Patents
Literature
Patsnap Eureka AI that helps you search prior art, draft patents, and assess FTO risks, powered by patent and scientific literature data.

73 results about "Search problem" patented technology

In computational complexity theory and computability theory, a search problem is a type of computational problem represented by a binary relation. If R is a binary relation such that field(R) ⊆ Γ⁺ and T is a Turing machine, then T calculates R if: If x is such that there is some y such that R(x, y) then T accepts x with output z such that R(x, z) (there may be multiple y, and T need only find one of them) If x is such that there is no y such that R(x, y) then T rejects x Intuitively, the problem consists in finding structure "y" in object "x".

Elastic distributed Nash equilibrium search method and device under FDI attack

The invention discloses an elastic distributed Nash equilibrium search method and device under FDI attack, and relates to the technical field of non-cooperative game decision, the method comprises the following steps: constructing a multi-agent network, forming a non-cooperative game, and defining an objective function of each agent in the multi-agent network; establishing a malicious FDI attack model; the malicious FDI attack model maps false data injection attack to control input of an agent; an elastic self-adaptive distributed Nash equilibrium search algorithm is designed based on the target function, so that behaviors of other intelligent agents are estimated under the condition that the intelligent agents cannot obtain global information and attack information; the elastic self-adaptive distributed Nash equilibrium search algorithm comprises a leader-follower consistent consensus estimation-based identifier signal designed by an observer and a self-adaptive dynamic gain part. The elastic self-adaptive distributed Nash equilibrium search method solves the problem of elastic self-adaptive distributed Nash equilibrium search when the non-cooperative game is attacked by unknown FDI.
Owner:BEIHANG UNIV

Superconducting quantum chip wiring and optimizing method

The invention discloses an automatic wiring and optimizing method for a superconducting quantum chip, and belongs to the field of automatic design of superconducting quantum chips. According to the method, in a modeling stage, a'combined wiring template 'model under a fine-grained grid is adopted, a problem is discretized through a combined wiring template, complex geometric operation of the arc wiring problem in the solving process is avoided, and the search space of an algorithm is greatly reduced; in the legal solution search stage, a single-side wiring algorithm is adopted, a line sequence search problem is extended according to wiring result characteristics of the algorithm, a line sequence tree search framework is designed, and on the basis, the blocking relation between lines is explored, and a series of search optimization strategies are designed; an exchange winding action is adopted to meet the special wire sequence requirement of a read wire; in the wiring optimization stage, an iterative optimization wiring scheme based on space limitation is adopted, and multi-stage target optimization is performed on the overall wiring result on the basis of a legal solution, so that the wiring result meets design requirements.
Owner:HUAZHONG UNIV OF SCI & TECH

Interactive command line tool-oriented intelligent task planning and execution method and system

The invention provides an interactive command line tool-oriented intelligent task planning and execution method and system, and the method comprises the steps: S1, generating a corresponding command sequence template for each operation of a target tool through a first large language model, and constructing an executable knowledge base; s2, calculating the semantic similarity between the task description and each command sequence template to obtain a candidate set, and selecting the most relevant template as a retrieval context by using a second large language model; s3, modeling a task execution process as a state diagram search problem, selecting candidate nodes from reachable nodes of a current node based on an algorithm, and selecting an optimal node according to a current state and a retrieval context by utilizing a second large language model; s4, executing an execution action corresponding to the optimal node, and analyzing tool output by using a second large language model to adjust a heuristic value of the corresponding node and execute a corresponding rollback strategy; and S5, repeating the steps S3 to S4 until the reasoning result shows that the task reaches the target node or the number of attempts exceeds a preset threshold.
Owner:FUJIAN NORMAL UNIV

