23results about How to "Achieve learning" patented technology

The invention provides an unstructured database federated learning collaboration method and system based on swarm intelligence, and relates to the technical field of structured data, and the method comprises the steps: training a query pattern recognition sub-model, a path planning sub-model and a resource allocation sub-model through federated learning, and constructing a swarm intelligence collaboration network. After a user query request is analyzed by the query mode recognition sub-model, candidate query paths are generated and optimized by fusing an ant colony algorithm and a particle swarm algorithm, and a query execution scheme is generated. And the resource allocation sub-model performs resource scheduling accordingly. And generating a performance optimization target for updating local model parameters of each node, and fusing the updated parameters through a security aggregation algorithm to guide the next query optimization. According to the method, the query efficiency of the unstructured database is improved, and the resource consumption is reduced.

The invention relates to the field of drug-target affinity prediction, in particular to a drug-target affinity prediction method based on a multi-scale mixed attention network. According to the method, global residue characteristics can be learned by protein pocket residues by utilizing a self-attention mechanism, so that the global characteristics of the protein are learned. In addition, a cross-modal feature fusion mechanism is utilized, protein and drug structural features are enhanced in sequence features, cross-modal feature fusion in entities is enhanced, meanwhile, a cross-entity interaction module is utilized to recognize key atoms of drug molecules or key residues of protein, and therefore the overall performance of the model is improved. Finally, multi-level protein feature extraction is further introduced to further extract protein features.

The invention relates to an IPv6 address generation method and device based on few-seed image coding, and belongs to the technical field of network security. The method comprises the following steps: preprocessing: extracting an active IPv6 address set of a few-seed area from collected active IPv6 addresses, and respectively constructing a graph structure and a coding vector for the set; extracting features; the method comprises the steps of active address generation, image generation, address generation and distributed detection, alias prefix detection and active IPv6 address detection are carried out through distributed detection, a result obtained through detection is subjected to preprocessing and feature extraction and then is sent to an image generation sub-module for fine adjustment, and after the operation is repeated for multiple times, the target of conversion rate is completed. According to the invention, a complex seed address configuration rule can be effectively learned, and the problem of low detection efficiency caused by sampling deviation is solved.

The invention provides a multi-source software-defined science and education resource recommendation method and system based on a knowledge graph, and the method comprises the steps: constructing an artificial intelligence knowledge graph based on the relation between artificial intelligence knowledge entities; obtaining science and education resource data of a science and education institution, extracting knowledge entities in the science and education resource data, and connecting the knowledge entities to the artificial intelligence knowledge graph to obtain a science and education resource knowledge graph; extracting local resource features / user features based on a graph neural network to obtain local resource features and user features of each client; constructing a local knowledge graph based on the science and education resource knowledge graph, and performing model training by taking the local resource features of each client as initial features until the model converges; and inputting the users and the resources into the graph neural network for prediction, and recommending the resources with high scores to each user. According to the scheme, the relationship between knowledge entities in the field of artificial intelligence is effectively extracted, a structured knowledge graph is formed, and the recommendation precision is improved.

The invention belongs to off-line reinforcement learning, and particularly relates to an off-line reinforcement learning method combining generalized advantage estimation and modal decomposition interaction. In order to solve the problems of insufficient track splicing capability and intra-modal interaction and inter-modal interaction of multi-modal information in a track in an off-line reinforcement learning method based on TransFormer, the track splicing capability of a model in an off-line data set is improved by adopting generalized advantage estimation, a modal decomposition interaction module utilizes an encoder-decoder structure, and a mode decomposition interaction module utilizes an encoder-decoder structure, so that the track splicing capability of the model in the off-line data set is improved. And the encoder effectively processes time sequence information in a single mode by utilizing modal internal interaction based on ConvFormer. Relative to this, the correlation between the state and the action is captured based on the interaction between the modes of the TransFormer, so that the characterization capability of the model on the state and action sequence is enhanced. The decoder further fuses multi-modal information by using convolution and cross-attention mechanisms, thereby realizing more accurate motion prediction.

