enhancement on the visual Transformer backbone network to obtain an image classification model. This application can extract discriminative local region features of targets under small sample conditions and enhance semantic consistency between similar samples, completing the training of the image classification model, thereby reducing background interference and image classification error rate in complex scenes and improving classification accuracy.
Owner:FUDAN UNIVERSITY
Prototype network feature reconstruction and classification method, device and equipment
The application provides a prototype network feature reconstruction and classification method, device and equipment, comprising: obtaining N prototypes of a prototype network; determining the Euclidean distance between the feature vectors of the i-th prototype in any two prototypes is less than a preset threshold; determining the to-be-inserted feature vector in the prototype according to K feature vectors in the two prototypes; inserting the to-be-inserted feature vector into the set position of the two prototypes respectively, reconstructing the feature of the to-be-identified object through the above method, performing three spline interpolations on the two prototypes after feature reconstruction and the feature of the to-be-identified object after reconstruction, and deriving the two prototypes and the feature of the to-be-identified object after three spline interpolations, and determining the category of the to-be-identified object according to the similarity between the derivative of the feature of the to-be-identified object and the derivative of the two prototypes. The problem that the prototypes in the prototype network are too far away from the target and cause classification errors is solved, and the classification accuracy is improved.
Owner:HISENSE GRP HLDG CO LTD
Cross-domain long-tail image classification method based on self-supervised learning and self-training mechanism
The application discloses a cross-domain long-tail image classification method based on self-supervised learning and self-training mechanism, comprising the following steps: firstly, acquiring a dataset of cross-domain long-tail images, and dividing the dataset into a training set and a test set; then, training a double-branch network model by using the cross-domain long-tail image data in the training set; for the unsupervised adaptive branch, using a self-supervised learning and global distribution alignment method for representation learning; for the supervised adaptive branch, self-training and mutual information maximization are performed on the classifier; using a dynamic weighted ensemble training strategy to automatically adjust the learning weights of the two branches, so as to realize the simultaneous optimization of the learning objectives of the two branches; finally, classifying the cross-domain long-tail image data in the test set by using the trained model. The application solves the problem that the existing classification method is prone to cross-domain long-tail image classification errors due to insufficient extraction of image discriminative features, model decision boundary deviation, error accumulation in the self-training process and the like.
Owner:YUNNAN UNIV
A system and a method for analysing log data for operational and security anomalies
PCT designated stageWO2026052997A1Hardware monitoringPlatform integrity maintainanceData integrityMisclassification error
A system (100) for analyzing log data for operational and security anomalies is provided. A preparation module (120) prepares log data by determining the necessary volume, evaluating suitability, and categorizing it. A pattern discovery module (125) identifies structures in the log data using a plurality of methods. A patterned-message structure module ( 130) creates a structured format for each log message and computes hashes for data integrity. A processing module (135) stores and validates the structured logs. A message pattern curation module (140) customizes patterned messages and offers suggestions. A log message grouping module (145) organizes log messages into groups based on patterns and binding variables. An error identification module (150) detects and categorizes errors, highlighting operational and security issues. Finally, a log storage module (155) stores configurations, static log data, and variable content efficiently.
Owner:MANVI ANAND
Method and apparatus for object classification
Owner:SHENZHEN IWIN VISUAL TECH CO LTD
Accurate operation area calculation method and device based on classification-clustering bidirectional feedback
ActiveCN121659011AData processing applicationsClimate change adaptationData setMisclassification error
The invention relates to the technical field of agricultural intellectualization, and discloses a method and a device for accurately calculating an operation area based on classification-clustering bidirectional feedback, and the method comprises the steps: determining target trajectory data generated in an operation process of an agricultural machine; based on the target trajectory data, performing operation state classification processing by using a target classification model to determine a target operation data set of the agricultural machine; wherein the target classification model is determined by performing iterative training on the initial classification model by using suspicious trajectory data in the target trajectory data; and performing area calculation based on the effective land parcel units in the target operation data set to obtain a target operation area of the agricultural machine. The target classification model can continuously learn and correct classification errors by performing iterative training on the initial classification model, so that compared with a single classification model, misjudgment and missed judgment can be effectively reduced, the precision of a final operation area calculation result is remarkably improved, and high-standard application requirements can be met.
Owner:ZHEDA ZHENGCHENG TECH CO LTD
A targeted attack method for YOLOv8
This invention discloses a targeted adversarial attack method against YOLOv8, belonging to the field of adversarial attack technology. The invention includes the following steps: analyzing the architecture, loss function, and input / output data information of the YOLOv8 object detection model to clarify the output result of the original clean image after model prediction; based on the model architecture, loss function, and input / output data information, constructing three targeted attack targets: object disappearance attack, redundant object attack, and misclassification attack, and designing a corresponding attack loss function for each target; by minimizing the attack loss function, iteratively updating the input image to achieve the three targeted attacks on the YOLOv8 model, generating adversarial sample images in batches; and inputting the adversarial sample images into the YOLOv8 model to obtain the detection result image after the attack. This invention can achieve the three targeted attack results of object disappearance, redundant objects, and misclassification.