Multi-objective optimization method for radiotherapy plan

The invention discloses a multi-objective optimization method for a radiotherapy plan, and the method comprises the steps: S1, building a multi-objective optimization model, and enabling the multi-objective optimization model to be used for the multi-objective intensity optimization of a radiotherapy reverse intensity modulation plan, and generating a Pareto solution; step S2, inputting a Pareto solution currently generated by the multi-objective optimization model into a weight search network model, wherein the weight search network model predicts a sub-objective weight of a next round of multi-objective intensity optimization of the multi-objective optimization model according to the current Pareto solution; s3, optimizing a single-target strength optimization problem with sub-target weights through a multi-target optimization model to update a current Pareto solution; and S4, circularly executing the steps S2 to S3 until a Pareto solution generated by the multi-objective optimization model forms a Hausdorff distance dHaus (Zin, Zout) between an upper bound and a lower bound of the Pareto in space, the distance dHaus (Zin, Zout) is less than threshold, and the threshold is a predefined threshold parameter which is greater than 0. According to the invention, a deep learning method is creatively used to solve the problem of constraint weight search in a radiation therapy reverse plan multi-objective optimization method.
Owner:SUPERACCURACY SCIENCE & TECHNOLOGY CO LTD

Intelligent question and answer method and system

The invention discloses an intelligent question and answer method and system. The method comprises the steps of obtaining a target question text input by a target user; determining a matching result of the target question text and the question and answer sheet; if the matching is successful, feeding back a first target structured answer result corresponding to the question text matched with the target question text to the target user; if the matching fails, analyzing the target problem text by utilizing a classification model, determining a service type corresponding to the target problem text, and determining at least one service keyword related to the service type in the target problem text; and taking each business keyword as a query condition, determining a second target structured answer result corresponding to the query condition in a preset database by adopting an inverse indexing method, and feeding back the second target structured answer result to the target user. The technical problem that a user needs to consume a lot of time to search for answers to questions in a document due to the fact that a related question answering system can only output a query list related to the questions and cannot output structured answer results is solved.
Owner:CHINA TELECOM CORP LTD

Self-adaptive data access account checking method and device and medium

The invention provides a self-adaptive data access account checking method and device and a medium, and belongs to the technical field of computer information processing. The method comprises the following steps: selecting a source database, a source data table, a target database and a target data table for account checking, selecting primary keys of the source data table and the target data table as a data comparison basis, and setting an initial account checking frequency; reading data of the source database and the target database through Spark SQL, and performing distributed preprocessing on the data of the source database; generating an MD5 abstract for each row of data in the source data table and the target data table through an MD5 algorithm, and adding the generated MD5 abstract as a new column to the data table; searching the MD5 abstract according to the main key comparison, determining a difference data line, searching problem data according to the difference data line, and recording an account checking result; and adjusting the account checking frequency according to the current data account checking result through a Q-learning algorithm. The account checking speed and accuracy are improved, computing resources are utilized more efficiently, and hardware resource consumption is reduced.
Owner:SHANDONG FUTURE NETWORK RES INST (PURPLE MOUNTAIN LAB IND INTERNET INNOVATION APPL BASE)

A human-like complex question retrieval method, electronic device and storage medium

The present invention provides a human-simulated complex question retrieval method, electronic device and storage medium, the method comprising: disassembling an original question statement to obtain a plurality of atomic question statements; determining the execution order between the plurality of atomic question statements according to the association relationship between the plurality of atomic question statements; determining the target retrieval source corresponding to each atomic question statement according to the statement type corresponding to each atomic question statement; searching the atomic question statement through the target retrieval source corresponding to each atomic question statement in turn according to the execution order between the plurality of atomic question statements to obtain the original answer statement corresponding to each atomic question statement; semantically integrating the plurality of original answer statements to obtain the target answer statement corresponding to the original question statement, so as to effectively solve the retrieval problem of complex questions through the collaborative work of multiple modules such as intention recognition, problem disassembly, logic planning, extension generation and dynamic execution.
Owner:北京中科闻歌科技股份有限公司

