11results about How to "Maximize accuracy" patented technology

A multi-user multi-exit DNN reasoning partitioning and task scheduling method includes the steps that trained multi-exit DNN models are deployed on each user device and an edge server, and in a time slot, tasks reach the user devices, meet Poisson distribution and are stored in task queues of the user devices; taking the strategy feasibility interval as a constraint, taking an exit point, a partition point and a scheduling decision of the task as optimization variables, and establishing an optimization problem P taking maximization of the total completion rate and the reasoning accuracy of task reasoning in a time slot as a target; decoupling the problem into a server scheduling optimization problem P1, establishing a graph model of a server task, and solving the problem P1 by using a double-greedy algorithm; and simplifying the problem into P2, establishing a Markov decision process, and solving an optimal strategy by using a multi-agent reinforcement learning algorithm. The method is suitable for multi-user equipment to carry out multi-outlet DNN reasoning partition, and the task completion rate and reasoning accuracy within a period of time can be effectively improved.

The invention relates to the technical field of satellite system data transmission, in particular to a satellite system model training method and device, equipment and a medium. The method comprises the following steps: acquiring a distillation data set sent by each distillation satellite; performing centralized training on each distillation data set to obtain a first average gradient; gradient data sent by all the federated satellites are obtained, and a second average gradient of all the federated satellites is generated according to all the gradient data; determining the global gradient of each running satellite; updating the global model parameters according to the global gradient to obtain the global model parameters of the current round, and sending the global model parameters of the current round to the running satellite; and if the global model parameter of the current round does not meet the preset convergence condition, repeating the above steps until the target global model parameter is output to meet the preset convergence condition. According to the method and the device, the problems of overlarge communication load, relatively large privacy security problem and relatively low satellite data analysis capability in the prior art can be solved.

The invention discloses a target tracking-oriented unmanned cluster formation configuration design method, which comprises the following steps of: constructing a target motion model and an observation model of an unmanned cluster to a target; constructing a detection performance evaluation index of the unmanned cluster and a target tracking performance evaluation index of the unmanned cluster; establishing an unmanned cluster initial formation configuration optimization model for target tracking; establishing an unmanned cluster formation track optimization model for target tracking; and respectively solving the unmanned cluster initial formation configuration optimization model facing the target tracking and the unmanned cluster formation track optimization model facing the target tracking in combination with an interior point method. According to the method, the detection performance and the target tracking precision of the unmanned cluster can be maximized, the requirements of unmanned cluster spacing and maneuvering limitation can be met, the optimal formation initial configuration and the optimal formation route planning scheme at the subsequent moment can be obtained, and the detection tracking performance of the unmanned cluster is effectively improved.

The invention discloses a method for constructing, updating and retrieving an action memory library, which belongs to the technical field of computer vision and comprises the following steps of: constructing a training data source with time sequence diversity; initializing an action memory library containing a plurality of learnable prototype matrixes and a double-flow interaction network; memory bank evolution is executed, an action prototype is retrieved by utilizing a query stream, a current memory state is dynamically generated by combining a memory state updating gate mechanism with a historical state, and dynamic memory is injected into a feature space by utilizing memory driving graph convolution; synchronously updating parameters based on multi-target loss, and driving a memory bank to evolve into optimal structured prior; and finally, performing structured reasoning on the to-be-detected sequence by using the optimal memory bank. According to the method, structured priori is constructed by mining a spatio-temporal topology mode of a human body action sequence, and hierarchical memory evolution and double-flow depth interaction are combined, so that the problem of depth ambiguity in a monocular vision task is effectively solved, geometric structure distortion is corrected, and the accuracy of action posture estimation is remarkably improved.

The invention provides a worst execution time path derivation method for an ARINC653 operating system. Only tasks to be tested are reserved in a partition to be tested; from the entrance of the to-be-tested task to top down, performing pile pitching on each branch possibly influencing the program trend in the to-be-tested task; running the to-be-tested task under the target machine, injecting a test case set into the to-be-tested task, and collecting time information of each branch in all pile points; subtracting the execution time of the pile point from the execution time of the to-be-tested task; and obtaining a theoretical worst path and an actual worst path of the to-be-tested project, fitting the bifurcation points of the two paths to obtain a fitting path, running the fitting path under the target machine again, and obtaining a final WCET path. The method solves the problems that a static analysis method is difficult to meet the time margin resource requirement of embedded software operation, a dynamic measurement method is difficult to prove that the obtained result is a real WCET value, and the algorithm threshold is too high.

