99 results about "Multiple experts" patented technology

The invention relates to the technical field of industrial process prediction, in particular to a tundish erosion prediction method based on a hierarchical hybrid expert framework. The method comprises the following steps: collecting time sequence physical field data of the tundish, and screening features to construct a unified feature space; shunting the feature space to obtain a time sequence feature and a statistical aggregation feature; a statistical aggregation feature training classifier is utilized to generate a calibration posterior probability, and a gating network is constructed; generating an initial mode subset based on a posterior probability and training a corresponding expert model; the confidence of data to be measured is obtained by the gating network, and a single expert model is selected for prediction or multiple expert models are fused through a self-adaptive strategy for weighted prediction according to whether the confidence exceeds a threshold value or not; and finally, reconstructing the predicted value into an erosion thickness absolute value through inverse transformation. According to the method, the accuracy and adaptability of tundish erosion prediction are effectively improved.

The invention provides a laser powder bed melting method for preparing a high-toughness TC4 titanium alloy based on machine learning. The method comprises the steps that performance data of the TC4 titanium alloy prepared through different SLM process parameters are collected; constructing a hybrid expert (MoE) machine learning prediction model framework; classifying the TC4 data by adopting a K-means clustering algorithm, and taking a clustering center as an initial weight of an input layer and a first layer of the gating network; presetting a TC4 performance prediction model through the gating network and the plurality of expert networks after clustering information initialization; performing small-range iteration through large-step network search in combination with a Bayesian optimization method so as to determine an optimal model structure and complete model training; the trained performance prediction model and a genetic algorithm are used for carrying out backstepping to obtain SLM printing parameters of the high-toughness TC4 alloy; and SLM forming is carried out, and the high-toughness TC4 titanium alloy is prepared. According to the method, a calculation framework which is high in precision and suitable for exploring a high-unknown characteristic space is constructed, and far-reaching influences on titanium alloy additive manufacturing and process optimization of other alloy systems are achieved.

The invention discloses a mechanical equipment residual life prediction method based on multiple expert models, and belongs to the technical field of mechanical equipment. Comprising the steps of receiving an original mechanical vibration signal, performing preprocessing, constructing a residual life prediction model of the mechanical equipment, training the residual life prediction model of the mechanical equipment based on a preset data set, inputting a mechanical vibration signal to be detected into the trained residual life prediction model of the mechanical equipment, and performing fault category prediction of the mechanical equipment. According to the method, the accuracy, generalization and practicability of residual life prediction of mechanical equipment are remarkably improved, and through multi-scale time-frequency domain feature fusion and a double attention mechanism, the identification capability and noise immunity of the model to a complex fault mode are enhanced; the domain adaptive routing network realizes cross-device efficient knowledge migration, greatly improves the prediction performance on unseen devices, has high precision and high real-time performance, and provides reliable technical support for industrial predictive maintenance.

The invention discloses a processing method, system and device for intelligent questionnaire survey and a medium, and the method specifically comprises the steps: obtaining questionnaire data and evaluation parameter setting inputted by a user, calling a main model to carry out universal dimension inspection on a questionnaire, and obtaining a preliminary analysis result; according to a preliminary analysis result, distributing different types of problems to corresponding expert models for deep analysis; in combination with culture feature data inquired in real time, a culture suitability score is generated to supplement an analysis result of the expert model; based on the analysis results of the plurality of expert models, performing weighted fusion processing on the score of each dimension by using a weight coefficient dynamically adjusted through a reinforcement learning mechanism to obtain a comprehensive evaluation score; the comprehensive evaluation score is returned to the user interface, feedback data are collected after the user completes interaction, and the feedback data are used for iteratively updating model parameters and setting evaluation parameters. According to the invention, efficient and accurate examination of questionnaire contents is realized.

Embodiments herein describe a system including a plurality of hardware accelerators including at least one mixture-of-experts (MoE) layer having multiple experts and a plurality of network interface cards (NICs) coupled to the plurality of hardware accelerators, wherein at least one expert of the multiple experts is offloaded from the plurality of hardware accelerators to the plurality of NICs. The plurality of hardware accelerators may be graphics processing units (GPUs). In one example, a subset of the multiple experts are selectively offloaded from the plurality of GPUs to the plurality of NICs based on memory and computational capacity available on the plurality of NICs. In another example, the multiple experts are designated as either hot experts or cold experts. The cold experts are offloaded from the plurality of GPUs to the plurality of NICs and the hot experts are duplicated for each of the plurality of GPUs.