The utility model discloses a touch and talk pen with an AI interaction function, and belongs to the technical field of touch and talk pens. Comprising a touch and talk pen point, a touch and talk pen holder, a display screen, a touch and talk module installed in the touch and talk pen holder, and a loudspeaker, a microphone and a control button which are installed on the side wall of the touch and talk pen holder. Wherein the touch and talk module is connected with the display screen, the touch and talk pen, the loudspeaker, the microphone and the control button, and communicates with an external AI server. The touch and talk pen provided by the utility model has multimedia interaction capability, and can realize AI autonomous teaching and learning.

The present invention discloses a quantitative data processing method based on cloud computing, which belongs to the technical field of cloud computing and quantitative data processing. By adopting a deep learning algorithm and a multi-level search algorithm to optimize the parameters of a quantitative data analysis model, the learning of historical data is realized, and rules that can process quantitative data are formed. The quantitative data are then processed according to the rules, thereby realizing automatic processing of quantitative data. This method can not only effectively improve the analysis efficiency of quantitative data, but also eliminate errors caused by human analysis.

The present invention is a method, medium and device for generating models based on AI agents and low-code, and relates to the field of data processing technology, including: building a target AI agent and configuring an output API interface format for the target AI agent, obtaining target demand data, inputting the target model data into a low-code platform, formatting the target model data based on the API interface format to generate candidate model fields, the low-code platform fills the generated candidate model fields into a graphical candidate model definition interface, and displays the target AI agent's related thinking process and attached reference documents and recommended questions, the target user determines the final model through the generated candidate model, it can be seen that the AI ​​agent is integrated with the low-code platform development, and interacts with the interface of the low-code platform and the AI ​​agent to realize the AI ​​agent's learning of the database and adaptation to the low-code platform, thereby improving the efficiency and accuracy of business model creation.

The invention is suitable for the technical field of digital medical treatment, and particularly relates to a safe interaction method and device for a federal training medical model, equipment and a medium. According to the method, a server side and a client side register in the same block chain, the server side broadcasts the weight of a current global model obtained by the last round of training to the block chain, the client side updates and trains a local model according to the weight to obtain a trained local model, the updated model weight of the trained local model is broadcasted to the block chain, and the updated model weight of the trained local model is broadcasted to the block chain. And the server performs accuracy calculation on the updated model weight and updates the current global model by using the verification data to obtain the first accuracy and the current global model of the current round of training, obtains the reward distribution amount of the client according to the first accuracy, realizes federated learning of the model, limits the client on the basis of the reward in the block chain, and improves the user experience. The invention aims to reduce the influence of a client on federal learning by means of forgery aggressive data and the like in an interaction process and improve the application accuracy of a model in medical treatment.

A multimodal knowledge graph completion method based on neighborhood semantic enhancement belongs to the field of multimodal knowledge graphs. Specifically, a neighborhood semantic enhancement module is first designed to fuse different neighborhood information into the representation of the central entity to enhance the semantic expression ability of the model. Then, the cross-modal attention mechanism is used to capture the interaction and common semantic features of information between different modalities to achieve the learning of neighborhood-guided features. Then, based on the contrastive learning method, the similarities and differences between positive and negative samples are compared to learn richer entity representations, and finally improve the multimodal knowledge graph completion results.

The application discloses a high-precision map dynamic detection and updating method adopting a general image recovery strategy, belongs to the technical field of high-precision map dynamic detection and updating methods, and through a pre-training process, a degraded image is input into a network, features are extracted and enhanced layer by layer by adopting a feature generation module and a feature selection module, image recovery is realized by combining multi-dimensional feature fusion and a cross attention mechanism, the detail restoration and global consistency of the degraded image are improved, according to a BEV representation generated by the pre-training model in step S1, in combination with the historical state of the high-precision map, a dynamic detection module is adopted to identify elements that are newly added, deleted or unchanged in the map, and a difference map is output, according to the difference map generated in step S2, a map updating module is adopted to perform geometric adjustment and semantic correction on the marked added and deleted areas, an accurate and coherent local high-precision map is generated, and through a model evaluation framework, the performance of dynamic detection and map updating is comprehensively evaluated.

The invention discloses a plug-and-play license plate image automatic correction module suitable for an open scene and a training strategy thereof. The license plate image automatic correction module comprises a license plate vertex offset regression subnet, a transformation matrix solving module, a grid generation module and an interpolation sampling module. The license plate vertex offset regression subnet is connected with a transformation matrix solving module, the transformation matrix solving module is connected with a grid generation module, and the grid generation module is connected with an interpolation sampling module; inputting a license plate image into a license plate vertex offset regression subnet and an interpolation sampling module; and the interpolation sampling module outputs the transformed license plate image to a license plate recognition network. The input license plate image can be geometrically corrected in a self-adaptive manner on the basis of the recognition accuracy; the license plate image automatic correction module has excellent performance when processing the license plate image with an improper shooting angle.