The invention provides an industrial robot welding process-oriented knowledge graph construction method and a related system thereof, and belongs to the technical field of knowledge graphs. Specifically, industrial robot welding process data are collected, and an industrial robot welding process data database is established; constructing a welding knowledge graph mode layer from top to bottom based on a welding process data database; constructing a knowledge graph data layer from bottom to top based on a welding knowledge graph mode layer, and forming an entity-relationship-entity triple data form by adopting a knowledge extraction algorithm; aiming at an entity-relation-entity triple data form, performing data alignment by adopting a similarity algorithm, and merging equivalent entities and relations; and storing all the triple data, and importing the triple data into a Neo4j graph database to realize visualization of the knowledge graph. Reliability and efficiency of the industrial robot welding process can be improved, and agility and intelligence of the industrial robot are promoted.

The invention relates to the technical field of electronic transaction processing, in particular to an air material transaction platform intelligent matching recommendation method and system, and the method comprises the following steps: obtaining an inventory record and manual data, building an interchange edge and a modification edge, constructing an air material configuration state topology network, obtaining a purchase request, and positioning a target state node, and performing breadth-first search on a reverse path, performing accumulative weighting to generate candidate alternative paths, obtaining an access rule to verify airworthiness certificates and life values, retaining compliant alternative paths, and performing ascending arrangement according to the total conversion cost to generate a recommendation list. According to the method, the directed topology network containing the configuration state, the interchange relation and the refitting cost weight is constructed, the breadth-first search is utilized to automatically calculate the composite alternative path, the airworthiness certificate and the life rule are deeply verified, the hidden available inventory is effectively mined, the compliance scheme is generated, sorting is carried out according to the total conversion cost, and the conversion efficiency is improved. And the inventory utilization rate and the transaction matching accuracy are maximized while the compliance risk is eliminated.

The embodiments of the present application provide a corpus matching method, apparatus, device and storage medium, the method comprising: obtaining a corpus to be matched and a required matching type of the corpus to be matched; obtaining a corpus that matches the required matching type according to the required matching type of the corpus to be matched; each sentence pair contained in the corpus has multiple attribute settings; and obtaining a target corpus that matches the corpus to be matched according to the multiple attribute settings of each sentence pair contained in the corpus. Based on the multiple attribute settings for each sentence pair in the corpus, while achieving corpus matching, the corpus in the corpus can also be divided into fine-grained parts, and the accuracy of corpus reuse can be maximized based on the minimum granularity corpus matching. At the same time, the corpus reuse can be maximized based on multi-dimensional reuse, which is convenient for improving the efficiency of corpus matching.

The application discloses a federal learning model compression method, a user terminal, a server and a system, and belongs to the technical field of communication. The method comprises the following steps: acquiring a local model corresponding to at least one target application; determining a compression strategy of the local model based on a performance index of the target application; performing compression processing on the local model based on the compression strategy to obtain a local compressed model and a compression identifier corresponding to the local compressed model; uploading the local compressed model and the compression identifier to a server for model aggregation processing, acquiring global model parameters issued by the server; and updating the local model based on the global model parameters. In this way, according to the performance index of the application, a suitable compression strategy is selected for model compression, the communication cost can be reduced on the basis that the federal learning model is deployed to multiple devices, the accuracy of the federal learning model parameters is maximized, and the experience of users is significantly improved.

The invention relates to the technical field of shielding optimization, in particular to a neutron moderation and collimation shielding optimization method based on BNCT, and the method comprises the following steps: calculating the thickness required by a shielding wall based on the construction information of a cyclotron installation room in combination with the radiation shielding effect of a shielding wall material, and obtaining the construction thickness analysis result of the shielding wall. According to the method, the construction accuracy and the shielding effect maximization are ensured by considering the construction information of the cyclotron, preselecting the shielding material and calculating the thickness of the shielding wall, the selection of the radiation blocking material is optimized by matching the pore channel requirements and the material characteristics, and the material consumption is evaluated, so that the radiation blocking effect is improved. Effective utilization of resources and optimization of project cost are ensured, and the most economical and effective shielding scheme can be selected for medical facilities by introducing cost-benefit analysis and combining material cost, construction cost, follow-up maintenance cost and service life evaluation.

A building automation system (306) for controlling conditions of a room (310) is described. The building automation system (306) includes room devices (328-332), a first interface (322), a second interface (324), and a management device (e.g., 404-408). The first interface (322) receives a voice command based on a voice utterance detected by a voice-enabled system (302, 304) in the room (310). The second interface (324) receives a hospitality user profile from a hospitality information system (308). The hospitality user profile identifies one or more user parameters associated with the room (310). The management device of the building automation system (306) includes a guest room profile that identifies one or more room parameters associated with the room (310). The management device controls the room devices (328-332) based on the voice command, the hospitality user profile, and the guest room profile.