14results about How to "Avoid learning" patented technology

The invention discloses a target object pose estimation method, system and device and a medium, and relates to the technical field of pose estimation, and the method comprises the steps of offline data set establishment and online real-time multi-target pose estimation. In the off-line process, continuous RGBD streams surrounding a single object are collected, the relative poses of front and back frames are calculated frame by frame, and meanwhile, the three-dimensional surface of a target object is reconstructed to obtain a key frame off-line data set; in the online process, a visible target set in a data stream is firstly detected, VOS segmentation is performed on each target to obtain real-time effective texture and point cloud, offline key frame images of to-be-detected targets are arranged to realize single matching of all feature points and find out a closest reference frame at the same time, and RANSAC is adopted to calculate relative poses of a current frame target and the reference frame; according to the method, an expensive data labeling process is avoided, rapid integrated deployment of detection of an object without a new target is realized, the matching speed is improved, and meanwhile, online real-time multi-target pose estimation can be realized.

The invention discloses a short message user abnormal behavior detection method, system and device for a short message user with an abnormal behavior and a storage medium, and the method comprises the steps: obtaining an original short message record, and constructing a complex network according to the original short message record; initial nodes of the complex network represent short message users, edges between the initial nodes represent that short message records exist between the initial nodes, a sub-network of each initial node is obtained according to the complex network, and statistical characteristics of the sub-networks are extracted; the statistical features comprise the actual node number and the actual edge number of the sub-network, calculating the deviation degree of the statistical features and the node feature model, and screening the initial nodes according to the deviation degree to obtain the short message user with the abnormal behavior. The method can reduce false detection probability, save learning time and calculation cost, and can be widely applied to the technical field of short message security detection.

The invention discloses a hyperspectral image anomaly detection method and device, an electronic terminal and a storage medium, and belongs to the technical field of hyperspectral remote sensing image processing. The hyperspectral image anomaly detection method comprises the steps that a feature map is obtained through a coding module; a self-attention mechanism module is used for inhibiting abnormal information expression of the feature map; the method comprises the following steps of: extracting a global feature through a Transform coding module; carrying out feature fusion on the feature maps containing different information; and performing image reconstruction by using a decoding module, and obtaining an anomaly detection graph by using a reconstruction error. According to the method, spatial multi-scale feature extraction is carried out, anomaly suppression is carried out through a self-attention mechanism, the contribution of abnormal pixel points to the network reconstruction capability is suppressed, the network is prevented from learning the features of the abnormal points, and a good reconstruction effect is generated for the abnormal points.

The present invention provides a batch process modeling and final product quality prediction method based on big data, which belongs to the technical field of fine chemical and pharmaceutical batch process modeling and final product quality prediction, and specifically includes using an encoder to nonlinearly map process measurement variables and control input variables to higher-dimensional state variables, using a recurrent neural network to build a state variable dynamic model; mapping the state variables back to the measurement variables through a decoder to form nonlinear mapping parameters; using batch process historical operation big data to learn nonlinear mapping parameters and dynamic models; and combining an attention mechanism to establish a batch process final product quality prediction model. The present invention avoids staged modeling and learning, and the final product quality prediction model explicitly considers the impact of historical state variables on the final product quality, overcoming the problem that the recurrent neural network predicts the final product quality performance as the input length increases.

This utility model relates to the field of desk technology, and discloses a student desk designed to prevent myopia, including a desk body; two sets of mounting bases installed on both sides of the desk body; support rods installed on the mounting bases; and a connecting rod between the two sets of support rods. The mounting bases are fixed to the side of the desk body with screws. This student desk, by using snap-fit ​​connectors to install the connecting rod between the two sets of support rods, forms a guardrail, effectively preventing students from studying in a posture of lying on the desk body or too close to it. This design guides students to maintain correct viewing distance and reading posture, reduces prolonged close-range eye use, effectively reduces the risk of myopia, improves the efficiency of the device, and meets practical application needs.

The application discloses a modal balance and semantic enhancement training method of a full-modal safety guardrail, which significantly reduces the missed detection and false rejection of cross-modal queries by improving the training data and designing the training target, and improves the safety and reliability of the full-modal guardrail. The application efficiently constructs modal de-biased data, constructs a multi-modal query quadruple based on semantic pairing of a text seed query, and enumerates combinations of modal subsets and safety semantic labels to construct risk variant query samples. The data construction method can efficiently expand the training data based on a small amount of text samples, so that the safety and unsafe states of various modal combinations are comprehensively covered, the model is prevented from learning modal shortcuts during the training process, the dependence of the guardrail on different modal safety semantics is rebalanced, fine-grained safety supervision signals are efficiently expanded, and the cross-modal safety semantic understanding of the guardrail model is improved.

This invention discloses an online prediction method for battery SOC based on an incremental correlation vector machine (RVM) strategy using a multi-kernel ensemble approach. The method includes the following steps: Step 1, data preprocessing; Step 2, training set sampling; Step 3, kernel function selection; Step 4, model training; Step 5, model validation; Step 6, adaptive kernel parameters; Step 7, RVM model ensemble; Step 8, model prediction; Step 9, incremental learning strategy; and Step 10, online incremental prediction. From a practical perspective, this invention addresses the complexities and diverse needs of various applications. Drawing on the ideas of incremental learning and ensemble learning, it generates highly differentiated RVM individual learning models containing multiple kernel functions through dual perturbation of training samples and kernel functions. Combined with a novel incremental ensemble strategy, it avoids the problem of model overlearning, improves the model's generalization ability and robustness, and expands its application scope.