The embodiment of the invention discloses a construction method of a cross-platform high-performance tensor program generation model and a related device. The method comprises the steps of obtaining a training data set and tensor program tuning features; training a large language model according to the training data set to obtain a base model; classifying the first hardware according to the tensor program tuning features to obtain a plurality of hardware categories; a plurality of expert layers and a knowledge aggregation layer connected with the expert layers are added in the base model, a cross-platform high-performance tensor program generation model is obtained, and each expert layer corresponds to one hardware category. And the knowledge aggregation layer is used for aggregating the relationship between the first hardware of the different hardware categories learned by the plurality of expert layers and the high-performance tensor program. By adopting the cross-platform high-performance tensor program generation model constructed in the embodiment of the invention, the efficiency and the performance of generating the high-performance tensor program on a multi-hardware platform can be remarkably improved.

The invention is suitable for the technical field of network security, and provides a network attack detection method and device, computer equipment and a program product, and the method comprises the steps: obtaining to-be-detected network traffic data; inputting the network flow data into a preset gating network to generate a routing weight value used for selecting from a plurality of expert models; processing the network flow data by using one or more of the plurality of expert models to generate an expert model processing result; wherein the plurality of expert models are arranged in parallel, and at least comprise: a first expert model configured to be used for analyzing time sequence features of the network traffic data; the second expert model is configured to be used for analyzing space-time correlation characteristics of the network flow data; and based on the routing weight value, fusing the expert model processing results to generate a network attack detection result. Therefore, diversified network attacks can be identified efficiently and accurately.

The application provides a miRNA-disease relationship prediction method, system and model based on a hybrid expert model and a storage medium. The method comprises the following steps: obtaining a miRNA-disease correlation matrix of multi-omics data of a miRNA-disease to be predicted, a miRNA similarity matrix and a disease similarity matrix; constructing a miRNA-disease heterogeneous graph, a miRNA homogeneous subgraph and a disease homogeneous subgraph, and fusing them into a multi-modal biological graph network; inputting the fusion network into a pre-trained gated hybrid multi-expert network model to output a correlation probability of a miRNA-disease pair to be predicted; and the gated hybrid multi-expert network model comprises multiple expert networks, a gating network and an output layer. The application adopts a gating mechanism to adjust the weights of each expert for miRNA-disease prediction, maintains high prediction accuracy on a small-scale data set, and has the advantages of portability and light weight.

The embodiment of the invention discloses a model training method and device. According to the embodiment of the invention, after an initial model and training sample sets are obtained, input of the initial model is determined according to task related information in each training sample set, and based on a multi-task structure of the initial model, embedded dense vectors corresponding to different tasks and shared embedded dense vectors are respectively calculated by utilizing a plurality of expert networks of the initial model; respectively calculating corresponding task mixing vectors through gating networks of different tasks, and then inputting the different task mixing vectors into corresponding task towers to respectively obtain an execution time length predicted value and a bearing parameter predicted value; and calculating multi-task joint loss as a training target of the initial model based on the predicted value, the task execution duration and the task bearing parameter, and training the initial model based on the multi-task joint loss to obtain a data prediction model. Therefore, the dependency of the model on the data richness can be reduced while the accuracy of task execution time length prediction can be guaranteed.

The invention discloses an acquisition terminal test case priority ranking method based on a fuzzy analytic hierarchy process, which comprises the following steps of: applying the fuzzy analytic hierarchy process to the field of acquisition terminal test, and quantizing fuzzy evaluation opinions of experts by triangular fuzzy through constructing a'target-criterion-scheme 'hierarchical structure; fusing a multi-expert judgment matrix by adopting a weighted geometric averaging method, and optimizing a criterion weight and performing defuzzification in combination with consistency check; and comprehensive priority scores of the test cases are calculated through weighted summation, a sorting sequence from high to low is generated, and meanwhile, dynamic adjustment during demand change is supported. According to the method, the fuzzy demand quantification accuracy can be improved by 60%, the expert dispute rate is reduced to 8% or below, the regression test error detection efficiency is improved by 40%, and the method adapts to the multi-scene test demand of the acquisition terminal.