Owner:BEIJING INST OF TECH
Intelligent goods sorting and tracing method and system
The invention provides an intelligent goods sorting and tracing method and system, and the method comprises the steps: S5, carrying out goods distribution operation through employing automatic sorting equipment according to a final classification result, recording a sorting path and a timestamp through a real-time monitoring system, and obtaining a complete data chain of a sorting process; step S7, acquiring a hash value in the data chain, performing association matching with a goods classification result through a query interface, judging whether the tracing request is consistent with an original record, and obtaining a credible tracing verification result; s8, extracting abnormal data from a tracing verification result, analyzing specific links of classification errors or data tampering by adopting an anomaly detection algorithm, and determining a source position of a problem; and S9, aiming at the source position of the problem, adjusting a weight parameter of a multi-modal fusion algorithm through a feedback mechanism, obtaining an optimized goods description vector, and judging the performance improvement of the sorting system.
Owner:GUANGDONG WULIU DIGITAL TECHNOLOGY CO LTD
Attention-based few-sample segmentation methods, devices, terminals, and media
This invention discloses a few-shot segmentation method, apparatus, terminal, and medium based on an attention mechanism, comprising: inputting a support image, a mask image corresponding to the support image, and a query image to be predicted; extracting features from the query image to be predicted and features from the support image using a weight-shared encoder; inputting the features from the support image and the mask image corresponding to the support image into a prototype generation algorithm to obtain a pair of complementary prototypes; expanding the pair of complementary prototypes to the size of the query image to be predicted and concatenating them with the query image to be predicted; predicting the foreground and background of the concatenated image using an FPN-structured decoder to obtain the segmentation result; and restoring the segmentation result to the original image size using a bilinear interpolation algorithm to obtain the few-shot segmentation result. This invention constructs a pair of prototypes in a weighted complementary manner, thereby preserving as much effective information as possible during prototype generation, reducing classification errors caused by prototype bias, and improving judgment accuracy.
Owner:PENG CHENG LAB
system
Provide a system. 【Solution means】 Means for collecting log data in real time, Means for converting the log data into a unified format, Means for extracting and classifying error logs from the log data using a generation model, Means for identifying the cause based on the error log and proposing a solution, Means for detecting and warning security incidents, Means for identifying system performance failures and proposing improvement measures, Means for adding the analysis results to the knowledge base and automatically notifying the management organization, A system including means for visualizing the analysis results using a mobile information terminal and notifying in real time.
Owner:SOFTBANK GROUP CORP
Scene perception modeling and verification method and system based on parallel intelligence
PendingCN121659779ADesign optimisation/simulationKnowledge representationOriginal dataMachine learning
The invention discloses a scene perception modeling and verification method and system based on parallel intelligence. The method comprises the following steps: acquiring an original data stream of a service scene through a multi-modal data acquisition unit; structured, semi-structured and unstructured data are acquired through the multi-modal data acquisition unit, and the dependence of a traditional method on a single data type is broken through; a self-adaptive weight distribution mechanism is adopted to carry out feature importance evaluation on the fusion feature matrix, a feature weight map is generated, and the mechanism can automatically distribute reasonable weights for different features according to data features and business requirements and highlight the influence of key features on classification; semantic constraint verification is conducted on the candidate classification set through the domain knowledge graph, classification deviation is corrected, the domain knowledge graph contains rich business domain knowledge and semantic information, the reasonability of the classification result can be judged from the semantic level, classification errors caused by data noise or algorithm limitation can be found and corrected in time, and the classification accuracy is improved. And verifying that the classification result accords with business reality.
Owner:XIAN UNIV OF TECH
Intention recognition method and device, electronic equipment, storage medium and product
The invention provides an intention recognition method and device, electronic equipment, a storage medium and a product. The method comprises the following steps: acquiring basic dimension features in a dialogue text, wherein the basic dimension features are stable features which do not change at will in a graph recognition system; performing intention recognition on the basic dimension features by using a recognition model to obtain a recognition result under the current intention recognition system; in response to the change of the current intention recognition system, determining a classification error sample according to a recognition result, and generating candidate features according to an error reason of the classification error sample; target features including the basic dimension features and the candidate features are determined; and training the recognition model based on the target features, and terminating the training and obtaining an intention recognition result corresponding to the target features when an iteration termination condition is satisfied.
Owner:BEIJING ZHONGKE JINDEZHU INTELLIGENT TECH CO LTD