The application relates to the technical field of energy storage, and discloses a hybrid energy storage system and method supporting mode switching, which comprises a data acquisition and communication module, an operation state analysis and mode decision module, a switching trajectory planning and cooperative control module, a safety monitoring and dynamic compensation module, a panoramic instruction integration and execution module, and a strategy self-learning and optimization module; target operation mode instructions are generated by inputting comprehensive operation state characteristic data of the system into a pre-trained mode research and judgment neural network model, combining a pre-set hysteresis comparison strategy and a minimum time interval constraint of mode switching, and realizing research and judgment on multiple operation modes such as power smoothing, peak clipping and valley filling, frequency modulation auxiliary services, so that frequent mode switching caused by state fluctuations is avoided; characteristic data, control sequences and safety evaluation data of the whole switching process are recorded to form a history library, and the optimization of the switching strategy is realized.

The invention discloses a driving subject identification method and system in an auxiliary driving environment. The method comprises the following steps: acquiring driving data of a vehicle; processing the acquired driving data, and screening lane changing time, steering wheel angle and acceleration fragment data; inputting the screened fragment data into a pre-constructed Stacking integrated driving subject recognition model, and recognizing the probabilities of driving subjects corresponding to different time periods in the fragment data; the Stacking integrated driving subject identification model comprises a primary learner and a secondary classifier, the primary learner comprises a lane changing working condition driving subject identification model and a steering working condition driving subject identification model, and the secondary classifier is a variable speed working condition driving subject identification model. According to the invention, the identification accuracy of the driving subject can be improved.

The present invention discloses a method for creating a personalized speech synthesis model, a speech synthesis method and a device. The method for creating a personalized speech synthesis model comprises: selecting a speaker of the same category as the user from multiple speakers of a multi-speaker speech synthesis model; training the multi-speaker speech synthesis model according to the training data of the user and the selected speaker of the same category to obtain a personalized speech synthesis model for the user. The present invention can synthesize speech of a specific speaking style of a target speaker, thereby improving the user experience.

The invention discloses a folding table and chair mechanism suitable for a narrow cabin, the folding table and chair mechanism comprises a folding dining table assembly and a folding dining chair assembly, the folding dining table assembly comprises a first mounting base and a table plate, and one side of the first mounting base and the folding dining chair assembly are arranged side by side; two sliding rails are installed on the two sides of the front surface of the first installation base, the rear end of the table plate is rotatably connected with two sliding blocks, the two sliding blocks are installed on the two sliding rails in a vertically sliding mode, and two supporting assemblies are rotatably connected between the two sides of the middle of the lower surface of the table plate and the two sides of the first installation base. When the sliding blocks slide upwards along the sliding rails, the table plate can be driven to rotate from the horizontal direction to the vertical direction, and when the sliding blocks slide downwards along the sliding rails, the table plate can be driven to rotate from the vertical direction to the horizontal direction. The folding ladder can be completely folded, the space utilization rate is greatly increased, space clearance is achieved in a narrow space, and the folding ladder is used for other guarantee functions.

The invention relates to the technical field of wearable intelligent equipment, and further relates to a sound effect recommendation method and device, a computer readable storage medium and wearable intelligent equipment. The method comprises the following steps: determining a first vector according to acquired behavior feature data and environment feature data; determining a current scene and a first confidence coefficient according to the multi-source sensing data; determining a plurality of target sound effect templates according to the current scene, the first confidence and the scene priority matrix; and determining a target divergence according to the first vector, the current scene and the plurality of target sound effect templates, and determining a target sound effect according to the target divergence. According to the method, on the premise that scene adaptability and brand specifications are guaranteed, learning of user preferences can be achieved, so that the problem that the recommended sound effect is different from the expectation of the user is effectively avoided, meanwhile, mistaken recommendation caused by low-reliability preferences is restrained through a confidence coefficient mechanism, and sound effect selection better fitting individual habits can be provided for the user.