The invention belongs to the technical field of industrial facility safety assessment, and particularly discloses an overdue service pressure pipeline risk assessment method and terminal equipment, and the method comprises the steps: constructing a fault tree with the failure of a pressure pipeline as a top event; obtaining a fuzzy number corresponding to a plurality of expert ratings of each bottom event in the fault tree, and determining a relative consistency degree of the expert ratings according to the fuzzy number; clustering the fuzzy number by using a similarity clustering method, and determining an optimal relaxation factor according to a clustering result; determining the failure probability of each bottom event according to the relative consistency degree of expert rating and the optimal relaxation factor; and determining the failure probability of the top event according to the failure probability of each bottom event. According to the method, the optimal relaxation factor is determined by combining the fuzzy number of expert rating, and the relaxation factor is improved, so that the influence of subjectivity is remarkably reduced, and the reliability of risk assessment of the overdue service pressure pipeline is greatly improved.

This invention discloses a microscopic denoising method based on multi-expert judgment, relating to the field of computer image enhancement technology. The method includes: acquiring original microscopic image data; preprocessing the image data; calculating initial expert weights through an expert weight threshold discrimination network; inputting the preprocessed data into multiple expert denoising networks for denoising; calculating result weights based on the denoising results and performing weighted fusion; and generating a denoised microscopic image. By introducing a multi-expert structure and threshold discrimination mechanism, this invention can adaptively denoise under various observation objects and imaging conditions, significantly improving the generalization ability of the denoising method. This invention, through a multi-expert mechanism, overcomes the shortcomings of current denoising methods, such as weak generalization and applicability limited to specific observation sample types or imaging conditions.

This invention relates to a method and system for detecting anomalies in wastewater treatment processes based on a hybrid expert model. The method includes: constructing a multimodal input vector for the wastewater treatment plant; inputting the multimodal input vector into a hybrid expert model, and performing anomaly inference through multiple expert networks selected by sparse gating to obtain anomaly prediction results; dynamically calculating the weight coefficients of each expert network based on the feature distribution of the multimodal input vector, and using the weight coefficients to perform weighted fusion of the anomaly prediction results to generate a preliminary comprehensive detection result; performing consistency verification on the anomaly prediction results, and triggering a thought chain inference mechanism for analysis if the verification fails, and updating the preliminary comprehensive detection result based on the analysis results; performing multi-level risk assessment and decision-making on the optimized comprehensive detection result, and generating a process anomaly report including risk level and disposal recommendations. This invention can reduce false alarms and false negatives, and improve the accuracy and efficiency of anomaly identification and decision-making in wastewater treatment plants.

The invention discloses a leg posture analysis and rating system for a football player in a ball kicking process, and the system comprises a data collection unit which collects four-stage kinematics parameters of kicking, kicking and stretching, backward swinging, forward swinging and ball touching through a three-dimensional motion capture system and a high-speed camera; the data processing unit is used for processing parameters and extracting characteristic parameters of each stage; the expert evaluation interface is used for receiving evaluation data of multiple experts on different ball kicking technology types on the multi-score influence factors; the weight calculation unit is used for calculating an influence factor attribute weight by using a Bayesian method based on expert data; the rating decision matrix construction unit is used for constructing a comprehensive matrix in combination with a weight rough projection method; the multi-angle analysis unit is used for performing multi-angle risk rating by fusing a rough set theory and an approximate ideal solution sorting method; and the result output unit is used for outputting the posture quality grade and the training suggestion. According to the method, the actual requirements of football training on precision, standardization and comprehensiveness of posture analysis can be met.

The application provides a resource allocation method, an electronic device, a storage medium and a program product, relates to the technical field of computers, and is used for predicting the probability of passenger cancellation of resources to determine whether to allocate resources to passengers. The method comprises the following steps: determining initial features based on passenger data of a target passenger and resource data of to-be-issued resources; inputting the initial features into each expert network respectively to obtain multiple expert extraction features; each expert network corresponds to a resource allocation target; the expert network is used for extracting the correlation between the multiple initial features related to the resource allocation target to obtain the expert extraction features; weighting and fusing the multiple expert extraction features based on the weight corresponding to each resource allocation target in the multiple resource allocation targets to obtain fused features; predicting the prediction probability of the target passenger using the to-be-issued resources based on the fused features; and if the prediction probability is greater than or equal to a preset value, allocating the to-be-issued resources to the target passenger.

