27results about How to "Reduce training" patented technology

The invention discloses an image generation method and system based on diffusion model adaptive reasoning. The method comprises the following steps: constructing a next noise intensity prediction module TPM; improving an original diffusion model, and adding a TPM; the improved diffusion model is trained by using a reinforcement learning method, the whole denoising process is expanded for multiple times, the next time step is predicted in each iteration, an award is calculated according to the quality of a finally generated image and the number of steps, and TPM parameters are updated by using a PPO loss function; in the actual reasoning process, noise is randomly sampled from Gaussian noise and input into the trained diffusion model, the diffusion model is used for predicting a denoising result, meanwhile, the TPM is used for outputting distribution of time steps based on next denoising, and a distribution mode is used as the next time step used in actual reasoning; and when the predicted time step is smaller than a set threshold value, a final image generation result is obtained based on model output. According to the method, the image generation process can be dynamically adjusted, and a reasoning result with both efficiency and quality is realized.

The invention discloses a real-time three-dimensional multi-machine conflict resolution method based on graph reinforcement learning, and the method comprises the steps: constructing a three-dimensional air traffic interaction scene, carrying out the modeling of the scene into a graph, and obtaining the convolution features of the graph as model input; and establishing a conflict buffer model, and designing a reward function to guide the aircraft to minimize an additional flight distance while ensuring a safety interval. A D3QN reinforcement learning algorithm is adopted for training, the optimal avoidance action of the aircraft is obtained, and real-time multi-aircraft conflict resolution in the high-density airspace is achieved.

The invention provides a data set keyword generation and screening method based on a large language model, and belongs to the field of natural language processing and artificial intelligence. The method comprises the following steps: collecting comment text data and known keywords thereof; dividing the text data into blocks; generating preliminary keywords by the large language model; performing preliminary screening on the preliminary keywords in the preliminary keyword list in combination with a TF-IDF algorithm and a TextRank algorithm; and performing semantic and emotion correlation screening on the preliminarily screened keywords through a large language model to obtain final keywords. According to the method, the keyword generation and screening technology is remarkably improved, and the method is particularly suitable for large-scale text analysis tasks needing high precision and high efficiency.

The invention provides a key information extraction and mass attitude evaluation method based on a large language model. The method comprises the following steps: collecting comment texts of the masses on a target decision, and extracting known keywords; screening a comment text related to the known keyword by utilizing a search model; further screening a comment text related to the known keyword by utilizing a semantic understanding model; rechecking the correlation between the comment texts and the known keywords by using a large language model, and filtering out the comment texts which are misjudged to be related to the known keywords to obtain a key comment set; and a multi-agent system is constructed, each agent independently evaluates the attitude of each comment text in the key comment set to the target decision, and the user agent integrates the evaluation results of all agents and outputs the evaluation results. According to the method, a multi-level screening mechanism and an attitude evaluation mechanism based on multi-agent consensus are adopted, the calculation amount in the similarity judgment process is reduced, and the attitude evaluation accuracy and the system processing efficiency are improved.

The invention relates to a method for generating a complex-shape three-dimensional building model based on frame field deep learning. The problems that in the prior art, when a three-dimensional building model is generated, training difficulty is large, the three-dimensional building model is easily affected by smooth corners, and artifacts are generated are solved. The method comprises the following steps: S1, acquiring remote sensing data and point cloud data; s2, preprocessing the acquired image data; s3, carrying out data normalization processing; s4, designing a frame field and optimizing details; s5, optimizing the activity skeleton model; and S6, generating a multi-shape contour and a three-dimensional model. The method has the advantages that the generation precision and the generation efficiency of the three-dimensional building model are improved, buildings with complex shapes can be processed, and the method has a wide application prospect.

The invention discloses a multi-scale fusion fast network traffic anomaly detection method based on comparative learning, and the method comprises the steps: carrying out the preprocessing of traffic data, extracting features, converting the traffic data into a traffic image, carrying out the normalization processing of the traffic image, and dividing the traffic image into small blocks; constructing a multi-granularity multi-layer perceptron model, inputting traffic images after normal traffic data and abnormal traffic data division into the multi-granularity multi-layer perceptron model, and extracting feature information; defining a comparative learning loss function, training the multi-granularity multi-layer perceptron model by comparing the characteristics of the normal traffic and the abnormal traffic, and iteratively adjusting the model parameters of the multi-granularity multi-layer perceptron model to reduce the comparative learning loss function value until convergence; and inputting the preprocessed traffic image into the trained multi-granularity multi-layer perceptron model, wherein the multi-granularity multi-layer perceptron model can distinguish normal traffic from abnormal traffic.