The present invention provides an API misuse detection method based on historical code change information, which belongs to the field of software engineering technology. The method first pulls the historical code change information of the open source project from the code hosting tool, and extracts the repair information related to the API misuse. The AUG before and after the program repair is used to reversely extract the program mutation operator, rather than manually designing the mutation operator, so as to effectively and efficiently generate a large number of program sets containing API misuse. After the API misuse set passes the test suite as a test case, the information in the termination trace stack is analyzed. If the termination stack trace information is not caused by the target API misuse, or also contains information caused by other API misuses, it means that the original program contains API misuse. It avoids learning the correct API usage pattern from a large number of programs and improves the efficiency of API misuse detection.

The embodiment of the invention aims to provide a two-layer load balancing method for forwarding a client request to corresponding processing equipment by calculating a hash value of a source MAC (Media Access Control), and a database firewall system using the method. Firstly, a hash value is calculated for a data source MAC (Media Access Control) from a client, then a hash result is used for carrying out remainder operation on a node number, so that a current client message is shunted to a firewall node with a corresponding serial number, two-layer load balancing under database firewall cluster deployment can be realized, and the reliability, availability and high-concurrency processing capability of a system are improved; meanwhile, for a firewall deployed in a bridge mode, a source MAC of a message of a database server is modified into a bridge port MAC of the firewall on a client side, so that a switch on the client side can be prevented from learning the MAC of the database server to a fixed port, and a scheduling algorithm is prevented from losing efficacy; moreover, the client side port of the firewall filters out the message of which the source MAC is the MAC of the port at the same side of other firewalls, and the server side port of the firewall filters out the message of which the source MAC is the client side MAC, so that the network loop problem can be avoided.

The present invention discloses a reinforcement learning method for controlling the motion of a bipedal robot based on trajectory self-correction, which uses a strategy model to output decision actions and interact with the bipedal robot online to obtain real environment samples; uses an environment dynamics model to simulate the motion state of the bipedal robot, and uses the obtained real environment samples and a supervised learning method to train the environment dynamics model; then interacts with the strategy model to obtain virtual samples, aligns the real samples with the virtual samples for trajectory self-correction, uses error control to filter the virtual samples, and after the environment dynamics model is updated, uses the strategy to interact with the environment dynamics model to generate new and more accurate training virtual samples, and uses the real environment samples and the new training virtual samples to jointly train the strategy model. The present invention only requires a small amount of interaction between the robot and the environment to achieve a cumulative reward similar to that of model-free reinforcement learning, thereby improving the learning efficiency and accuracy of the bipedal robot motion strategy.

The invention belongs to the technical field of short-term load prediction, and discloses a residential community short-term load prediction model construction method and application, and the method comprises the steps: employing a data set to train a load prediction base model, and obtaining a trained short-term load prediction base model; wherein each training sample in the data set comprises load data and meteorological data of a day before a prediction day, two days before and seven days before the prediction day, and meteorological data of a prediction moment of the current day of the prediction day; and carrying out online incremental training on the trained short-term load prediction base model by utilizing real load data and meteorological data of the day before the prediction day and a meteorological data forecast value of the prediction moment of the day on the prediction day and adopting online learning, and taking the model obtained after the online incremental training is finished as a final short-term load prediction model. The method is used for predicting the target resident community load. According to the invention, the accuracy of short-term load prediction of the resident community can be improved, and the real-time performance of prediction is improved.

The purpose of the present application is to provide a data operation method and device based on a full-text search engine. The present application uses the preset operation rules of the client of the full-text search engine to add, delete, modify and query different data; the client obtains the target data to be processed, and performs parameter analysis on the target data to obtain the target operation parameters corresponding to the target data; calls the target operation rules corresponding to the target operation parameters from the preset operation rules, and performs data operations on the target data according to the target operation rules to obtain the data operation results corresponding to the target data, and sends the data operation results corresponding to the target data to the server corresponding to the full-text search engine, which not only avoids the investment of manpower in learning the DSL statement syntax of the full-text search engine, but also improves the efficiency of data operations and saves manpower costs.

The present invention relates to the field of machine learning, specifically to a data balancing detoxification method based on data decoupling of a generative adversarial network, wherein a data set is subjected to dimensionality reduction mapping and cluster analysis to obtain a number of data clusters; adversarial disturbances are added to each data cluster to generate adversarial data clusters; transfer rate calculations and poisoned sample removal are performed on each adversarial data cluster to obtain detoxified data clusters, and all detoxified data clusters are combined into a detoxified data set; data decoupling and balancing processing are performed on the detoxified data set to obtain a balanced data set; based on the balanced data set, a detoxification model is fine-tuned and trained to obtain a detoxified model. The present invention can reduce the poisoned samples of training data, avoid learning and fitting the features of backdoor triggers during model training, and improve the stability of the model's backdoor attack defense effect and the reliability of the model in practical applications.

The application discloses a time series classification method based on frequency domain feature extraction, which first acquires time series information of different dimensions in a data set and applies standardization. Secondly, the standardized time series information is put into a time block layer for decomposition, and the decomposed data is input into a self-attention layer in a Transformer to extract time domain features. Then, Fourier decomposition is performed on the time domain features, and high-frequency components and low-frequency components on the frequency domain are retained simultaneously by selecting an operator, and then inverse Fourier transformation is performed to convert the time series data to the time domain dimension. Finally, according to the time domain feature and the frequency domain feature information, the time series data is classified by a classifier. Thus, the application solves the problem of lacking learning of frequency domain information in the classification process of time series data, and improves the accuracy of time series classification.