Prompt word processing method and device, equipment and storage medium

The invention provides a cue word processing method and device, equipment and a storage medium, and relates to the field of artificial intelligence, in particular to the fields of large models, big data, intelligent search and the like. According to the specific implementation scheme, multiple pieces of sample data are obtained, wherein the sample data comprise search problems belonging to a target field and target categories of the search problems; based on the multiple pieces of sample data, determining classification standard information conforming to each preset category in the target field; based on the plurality of candidate categories of each search question and the classification standard information of the candidate categories, splicing to obtain an initial cue word of each search question; the initial prompt word is used for indicating a category which the search problem conforms to is determined from a plurality of candidate categories; the plurality of candidate categories are at least partial categories in a plurality of preset categories; detecting the initial cue word of each search question to obtain a total detection result which is used for evaluating the cue effect of the initial cue word of each search question so as to adjust the initial cue word based on the total detection result.
Owner:BEIJING BAIDU NETCOM SCI & TECH CO LTD

Multi-agent visual reasoning method, device and equipment and storage medium

The application provides a multi-agent visual reasoning method, device and equipment and a storage medium. The method comprises the following steps: formalizing a visual reasoning task into a Monte Carlo tree search problem in a Markov decision process framework; using a multi-agent collaborative framework, taking each to-be-processed image and an initial query as a root node state, iteratively performing selection, expansion, simulation and back propagation steps of the Monte Carlo tree search to solve the Monte Carlo tree search problem, and obtaining a constructed search tree; stopping the search when the number of iterations reaches a preset threshold, and using a global thinking agent in the multi-agent collaborative framework to generate a final answer of the visual reasoning task according to a plurality of terminal node states of the constructed search tree. The application suppresses cascade errors caused by single chain reasoning path through multi-path exploration, and avoids the inefficiency of wide search in tree reasoning through a reward back propagation mechanism.
Owner:BEIJING UNIV OF POSTS & TELECOMM

Extensible deep neural network channel pruning method and system

The invention relates to the technical field of artificial intelligence, in particular to an extensible deep neural network channel pruning method and system. According to the method, a channel pruning problem is converted into a structure search problem, and an artificial bee colony algorithm is provided to find an optimal pruning structure. A coupled model structure is mapped into a non-coupled search structure through structure mapping, and an optimal pruning structure is searched and fine-tuned based on an artificial bee colony algorithm, so that the search space of the deep neural network is reduced, and the pruning problem of the coupled structure is solved.
Owner:NORTHWEST A & F UNIV

A two-stage key region semantic map construction method based on Hough transform

The present invention discloses a two-stage key area semantic map construction method based on Hough transform. Aiming at the problem that existing UAV environment modeling involves a large amount of calculation and cannot always guarantee the real-time processing of actual ranging images on currently available airborne computing devices, a two-stage linear plane extraction method is proposed. In the first stage, the three-dimensional space of the original point cloud is decomposed into a two-dimensional plane subspace, and the search problem of the plane object in the three-dimensional space is converted into the search problem of the straight line segment in the two-dimensional subspace (scanning plane), thereby significantly reducing the amount of calculation by reducing the dimensions of the search object and the search space; the second stage also does not involve a large amount of calculation, because the search for the plane object is not based on the original point cloud, but on the straight line segments extracted in the first stage, and the number of these straight line segments is much less than the number of points in the original image, which further reduces the amount of calculation, thereby improving the autonomous task execution efficiency of the UAV in the industrial city environment.
Owner:SOUTHWEAT UNIV OF SCI & TECH

A mobile robot two-dimensional full coverage path planning method and related equipment

The application discloses a two-dimensional full coverage path planning method for a mobile robot and related equipment, and belongs to the technical field of robot full coverage path planning. In a complex two-dimensional environment model, for example, a two-dimensional map in which irregular obstacles and concave-convex polygon obstacles exist, firstly, Maklink lines generated by Maklink graph theory can effectively divide the map into multiple sub-regions, which is helpful to simplify the path search problem and improve the search efficiency; then, the order in which the robot visits each sub-region is calculated, and the optimal path for the robot to switch between each sub-region is calculated; finally, the Boustrophedon path is used for "bow" shape traversal, which can ensure that the robot efficiently covers the entire sub-region, reduces the possibility of omission and repeated cleaning, and moves between sub-regions according to the optimal switching path and the visiting order, which is helpful to reduce the repeated path and invalid movement of the robot when traversing the sub-region.
Owner:CHENGDU UNIVERSITY OF TECHNOLOGY

