6 results about "Confidence region" patented technology

This invention provides a method and system for large-scale model preference optimization based on reward margin constraints. It constructs a third-order margin partitioning criterion and dynamically estimates the quantile threshold of the reward margin using exponential moving averages, adaptively dividing preference pairs into uncertainty, buffer, and confidence regions. Subsequently, a differentiated mapping strategy is applied to different regions. Finally, the differentiated mapping is integrated into a truncated sigmoid function, and a TruncPO loss function is constructed to optimize model parameters for human preference alignment tasks in large language models. This invention can improve the original win rate of different models by 4% in benchmark tests such as AlpacaEval2 and Arena-Hard, while reducing KL divergence fluctuation by more than one-third, effectively balancing the order consistency and probabilistic calibration of preference optimization. It is suitable for deployment in large language model application systems requiring accurate preference alignment, such as intelligent dialogue, automatic problem solving, and intelligent education.

This invention discloses a data annotation method and system for autonomous driving, relating to the field of data annotation. First, multi-source sensor data is acquired and initially annotated using an automated model. Then, a joint optimization algorithm decomposes and reconstructs features, improves boundary annotation accuracy, and identifies low-confidence regions. Based on a deep active learning strategy, prediction entropy, Bayesian divergence, and task-level uncertainty are fused to screen high-value samples. Ground truth labels are obtained through manual verification, while low-confidence regions are optimized to generate supplementary labels. These two types of labels are used as incremental training data, and the model mapping matrix is ​​updated through topological residual projection. Finally, the model is deployed for road testing, and problematic data is collected, triggering a new annotation optimization process to form a closed-loop iteration. This invention improves annotation accuracy and efficiency, achieves efficient incremental model updates, and constructs a continuously evolving annotation closed loop, providing support for the iteration of autonomous driving models.

The application belongs to the field of computer data processing, and more particularly relates to a weakly supervised semantic segmentation method based on shape block semantic correlation degree. The method comprises the following steps: S1, inputting an original image into a classification network to obtain a class activation map; S2, obtaining graph structure data with shape blocks as nodes by dividing the original image through a shape division module; S3, performing shape block pooling on the class activation map by using the shape block division result in S2 to obtain a pooled class activation map; S4, training a semantic correlation degree network by using the confidence region in the pooled class activation map; S5, performing semantic classification on the graph nodes by using the adjacency matrix output by the semantic correlation degree network, and aggregating the nodes into pseudo labels; and S6, training a semantic segmentation network by using the pseudo labels, wherein the network receives an original image and outputs a predicted semantic segmentation result.

The application relates to the technical field of intelligent positioning, and discloses a combined positioning method and device based on 5G communication technology and RTK technology. According to the application, the initial terminal position and a two-dimensional confidence area are determined based on 5G communication technology in response to the positioning demand, and then a three-dimensional confidence area integrating elevation information is constructed, so that the positioning dimension is expanded. The application can also accurately screen satellites, effectively constrain and optimize the RTK ambiguity search space through the screened satellites, improve the calculation efficiency and calculation accuracy of subsequent execution operation based on the optimized ambiguity search space, and finally perform high-precision ambiguity resolution by using the RTK technology to obtain a high-precision target terminal position, thereby effectively improving the real-time performance, accuracy and efficiency of positioning and improving the applicability and practicability of the method to various application scenarios.

This invention provides a product quality consistency inspection method. Before data comparison, the number of samples to be tested is differentiated to avoid unreasonable confidence regions caused by excessively large sample sizes. Secondly, when using the t-test statistic, the sample variance of the sample data is not considered, resulting in a more reasonable confidence region. Furthermore, this invention selects different confidence regions based on the importance of product quality parameters, making the final comparison results more accurate. This product quality consistency inspection method can be used to compare product parameters between a new production line and the original production line after product technology transfer. Based on the comparison results, it can accurately determine whether the products produced by the new production line are qualified after technology transfer, with small judgment errors and accurate results, providing an accurate method for judging whether products are qualified before and after technology transfer.

This invention relates to the field of image recognition technology, and discloses a method and system for detecting defects in the inner wall of pipes based on image recognition. The method includes: decoupling the original image sequence of the inner wall of the pipe to obtain a structural layer image sequence and a texture layer image sequence; unifying the structural layer image sequence and the texture layer image sequence to obtain an enhanced image sequence; mapping the enhanced image sequence onto a preset two-dimensional plane to obtain a two-dimensional plane unfolded map to obtain a candidate region mask; performing deep feature fusion on multi-scale contextual information to obtain a discriminative deep feature vector; performing preliminary defect classification and confidence assessment on the discriminative deep feature vector, and optimizing low-confidence regions based on the confidence assessment results to obtain defect category labels and pixel-level semantic segmentation contours; determining the actual geometric parameters and spatial pose of the defect to generate a structured defect detection report. This invention can improve the efficiency of defect detection in the inner wall of pipes.