The invention belongs to the field of social risk prevention and control, and discloses a credit card risk prevention and control method and system based on Stacking ensemble learning, and the method comprises the steps: preprocessing a data set, extracting key features, and employing a resampling technology to balance positive and negative examples; constructing a shallow feature rule by means of a diversified machine learning method, and establishing a deep neural rule by using deep learning technologies such as a multi-layer perceptron; based on a Stacking ensemble learning method, a feature rule and a neural rule are fused. According to the invention, the risk can be identified and predicted more accurately, so that the accuracy and timeliness of early warning are improved; through fusion optimization of multiple rules, the problems of ambiguity, conflict, non-uniformity and the like in current social governance and risk prevention and control rules are solved.

The invention discloses a gene amplification fluorescence in situ hybridization image intelligent identification system based on deep learning. According to the technical scheme provided by the invention, in the system, an image acquisition module is used for performing image acquisition on a sample marked by a FISH technology by using a microscope to obtain an initial image; the SAM segmentation module is used for carrying out image segmentation on the initial image based on an SAM model to determine a segmented image; the ResNet comparison module is used for comparing each quasi-cell image by using the standard cell image, and topping a real cell image; the fluorescent dot counting module is used for carrying out fluorescent dot identification and counting based on a color filtering technology, and carrying out gene amplification detection based on a target gene signal and a chromosome centromere signal. According to the method, more accurate intelligent segmentation of the FISH image is realized based on the SAM model, the accuracy of cell judgment is further improved by using a deep learning technology, the identification and counting of target gene signals are quickly realized, and the gene amplification level is better judged.

The application provides a location-based search processing method, device and equipment and a storage medium. The method comprises the following steps: in response to a location search request of a user, inputting search information into a preset multi-task search model; simultaneously processing the search information based on a city jump decision task and a search suggestion task in the multi-task search model to obtain target city location features corresponding to the city jump decision task and search suggestion features corresponding to the search suggestion task; obtaining a search processing result based on the target city location features and the search suggestion features; obtaining a search suggestion list with the highest matching degree with the search processing result in a preset search suggestion library, and displaying the search suggestion list on a search interface. Through the above method, the training and development and maintenance costs of multiple models can be effectively reduced.

The application discloses a Transformer oil well production prediction method and device based on gating constraints and Kalman filtering, and a medium, which comprises the following steps: first, obtaining oil well historical production data and preprocessing; then, using a mutual information method to screen input features related to target production, and constructing conditional gating constraint features representing the opening and closing states of the oil well; then, inputting the processed time sequence samples into a Transformer prediction model to obtain initial prediction results of the oil well production; then, performing state constraint on the initial prediction results according to the conditional gating constraint features, and performing causal recursive residual correction on the prediction results of subsequent time based on the current time prediction residual by using Kalman filtering to obtain final prediction results; further, performing grid search on Kalman filtering parameters Q and R for optimization, and determining the optimal parameter range by combining a heat map. The application can effectively reduce the prediction error, and improve the accuracy, stability and robustness of the oil well production prediction.

The application discloses a high-efficiency beam training method, device, medium and product under near field communication, relates to the technical field of communication, and comprises the following steps: selecting code words from the first layer of a stage one codebook according to the index of optimal code words; finding out the optimal code words from the selected code words and updating the index; feeding back the updated index to a base station; determining whether the first cycle parameter is greater than the first maximum cycle parameter, executing the above steps when the first cycle parameter is greater than the first maximum cycle parameter, and setting the second cycle parameter otherwise; selecting code words from the first layer of a stage two codebook according to the index of the current optimal code words; finding out the optimal code words from the selected code words and updating the index; feeding back the updated index to the base station; determining whether the second cycle parameter is greater than the second maximum cycle parameter, outputting the code word at the optimal code word index after updating when the second cycle parameter is greater than the second maximum cycle parameter, and returning to execute the above steps otherwise. The application can effectively improve the reachable rate performance in near field communication and significantly reduce the training and feedback overhead.