Direct location method for non-circular signals based on polynomial root finding and weighted fusion

The present invention discloses a non-circular signal direct positioning method based on polynomial root finding and weighted fusion, which is characterized by comprising the following steps: first, using the elliptical covariance information of the target signal to expand the spatial information and obtain an enlarged virtual array aperture; second, the high-dimensional spectrum function search problem of each base station is converted into a polynomial root finding problem by a root finding method; then, combining the information of all base stations to construct a spectrum function without complex multiplication to reduce the computational complexity; then, by weighting the received data of each base station, the algorithm estimation performance is improved; finally, the weighted fusion cost function is searched for the spectrum peak, and the one with the largest function value is selected. K The point is the target position. A non-circular source direct positioning method designed by the present invention has higher spatial freedom and positioning accuracy than the traditional direct positioning algorithm, and can estimate more targets. In addition, the method significantly reduces the computational complexity while ensuring the estimation performance.
Owner:NANJING UNIV OF AERONAUTICS & ASTRONAUTICS

A deadline-aware deep learning training task scheduling method and system for efficiency

The present invention discloses a deadline-aware deep learning training task scheduling method and system for efficiency. The method mainly includes two steps: deep learning training task running time estimation and deep learning training task scheduling optimization. First, a threshold-based adaptive sampling technology is used to obtain an accurate estimation result of the deep learning training task running time. Then, the scheduling problem in the heterogeneous GPU cluster is converted into a search problem for task arrangement on each GPU. The optimized scheduling result that meets the scheduling goal is searched, and the search space is reduced through a variety of pruning methods. Compared with the prior art, the present invention innovatively proposes a unified expression form of scheduling goals, which can simultaneously meet the scheduling goals of minimizing the deadline violation amount of the task and the overall task completion time, and can efficiently schedule deep learning tasks, which is conducive to improving the efficiency of task deployment and operation in heterogeneous GPU clusters.
Owner:INST OF SOFTWARE - CHINESE ACAD OF SCI

Intelligent task planning and execution method and system for interactive command line tools

The application provides an intelligent task planning and execution method and system for an interactive command line tool, comprising the following steps: S1, using a first large language model to generate a corresponding command sequence template for each operation of the target tool, and constructing an executable knowledge base; S2, calculating the semantic similarity between the task description and each command sequence template to obtain a candidate set, and using a second large language model to select the most relevant template as a retrieval context; S3, modeling the task execution process as a state graph search problem, selecting a candidate node from the reachable nodes of the current node based on an algorithm, and using the second large language model to select the optimal node according to the current state and the retrieval context; S4, executing the execution action corresponding to the optimal node, using the second large language model to parse the tool output, adjusting the heuristic value of the corresponding node and executing the corresponding rollback strategy; and S5, repeating steps S3 to S4 until the reasoning result shows that the task reaches the target node or the number of attempts exceeds the preset threshold.
Owner:FUJIAN NORMAL UNIV

A Vehicle Route Planning Method in a Complex Environment

The present invention relates to a vehicle path planning method in a complex environment and belongs to the field of intelligent transportation technology. This method adopts an incremental loading and search path planning algorithm, reusing the planning results of previously loaded areas in the incremental loading path planning algorithm to further improve the speed of path planning. An incremental method is used when loading map data to avoid loading a large amount of map data at one time, significantly reducing the geographical information processing time. Within the loaded area range, an incremental search path planning algorithm that breaks path symmetry is used to further improve the path planning speed. At the same time, aiming at the symmetric repeated search problem of the existing heuristic search algorithm, a straight-line trajectory deviation value is introduced to distinguish symmetric points with the same original priority to break the path symmetry, reduce the exploration of redundant grids by the algorithm, and improve the planning speed.
Owner:BEIJING INST OF TECH