The invention discloses a radar intelligent interference suppression software test evaluation method, and relates to the technical field of radar test, and the method comprises the steps: S1, constructing a radar intelligent interference suppression software evaluation index system, and storing multi-dimensional indexes in the form of a knowledge graph; s2, designing a radar intelligent interference suppression software comprehensive evaluation algorithm, and quantitatively representing randomness and fuzziness in an evaluation process through effective data analysis, screening and integration of scoring opinions of multiple experts; s3, performing radar intelligent interference suppression software test evaluation; according to the method, a radar intelligent interference suppression software evaluation index system in a real application environment is constructed, the defects of a traditional index in the aspect of evaluating the operation capability and intelligence of radar software are overcome, and a comprehensive evaluation algorithm is provided; according to the method, the problem of evaluation deviation caused by large cognitive difference of people on the new concept of radar software intelligence can be solved, so that the evaluation result is more in line with subjective cognition.

Technologies for efficiently generating issue tracker tickets include a compute device. The compute device includes circuitry configured to obtain a request message. The request message may be indicative of a request for assistance with a corresponding technical problem. The circuitry may be further configured to simulate a conversation among multiple experts to identify an approach to resolving the technical problem. Additionally, the circuitry may be configured to generate an issue tracker ticket based on the simulated conversation. Other embodiments are also described.

The invention discloses a large model dialogue control method and system, a storage medium and computer equipment. The large model dialogue method comprises the steps of converting user input data and dialogue historical data into state perception vectors; selecting a plurality of expert models matched with the state sensing vector from a preset expert model pool; respectively inputting the user input data and the dialogue historical data into the selected expert models to obtain a plurality of candidate replies; and calculating a value score of each candidate reply, and outputting the candidate reply with the highest value score as a final dialogue reply. Through the above method, dialogue context semantics and user core intentions can be accurately captured, dynamic sparse activation and integration of knowledge of a multi-expert model can be realized, computing resource consumption and reasoning delay can be significantly reduced, dialogue reply accuracy and reliability and cross-style and cross-field adaptivity can be improved, and dialogue reply experience can be improved. And the process has relatively high interpretability and controllability, so that debugging and optimization of a large model are facilitated.

This application provides a multi-task recommendation method and apparatus that integrates prior information, relating to the field of artificial intelligence and also applicable to the financial field. The method includes: generating expert representation vectors for each expert and a user representation vector for the current questioning user based on embedded data corresponding to question-and-answer prior information; wherein the embedded data includes all user question data and all expert answer data; the question-and-answer prior information includes the current question information of the current questioning user; determining multiple experts corresponding to the current question information based on the user representation vector and the expert representation vector; and recommending the optimal answer corresponding to the current question information from all answers provided by the multiple experts. This application can integrate pre-stored question-and-answer prior information for multi-task recommendation, utilizing the strong correlation between data to help users quickly retrieve the optimal answer to their questions.

The application discloses an automatic diagnosis algorithm for adaptive multi-agent cooperation for explainability. The application comprises the following steps: firstly, a two-way probability network of symptoms and diseases is constructed based on a data set, and a disease weight is calculated through a pre-diagnosis module to determine whether to enter a Doctor Agent module or a Med-Team Agents module. The Doctor Agent simulates single expert diagnosis, combines a medical knowledge base and the two-way probability network, and judges whether a patient can be definitely diagnosed or continue to ask related symptoms to assist in diagnosis. The Med-Team Agents simulate the consultation of multiple experts, including a Manager Agent and multiple Disease Agents, which respectively represent a chief physician and multiple disease experts. The Manager Agent dynamically manages multiple Disease Agents, selects appropriate experts according to the symptom information of the patient to generate related symptoms, and finally completes diagnosis. The application is suitable for medical automatic diagnosis tasks, combines multi-agent cooperation with two-way probability network prompts, improves the explainability of diagnosis, and realizes efficient inquiry and accurate diagnosis.