The invention provides a welding seam segmentation method based on stripe feature enhanced extraction. The welding seam segmentation method comprises the following steps: carrying out efficient encoding and decoding learning on welding seam stripe information in an encoder and a decoder by using transfer learning and cascade information learning thoughts respectively; through the two structures, richer low-layer and high-layer semantic information can be extracted, and the extraction capability of welding seam stripes of the segmentation model is ensured; besides, the weld stripe image contains complex noise such as arc light, smoke dust and molten drop splashing, the defects of an existing jump connection mode are analyzed, a multi-layer feature fusion mechanism fusing a global attention mechanism is designed, information association in different coding layers can be learned through the model, and the fusion precision is improved. And the influence of noise can be greatly weakened, so that effective fusion of shallow-layer semantic information and high-layer semantic information is realized, and the segmentation capability of the model to weld seam stripe features is enhanced.

The invention belongs to the technical field of garbage cleaning, and particularly relates to a sweeper garbage collecting and leading-in mechanism which comprises a mounting frame, a sweeping cavity, a garbage collecting and leading-in device and a garbage collecting and leading-in device, the mounting frame is internally provided with a conveying cavity, the right side wall of the conveying cavity is provided with a discharging port, and the conveying cavity is internally provided with a conveying belt; a collecting and guiding-in assembly is arranged in the sweeping cavity and used for collecting and guiding in garbage. Garbage of different sizes can be guided into the sweeping mechanism without difference, the size and the type of the garbage do not need to be perceived and recognized through an AI technology, and efficient sweeping is achieved; in addition, the research, development and manufacturing cost of equipment is saved, and the cost of large model training, vehicle-mounted computing power and sensing parts is saved.

This application discloses a cross-domain multimodal recommendation method and system based on multi-agent attribute-driven architecture. Embodiments of this application can be applied to product recommendation scenarios in an e-commerce platform. The method includes: receiving multimodal features; processing the multimodal features using a first operation to generate reconstructed original multimodal features; the reconstructed original multimodal features include attribute-related representations and attribute-independent representations; then, processing the reconstructed original multimodal features using a second operation to generate a modal cue vector; then, processing the modal cue vector and the attribute-related representation to generate a new feature vector; and finally, processing the new feature vector using a sequence encoder to generate attribute-level multimodal sequence states. By sharing basic embeddings, the parameter overhead is significantly reduced, solving the problem that the parameter size of traditional cross-domain multimodal encoders increases linearly with the number of domains.

The present application relates to a kind of triangular mesh autoregressive generation method based on geometric curvature perception state space modeling, belong to artificial intelligence, computer vision and computer graphics field.First, the input triangular mesh is serially represented, obtains the grid mark sequence and feature for state space modeling;Second, by establishing geometric curvature perception module, the curvature information and multi-scale geometric statistical information of input triangular mesh are extracted, obtain geometric condition vector;Then, grid mark sequence and feature, geometric condition vector are modeled in state space, obtain the output feature for predicting the conditional probability distribution of next grid mark;Finally, by constructing autoregressive generation model based on geometric quality constraint and adaptive decoding, combined with conditional probability distribution generates target grid mark sequence, finally generates target triangular mesh.The present application not only can realize adaptive generation of triangular mesh, but also can consider the geometric rationality of sequence modeling efficiency and generation result.

The application discloses a kind of based on knowledge graph's cancer cell line drug response prediction large language model construction method.The method fuses the structured information of biological knowledge graph and the semantic reasoning ability of large language model, realizes the multi-level modeling and explainable prediction of drug-cell reaction mechanism.The method comprises the following steps: first, integrate multi-source data to construct multi-modal dataset;Second, establish the drug response knowledge graph containing drug, gene and other entities;Then, fine-tune the model using LoRA technology and inject graph embedding, realize cross-modal alignment;Further, dynamic knowledge retrieval and enhancement are carried out using RAG technology;Finally, output drug sensitivity prediction and natural language explanation.The application significantly improves the prediction accuracy and explainability, and can reveal the key gene pathway, providing an efficient intelligent tool for precision cancer treatment.