A control law automatic design method based on long short-term memory network

ActiveCN117826591BAchieve integrated optimizationreduce dependenceAdaptive controlSearch problemNetwork architecture
The application discloses a control law automatic design method based on a long short-term memory network. The method draws on the experience of network architecture search method in deep learning, utilizes the advantage of a recurrent neural network in time sequence correlation exploration, converts a control law structure design problem into a directed acyclic graph topological relation automatic search problem, realizes automatic generation of a flight control law control structure, and realizes automatic setting of parameters under a given control law structure based on a genetic algorithm. The method aims at overcoming the limitation that current control law automatic optimization can only utilize a heuristic algorithm to automatically set controller parameters for a known controller structure, reducing the artificial design workload, and improving the control effect under complex design input conditions.
Owner:BEIJING AEROSPACE AUTOMATIC CONTROL RES INST

Information processing device, information processing system, program, and ising model development assistance method

With respect to an information processing device that supports creation of an Ising model for causing an annealing-type optimization machine to solve an optimum solution search problem, the information processing device includes a transforming unit configured to binarize an explanatory variable included in a training data set created using a trained machine learning model; a training unit configured to train an Ising model by performing machine learning with a relationship between the binarized explanatory variable and a predicted value of the training data set; and an output unit configured to output the trained Ising model.
Owner:RESONAC CORP

Question and answer search method and device, equipment and storage medium

The invention discloses a question and answer search method, device and equipment and a storage medium, and relates to the technical field of computers, and the question and answer search method comprises the steps that in response to a to-be-searched question input by a user, intention information corresponding to the to-be-searched question is recognized; based on the intention information, determining at least one target search platform of which the search attribute is matched with the intention information in a plurality of preset search platforms; searching the to-be-searched question in each target search platform to obtain candidate answers corresponding to the to-be-searched question; and based on the candidate answers, generating an output answer corresponding to the to-be-searched question. According to the method, the proper search platform can be accurately matched through intention recognition for directional question and answer search, answers meeting user requirements can be more accurately searched, the question and answer search accuracy is effectively improved, in addition, the search accuracy is guaranteed, meanwhile, accurate matching and directional search are achieved, and the search efficiency is improved. And the search efficiency and effect can be effectively improved.
Owner:BEIJING QIHOOD TECHNOLOGY CO LTD

Neural network channel number search method and apparatus

A neural network channel number searching method and device can realize a differentiable search technology to search network channel numbers, reduce the calculation complexity of the network while ensuring the network performance. The method comprises the following steps: determining the output channel number N of a convolutional layer, wherein N is a positive integer (S301); dividing the feature tensor output by the convolutional layer into n sub-feature tensors, wherein the channel number of each sub-feature tensor is N / n, n is an integer that can be divided by N, and n is greater than or equal to 2 (S302); determining n sets of weighting coefficients, wherein each set of weighting coefficients comprises a plurality of weighting coefficients, and the plurality of weighting coefficients correspond to a plurality of sub-feature tensors in the n sub-feature tensors one by one (S303); determining the sub-feature tensor corresponding to the maximum value in each set of weighting coefficients (S304) to obtain the sub-feature tensors corresponding to the n maximum values; and re-determining the output channel number of the convolutional layer according to the sub-feature tensors corresponding to the n maximum values (S305).
Owner:HUAWEI TECH CO LTD

Hough transform-based two-stage key region semantic map construction method

The invention discloses a two-stage key area semantic map construction method based on Hough transform, and provides a two-stage linear plane extraction method for solving the problem that existing unmanned aerial vehicle environment modeling relates to a large amount of calculation and does not always ensure that real-time processing of actual ranging images is achieved on currently available airborne computing equipment. In the first stage, a three-dimensional space of an original point cloud is decomposed into a two-dimensional plane subspace, and a search problem of a plane object in the three-dimensional space is converted into a search problem of a straight line segment in the two-dimensional subspace (a scanning plane), so that the calculated amount is remarkably reduced by reducing the dimensions of the search object and the search space; the second stage also does not involve a large amount of calculation, because the search of the planar object is not based on the original point cloud but based on the straight line segments extracted in the first stage, and the number of the straight line segments is far less than the point number in the original image, the calculation amount is further reduced, and the autonomous task execution efficiency of the unmanned aerial vehicle in the industrial city environment is further improved.
Owner:SOUTHWEAT UNIV OF SCI & TECH