The invention provides a multi-modal fusion method, module and system based on learnable routing, relates to the technical field of information fusion, and solves the problems of limited model performance and insufficient robustness in multi-modal fusion. The method comprises the following steps: firstly, processing coding features of each mode to obtain preliminary fusion features; inputting the preliminary fusion features into a routing network, performing linear projection and temperature parameter scaling, and normalizing to generate routing weight distribution corresponding to a plurality of expert modules; inputting the preliminary fusion features into a plurality of expert modules in parallel, carrying out nonlinear transformation, and outputting corresponding transformed features; and performing weighted summation on the routing weight distribution and the transformed features output by each expert module to generate final fusion features. According to the method, efficient, robust information fusion suitable for multiple modes is carried out for encoding results of multiple mode encoders, and therefore downstream tasks such as time sequence-image collaborative prediction, multi-mode classification and multi-mode retrieval can be better adapted.

The application discloses a language model structure suitable for super-long text, comprising: after receiving a multi-modal data stream, an asynchronous multi-modal fusion layer performs cross-modal feature fusion through a cross-modal attention gate unit when semantic correlation exceeds a threshold to obtain a fusion input feature; a recurrent state space unit processes the fusion input feature to obtain a state vector; a hybrid expert attention module is used to combine a global sparse attention mechanism and a local window attention mechanism, and an attention routing network is used to select an optimal attention calculation mode for different Token features; a routing prediction network of an MoE dynamic routing layer predicts the load demand of subsequent Token according to the Token feature distribution in the current window, adjusts the routing weight of the dynamic gate network, and uniformly distributes the calculation task to multiple expert networks. The application is efficient in calculation, economical in memory, has long-term memory capacity, is balanced in load, and can accurately process multi-modal information.

The application discloses a kind of robust preference aggregation method and system of multi-expert group decision-making, take the expert condition preference matrix of multiple experts to multiple candidate scheme, execute tailing robustness processing and normalization processing with numerical protection according to expert line, obtain robust initial opinion matrix;Similarity trust matrix and uniform trust matrix with diagonal line as zero and non-diagonal uniform distribution are constructed based on the matrix, and the final trust matrix is obtained by convex combination fusion of the two through shrinkage coefficient;Robust initial opinion matrix and final trust matrix are input Friedkin-Johnsen opinion evolution model to solve balanced opinion matrix, and the group score is obtained by weighting aggregation according to preset expert aggregation weight, and the sorting result is output.

The invention belongs to the technical field of dexterous hand grabbing models, and particularly relates to a reinforcement learning-based and multi-expert hybrid model which is established by the following steps: step 1, preparing a diversified data set: constructing an object data set containing various morphological and geometric features, a simulation model of the dexterous hand is combined for training and evaluation; and step 2, training an expert strategy by using a reward function enhanced PPO: constructing a universal-expert hybrid model of dexterous hand operation based on reinforcement learning. According to the method, efficient generalization of dexterous hand grabbing is achieved through a staged learning strategy; the method comprises the following steps of: firstly, acquiring a high-performance expert model with pertinence to different objects and operation modes by utilizing a basic expert strategy of reinforcement learning training dexterous operation; then, under a Generalist-Specialist Learning framework, a plurality of expert strategies are distilled step by step into a Generalist strategy with a more compact structure, and then the Generalist strategy with the more compact structure is obtained; the method is different from a traditional direct distillation method.

This application relates to a method and related apparatus for generating long-duration videos based on an adaptive video world model. The method includes: inputting an initial video block and semantic instructions into an adaptive video world model, which represents a comprehensive model pre-trained based on adaptive block partitioning to mix multiple expert models; dynamically routing among the multiple expert models to match a target expert model based on the initial video block and semantic instructions, and generating the next video block according to the diffusion computation process of the target expert model, the length of the next video block corresponding to the processing granularity of the target expert model; for the newly generated video block, iteratively matching the corresponding expert model and generating corresponding video blocks through an autoregressive approach until a task termination condition is met, and then sequentially combining all video blocks to obtain the target long-duration video. This method can improve the long-term stability and semantic coherence of the output results in long-duration video generation scenarios.