The invention relates to a polarization sensor modeling method considering sunlight interference, which comprises the following steps of: dividing light incident to a polarization sensor into three parts, namely linearly polarized light, vertically incident unpolarized light and sunlight, and representing sunlight interference transmission light intensity by considering the influence of sunlight interference on the polarization sensor on the basis of the Malus law; based on a long short-term memory network, polarized light intensity information, a sun vector, polarization sensor calibration parameters, a polarization sensor field angle, an inner wall reflection coefficient and a diffuse reflection coefficient are used as inputs, sunlight interference transmission light intensity and a sunlight proportion are used as outputs, and a polarization angle after sunlight interference compensation is used as a training target. Learning a sunlight interference transmission light intensity equation; influences of linearly polarized light, vertical incidence unpolarized light and sunlight on the polarization sensor are comprehensively considered, a polarization sensor model considering sunlight interference is established by combining a light intensity gain coefficient of the polarization sensor, and polarization information such as a polarization angle and a polarization degree is solved.

This invention relates to the field of defect detection technology, specifically to a method for improving the YOLOv8 network and its application in strip steel surface defect detection. Based on YOLOv8, this invention first proposes an improved coordinate attention mechanism, the three-channel coordinate attention mechanism TCCA. Addressing the overly simplistic offset mask generation method in DCNv2, which leads to insufficient deformation modeling capabilities, this invention proposes deeply embedding TCCA into DCNv2. Furthermore, this invention introduces a global attention mechanism to improve the model's ability to extract global features. The invention also introduces a BiFPN structure and dynamic serpentine convolution to enhance the model's multi-scale feature fusion capabilities. MDPioU replaces the original loss function of YOLOv8, solving the problem of loss of effectiveness due to identical aspect ratios in predicted bounding boxes, while increasing convergence speed and localization ability. Extensive comparative and ablation experiments on the NEU-DET dataset demonstrate that the improved algorithm of this invention achieves higher defect detection accuracy and faster speed.

The invention discloses a time sequence anomaly detection method, device and equipment and a storage medium, and the method comprises the steps: initializing a system, taking an original time sequence T as input, and initializing TL and Ts to be null; performing assignment through TL <-T-Ts, using a gradient descent method through an objective function # imgabs0 #, and optimizing parameters theta of an encoder and a single classification method; fixing the optimized parameter theta, starting to optimize the separated abnormal value part Ts, and reducing Ts1 through a near-end approximation sparse function; and judging whether the steps S1 to S3 are continuously and repeatedly executed or not by calculating whether circulation stop conditions, namely TL and Ts, are changed or not, and finally returning to a clean time sequence obtained by separation. The method can be applied to multiple fields, the influence of abnormal values can be effectively processed, and the time sequence abnormity can be detected more accurately.

The application discloses a three-dimensional point cloud few-shot classification method and system based on CLIP, which takes point cloud data and text description as original input, obtains multiple view images by projecting the point cloud at multiple angles, takes the multiple view images and the text description as the input of a pre-training model CLIP, obtains the corresponding text category of the multiple view images, and thus obtains the category of the original point cloud corresponding to the multiple view images. A learnable projection module is used to obtain several different optimal projection angles of the point cloud, and then a rotation matrix of the point cloud is obtained through the projection angle; secondly, a perspective projection method is used to obtain the projected multiple view two-dimensional images, and a ResNet network is used to extract the multiple view image features which can best reflect the object features, so that the classification is more simple and convenient; and the point cloud few-shot classification learning method has the advantages of simple training and strong universality.

The invention provides a power distribution network-traffic network fusion optimization scheduling method. The method comprises the following steps: establishing a scheduling model based on a profit maximization objective function and a consumer utility maximization objective function; according to a plurality of constraint conditions, optimizing the scheduling model to obtain an optimized scheduling model; carrying out data processing on the optimized scheduling model, optimizing the data processing process by adopting a federal learning mechanism, and protecting data in the scheduling model processing process; in the information interaction process, safety and privacy of information interaction are considered, a federal learning framework is utilized to guarantee data privacy, meanwhile, learning and optimization of the model are achieved, in the co-decision process between the electric vehicle and the energy consumer, negotiation or market interaction can be carried out on the basis of respective objective functions, and the decision-making efficiency of the electric vehicle and the energy consumer is improved. And meanwhile, the central transaction platform is responsible for coordinating transactions, and the scheduling model is updated through federal learning.