The present invention provides an indoor heading estimation method, device and medium, wherein the method includes grouping IMU data to obtain local features, integrating them with preset additional features, inputting the local features into a deep learning model to obtain correlation to update the additional features, and finally reducing the dimension of the integrated features to obtain the heading estimation result. The method can carefully mine the local features of the data and capture specific information in different regions; update the additional features through correlation to achieve global feature extraction and enrich feature representation; the dimensionality reduction process can reduce the computational complexity, reduce the training and inference time, remove redundant information and noise, and improve the generalization and robustness of the model. Compared with the existing technology, it can better adapt to the complexity of the indoor environment and improve the accuracy and reliability of heading estimation.

The invention belongs to the field of artificial intelligence, and particularly relates to a security defense method, device and equipment of a large language model for malicious query. The method comprises the following steps: obtaining a query question; obtaining a candidate lexical element set of a current word responding to the query problem by the large language model, and adjusting probability distribution of each lexical element in a safer direction in the candidate lexical element set; and selecting the lexical element with the highest probability from the candidate lexical element set as the current word of the response. According to the invention, the security defense method for the malicious query is efficient and does not obviously influence the performance.

The invention discloses a dynamic routing method and device based on a self-purification expert pool, and the method comprises the steps: firstly, constructing an expert feedback pool supporting dynamic extension, driving a group of heterogeneous domain expert models to carry out the reasoning of a cross-domain data set problem, and storing a result; secondly, implementing a double-layer self-purification mechanism, performing data cleaning and expert model cleaning on a feedback pool based on real answers, and automatically eliminating low-quality data and low-performance experts; further, on the basis of the purified high-quality data, a lightweight BERT model is trained to serve as a routing learning device, so that the routing learning device learns an expert which predicts the most adaptive expert according to the semantic features of the problem; and finally, in practical application, the route learner is used for performing rapid reasoning and routing on a new input question, and the new input question is distributed to the predicted optimal expert model to obtain an answer. Through dynamic self-purification of data and resources and data-driven routing decision, intelligent management and efficient utilization of heterogeneous expert model resources are realized, routing accuracy and interpretability are ensured, and meanwhile, the capability of a system for processing complex cross-domain problems in a scene with limited computing power is remarkably improved.

The invention relates to the technical field of abrasive disc processing equipment, in particular to a method for intelligently manufacturing a metal mineral polished section, which comprises the following steps: S1, roughly grinding a sample through water and sand adding, surface grinding and plane treatment operations; s2, the sample is subjected to advanced fine grinding treatment through a grinding material; s3, performing high-precision wafer grinding treatment on the sample through timed grinding with grinding powder and polished wafer thickness fine detection operation, and using a thickness sensor and a roughness sensor in combination to ensure that the surface quality of the sample meets a high standard; and S4, under the condition that the thickness reaches the standard, the sample is polished, a roughness sensor is used for detecting the surface smoothness of the sample, the polishing machine stops working after the specified smoothness is achieved, and the metal mineral polished section meeting the requirement is obtained. Detail parameters in the grinding and polishing processes are monitored and adjusted in real time through a sensor system, and high consistency and high standard quality of final products are ensured.

The invention discloses an intelligent network attack detection method based on traffic analysis and electronic equipment. The method comprises the following steps: acquiring traffic data in a network; a pre-trained automatic encoder AE is constructed, the automatic encoder AE comprises an encoder module, a potential space and a decoder module, the encoder module adopts two layers of long short-term memory networks, and a Dropout layer is inserted between the two layers of long short-term memory networks and is used for receiving input traffic data and performing data compression and feature extraction to obtain key traffic features; and the potential space adopts a bidirectional gating circulation unit to carry out key traffic feature classification, and outputs a detection result indicating whether the traffic data has an abnormal traffic type or not through a decoder module. According to the method, the important traffic characteristics in the traffic can be fully extracted, and the problems of low detection rate and high false alarm rate caused by solving the problem of data imbalance by using a sampling technology are avoided while data dimension reduction is completed.