Information processing apparatus, display control method, and storage medium

Technical ProblemTo provide a time efficient Neural Architecture Search for the Backbone block of Computer Vision task.Solution to ProblemA neural architecture searching apparatus comprises building means (11) to build a supernetwork, wherein a target layer of the supernetwork to be optimized is replaced by a plurality of candidate layers, and the supernetwork comprises a plurality of fully-connected layers; training means (12) to train the supernetwork, wherein the plurality of candidate layers are trained part by part, and the plurality of fully-connected layers are trained correspondingly to the part of the plurality of candidate layers; and selecting means (13) to evaluate the trained supernetwork and select a part of the plurality of candidate layers which corresponds to the best performing part of the plurality of fully-connected layers.
Owner:NEC CORP

Data processing program, data processing device, and data processing method

A data processing program causing a computer to execute: receiving user definition information including: function definition information defining an objective function of a combinatorial optimization problem and neighborhood definition information defining neighborhood states at a time of searching for a solution to the problem by a local search; generating an optimization system using the user definition information and base information indicating processing independent of: a definition of the function and a definition of the neighborhood states; receiving a coefficient in the function; and executing the optimization system to: calculate a change amount of a value of the function when a state represented by state variables in the function transitions to the neighborhood states, using the function definition information, the neighborhood definition information, and the coefficient, select a first state among the neighborhood states based on the change amount, and update the state variables to the first state.
Owner:FUJITSU LTD

A superconducting quantum chip wiring and optimization method

The present invention discloses a method for automatic wiring and optimization of superconducting quantum chips, belonging to the field of automated design of superconducting quantum chips. In the modeling phase, the method adopts a "combined wiring template" model under a fine-grained grid. The combined wiring template is used to discretize the problem, avoiding the complex geometric operations in the arc wiring problem solution process and significantly reducing the algorithm's search space. In the legal solution search phase, a one-sided wiring algorithm is adopted. Based on the wiring result characteristics of the algorithm, the line sequence search problem is extended and a line sequence tree search framework is designed. On this basis, the blocking relationship between lines is explored and a series of search optimization strategies are designed. An exchange winding action is adopted to meet the special line sequence requirements of the read line. In the wiring optimization phase, an iterative optimization wiring scheme based on space constraints is adopted. Based on the legal solution, the overall wiring result is optimized at multiple levels of objectives to ensure that the wiring result meets the design requirements.
Owner:HUAZHONG UNIV OF SCI & TECH

Maximum inner product vector retrieval-oriented graph indexing method

The invention discloses a graph indexing method oriented to maximum inner product vector retrieval, which utilizes query vector scaling to prove that a maximum inner product search problem can be equivalently converted into a nearest neighbor search problem, so that an efficient Euclidean graph indexing structure is reused on the premise that a vector space does not need to be changed; according to the method, a graph structure based on a spherical path is constructed, a cross-norm connecting edge is introduced on the basis of a traditional sparse graph to relieve norm bias, and a normalized clustering and high norm screening strategy is adopted to generate an entry point to improve search coverage. Meanwhile, the invention provides a self-adaptive termination mechanism based on search process features, and whether the search is ended in advance is judged through a lightweight decision tree model, so that the search overhead is effectively reduced. Experiments show that the retrieval efficiency is remarkably improved on the premise of ensuring the accuracy, and the method is suitable for high-performance vector retrieval scenes such as recommendation systems and e-commerce search.
Owner:ZHEJIANG UNIV

Shopping guide method and device based on artificial intelligence, electronic equipment and storage medium