The invention discloses a power grid control method and device based on a hybrid expert network and a medium, and belongs to the technical field of power grid optimization, and the method comprises the steps: obtaining power grid data which comprises node voltage, load demands, generator output and energy storage charging and discharging data; based on the power grid data, an intelligent scheduling model driven by multi-modal data is constructed, and the intelligent scheduling model comprises a plurality of expert sub-networks composed of a steady-state regulation and control model, a deep reinforcement learning model, a model prediction control model, a new energy consumption model and dynamic safety evaluation, and dynamically allocating expert weights through the gating network to realize self-adaptive decision making. Through the combination of the dynamic expert network and the lightweight gating mechanism, the multi-modal data of the power grid can be analyzed in real time, the optimal expert combination is activated in a self-adaptive manner, and dynamic spatial-temporal characteristic capture is realized. Compared with a traditional deep reinforcement learning method, the method is remarkably improved, and excellent self-adaptability and real-time performance are shown.

The application discloses a kind of ancient book Chinese character recognition methods based on lightweight convolutional neural network.Integrating multiple expert models, in the training process, the prediction probability distribution difference item is added to the loss function, the prediction deviation and variance on the whole class are reduced, so as to improve the recognition accuracy on all classes, while introducing model lightweight technology reduces the calculation overhead and parameter quantity of integrated model, realizes the long-tail sample recognition model of lightweight ancient book Chinese character.The model obtained by the application can effectively improve the recognition accuracy on the rare word sample set on the seriously unbalanced ancient book Chinese character dataset, and the calculation complexity and parameter quantity of the conventional lightweight model are similar.

The application relates to the technical field of industrial process prediction, in particular to a tundish erosion prediction method based on a hierarchical hybrid expert framework. The method comprises the following steps: collecting tundish time series physical field data, screening features and constructing a unified feature space; shunting the feature space to obtain time series features and statistical aggregation features; training a classifier using the statistical aggregation features to generate calibrated posterior probability and construct a gating network; generating an initial mode subset based on the posterior probability and training a corresponding expert model; for the to-be-tested data, obtaining the confidence thereof from the gating network, and selecting a single expert model for prediction or fusing multiple expert models for weighted prediction through an adaptive strategy according to whether the confidence exceeds a threshold; and finally inversely transforming and reconstructing the prediction value into an absolute value of the erosion thickness. The application effectively improves the accuracy and adaptability of tundish erosion prediction.

The application belongs to the technical field of dexterous hand grasping model, and in particular to a reinforcement learning and multi-specialist hybrid model, and the establishment of the reinforcement learning and multi-specialist hybrid model comprises the following steps: step 1: preparation of a diversified data set: an object data set containing various morphological and geometric characteristics is constructed, and a simulation model of a dexterous hand is combined for training and evaluation; step 2: training of an expert strategy using a reward function enhanced PPO: a generalist-specialist hybrid model for dexterous hand operation based on reinforcement learning is constructed. The application realizes efficient generalization of dexterous hand grasping through a phased learning strategy; first, a basic expert strategy for dexterous operation is trained using reinforcement learning, and a high-performance expert model targeted at different objects and operation modes is obtained; subsequently, under the Generalist-Specialist Learning framework, multiple expert strategies are gradually distilled into a more compact generalist strategy; unlike the traditional direct distillation method.

This invention provides an intelligent question-answering method and apparatus that combines experts with weight decomposition low-rank adaptation, relating to the field of natural language processing technology. The method includes: a multi-task question-answering model generating answers to questions; the training process of the multi-task question-answering model includes: acquiring historical question-answering text data; adding task identifiers and supervision labels to the historical question-answering text data and performing preprocessing to obtain a sample dataset; obtaining expert output results based on the samples; the multi-task question-answering model includes a shared backbone network module, an expert allocation module, and multiple expert sub-modules; introducing a weight decomposition low-rank adaptation structure into the shared backbone network module and multiple expert sub-modules; applying Steenfer manifold constraints and Riemann optimization to the direction parameters in the weight decomposition low-rank adaptation structure; and training the multi-task question-answering model based on the sample dataset to obtain a trained multi-task question-answering model; and outputting the generated answers to the user. This invention can achieve efficient and reliable intelligent question answering.