The application relates to the technical field of artificial intelligence, in particular to an AI model full life cycle collaborative management method and system. The method is characterized in that: a plurality of source heterogeneous data are acquired, dynamic data management is performed on the plurality of source heterogeneous data, a standardized data set is formed, a metadata graph is constructed, and unified correlation management of data, models and training processes is realized. AI model construction and configuration are completed according to a preset development mode, and model training is performed based on the configuration to schedule computing resources, so that a target AI model is generated. After the model completes training, deployment and inference service publishing are performed, and the model running performance is continuously monitored, and data management updating or model retraining is automatically triggered when the performance changes, so that closed-loop collaborative management of the AI model full life cycle is realized. The application effectively guarantees the stability, accuracy and business adaptation capability of the long-term operation of the AI model, and improves the automation level and overall efficiency of the AI model full life cycle management.

The application provides a prompt learning method based on unsupervised knowledge distillation, comprising: a supervised fine-tuning stage, taking a first visual language model as a teacher model, freezing first pre-training parameters of the teacher model, and performing supervised fine-tuning on the teacher model through labeled samples to optimize first learnable prompt parameters of the teacher model; an unsupervised distillation stage, taking a second visual language model as a student model, freezing second pre-training parameters of the student model, aligning inference results of the student model and the teacher model on unlabeled samples, and migrating discriminative knowledge of the teacher model to second learnable prompt parameters of the student model. The application also provides a prompt learning device based on unsupervised knowledge distillation, a storage medium and an electronic device. Therefore, the application significantly improves the adaptation effect of the visual language model on downstream tasks, improves the generalization performance of the visual language model, and has low training and inference costs.

The present invention relates to the technical field of coal quality identification and monitoring, and discloses a method and system for dynamic detection of coal quality based on image identification of mining face, wherein the method comprises: in the process of mining operation by a coal mining machine, obtaining images of the first, second and third regions from top to bottom of the mining face; extracting bedding features from the first region image and the second region image to obtain a bedding line distribution map; calculating the maximum interlayer spacing and the ordinate variance based on the bedding line distribution map; determining the uniformity of the coal seam based on the maximum interlayer spacing; determining the flatness of the coal seam based on the ordinate variance; obtaining the quality of the coal seam based on the uniformity of the coal seam and the flatness of the coal seam; extracting geometric features from the third region image and generating a bright spot distribution map; calculating the average area of ​​the bright spots based on the bright spot distribution map; determining the degree of crushing of the coal rock according to the average area of ​​the bright spots, and then determining the quality of the coal rock. The present invention can analyze the quality of the coal seam and the quality of the coal rock.

The application provides a railway sectional insulation device arc extinction angle defect identification method, belonging to the technical field of intelligent defect identification. The method obtains a sectional insulation device image to be detected; inputs the sectional insulation device image to be detected into a pre-trained arc extinction angle defect identification model, uses a backbone network of the model to extract a local feature map and a global feature map of the sectional insulation device in parallel; uses a bridging fusion module in the backbone network to fuse the global feature into the local feature and output a first feature map, and simultaneously fuse the local feature into the global feature and output a second feature map; inputs the first feature map and the second feature map into a classifier of the arc extinction angle defect identification model for defect identification classification, and outputs an arc extinction angle defect identification result. The application improves the identification effect of the model on the arc extinction angle defect, enhances the local and global perception ability of the model, simultaneously realizes parallel calculation of local feature extraction and global feature extraction, and thus reduces the delay of training and inference.

The utility model relates to the field of steel pipe container lifting appliances, in particular to a steel pipe container lifting appliance device with a clamping function. The lifting device comprises a main frame, a lifting lug, a balancing weight and a clamping mechanism. The main frame is transversely placed, the number of the lifting lugs is two, the two lifting lugs are fixedly installed on the top of the main frame, the balancing weights are located on one sides of the lifting lugs and fixedly arranged on the edge of the top of the main frame, the number of the clamping mechanisms is two, and the two clamping mechanisms are fixedly installed on the main frame.

The invention discloses an enhanced IP geographic positioning method based on uncertainty quantization, which relates to the technical field of IP positioning, has obvious advantages in the aspects of positioning accuracy, robustness, uncertainty quantization and calculation efficiency, and particularly has the advantages of high positioning accuracy, high robustness, high uncertainty quantization and high calculation efficiency in a complex and dynamic network environment. The technical scheme of the invention provides a more excellent positioning solution compared with the prior art. By innovatively combining a deep integration model, an energy regression model and an uncertainty quantification technology, the method can provide more efficient and more robust geographic positioning services in different application scenarios.