The embodiment of the invention discloses a shopping guide method and device based on artificial intelligence, electronic equipment and a storage medium, belongs to the technical field of artificial intelligence, and solves the problem of insufficient shopping guide personalization. Comprising the steps of obtaining a search question which is input by a user and is related to a target commodity, obtaining user preference information of the user in preset multiple dimensions, and generating a user preference feature vector based on the user preference information in each dimension; performing serialized splicing processing on the search question and the user preference feature vector to obtain target data, and inputting the target data into a trained large language model; and based on the trained large language model, outputting a target shopping guide answer of the search question, so as to determine a target commodity based on the target shopping guide answer.
Owner:ZHENGZHOU APUS DIGITAL CLOUD INFORMATION TECH CO LTD

Unsupervised rapid radio storm searching method based on time sequence anomaly detection

The invention discloses an unsupervised rapid radio storm searching method based on time sequence anomaly detection. The method comprises the following steps: carrying out achromatic processing on observation data of a radio telescope to obtain a signal intensity time sequence; embedding the signal intensity time sequence into a low-dimensional space and constructing a graph structure, and calculating a sequence abnormal score based on the graph structure to obtain a fast radio burst pulse signal candidate body; and screening the fast radio burst pulse signal candidates by using a barrel counting method, and confirming the arrival time of the fast radio burst pulse signal and the dispersion value corresponding to the fast radio burst pulse signal. According to the method, the fast radio storm searching problem is innovatively solved from the perspective of time sequence anomaly detection, the method has the advantages of being high in robustness and low in false detection rate, the number of candidate bodies is greatly reduced, and the searching efficiency is improved; the consumption of computing resources is greatly reduced by using coarse-grained grid achromatic dispersion; the method does not depend on prior knowledge related to signal shapes, and can be expanded to be used for detecting more unknown universe signals.
Owner:ZHEJIANG LAB

Automatic architecture design method and system for multi-modal fusion neural network

The invention discloses an automatic architecture design method and system for a multi-modal fusion neural network, and the method comprises the steps: defining an original operation pool for the structure search of the multi-modal fusion neural network, carrying out the search processing, and determining a multi-modal fusion cellular network structure; obtaining a basic component for constructing a multi-modal fusion cellular network structure unit, and inserting a residual structure to obtain a multi-modal fusion cellular network unit library; carrying out topology complexity and accuracy evaluation, and carrying out Shapley value calculation on the residual error structure to obtain a Shapley value of the residual error structure; carrying out residual error optimization processing on the multi-modal fusion cellular network structure; and fusion of various sensor data is realized. According to the method, a multi-modal fusion problem can be directly solved as a framework search problem by utilizing a residual structure, and data fusion is realized by fully utilizing each modal information of heterogeneous data. The automatic architecture design method and system for the multi-modal fusion neural network can be widely applied to the technical field of neural network architecture search.
Owner:FOSHAN UNIVERSITY

Classifier-assisted neural network architecture search multi-target multi-level learning optimization method

The invention discloses a classifier-assisted neural network architecture search multi-target multi-level learning optimization method, which comprises the following steps of: modeling by taking a neural network architecture search problem as a multi-target optimization problem, and constructing a search space; constructing a test network for training; screening a Pareto optimal solution through non-dominated sorting, and constructing a special search space based on a Pareto leading edge reserved operator; performing a classifier-assisted hierarchical learning-based multi-objective evolutionary optimization algorithm, selecting an updated population after evolution based on a candidate solution screening strategy, performing objective function evaluation based on a screened solution, updating an archive library, outputting a Pareto optimal solution in the archive library after iterative training, and performing multi-objective evolutionary optimization on the basis of the Pareto optimal solution in the archive library. And selecting a network in the optimal Pareto frontier of the last generation of population as an optimal neural network architecture. According to the method, efficient search of the neural network architecture is realized through the classifier-assisted hierarchical particle swarm optimization algorithm, and a solution meeting the double targets of classification accuracy and reasoning delay is obtained.
Owner:SOUTH CHINA UNIV OF TECH