20 results about "Parallel search" patented technology

This invention discloses a GNSS integer ambiguity fixing method integrating intelligent optimization algorithms and VIB strategies, belonging to the field of satellite navigation and positioning technology. This method combines the advantages of VIB strategies and swarm intelligence fusion optimization algorithms. By directionally grouping the ambiguity floating-point solution covariance structure, the ambiguities are guided into groups. The PSOAF and AWDE intelligent optimization algorithms are used to perform parallel searches on the grouped ambiguities, improving search efficiency and positioning accuracy. Simultaneously, the probability assessment and fixing judgment of the integer solutions of each sub-block ambiguity and the overall ambiguity ensure the reliability of the ambiguity solutions, making it suitable for complex and dynamic GNSS positioning scenarios.

The application discloses a GPU-based multi-modal index integration method. The method comprises the following steps: acquiring multi-modal original data and extracting a feature vector; mapping the feature vector to a unified common semantic space; constructing a double-layer hybrid index structure on a GPU, wherein a first layer adopts IVF-PQ index for coarse screening, and a second layer adopts graph index for fine screening; after receiving a query request, performing two-stage parallel search on the GPU, first screening a candidate set through IVF-PQ index, and then fine screening a final result through graph index. The application solves the cross-modal semantic alignment problem through the unified semantic space, realizes high-precision and low-delay retrieval of massive multi-modal data by using GPU parallel computing and a double-layer index structure, and effectively overcomes the technical bottleneck that precision and efficiency are difficult to be considered in the prior art.

This invention provides a method, apparatus, and vehicle for constructing a KD-tree from a vehicle-mounted LiDAR point cloud. The method includes: determining the first median of the point cloud data; performing KD-tree construction steps based on the left and right interval data until the unprocessed data variable is 0, and outputting the KD-tree. Thus, by obtaining the median using a quicksort algorithm, a full sort of the point cloud data is not required. Instead, the data is partitioned into two parts: less than a benchmark value and greater than a benchmark value. The median is then recursively searched, improving the efficiency of finding the first median. Since the first median divides the point cloud data into independent left and right interval data, multi-threaded parallel searching of left and right subtree nodes can be used during KD-tree construction, thereby improving the efficiency of KD-tree construction for the vehicle-mounted LiDAR point cloud and ensuring the response speed and operational sensitivity of the autonomous driving system.

The application provides a full-size subsonic wind tunnel loop aerodynamic profile multi-objective optimization method based on steady CFD, and belongs to the technical field of wind tunnels.The application reduces the initial geometric parameters to a sensitive parameter subspace through principal component analysis, generates initial sample points by using Latin hypercube sampling, extracts response values by executing steady Reynolds average Navier-Stokes equation solving, constructs a Kriging surrogate model, performs multi-start parallel search based on an expected improvement function, extracts a separation zone length and a backflow zone height to input an unsteady correction factor model to calculate a correction coefficient for correcting a diffusion section pressure recovery coefficient, searches a Pareto front solution set by using a non-dominated sorting genetic algorithm, selects an optimal parameter combination through a comprehensive performance index, and reconstructs the wind tunnel loop aerodynamic profile through a hierarchical parameterization method and performs verification calculation, so that the problem that the steady calculation framework cannot accurately predict the separation flow and the optimization result deviates from the true performance is solved.

The application discloses a fusion system of automatic activation value checkpoint search and automatic tensor parallel search, relates to the technical field of deep learning, and comprises network linearization, meta information extraction, an automatic activation value checkpoint solver, an automatic tensor parallel solver, and a two-stage solver. The meta information extraction is used to provide required memory overhead and operation overhead information for the automatic tensor parallel solver, the automatic tensor parallel solver solves according to the given memory budget and the meta information, the two-stage solver linearizes a series of tensor parallel strategies to a network, and the linearized network is transmitted to the automatic activation value checkpoint solver. The automatic activation value checkpoint solver controls the memory overhead below the real memory budget. The strategies of the automatic tensor parallel and the automatic activation value checkpoint are combined, the memory wall problem encountered in deep learning training is further slowed down, and meanwhile, the two-stage solver can guarantee that the memory problem is solved while the performance of model training is considered.

This invention relates to the field of multimedia content retrieval, specifically a topology-aware hierarchical index construction and load balancing method for multimedia content retrieval. It comprises two core processes: index construction and querying. Index construction includes three stages: topology graph construction, METIS-based multi-level graph partitioning, and parallel HNSW sub-index construction. These steps are executed sequentially and are closely interconnected. Querying includes three stages: feature modulation based on a gating mechanism, cluster association probability inference, and probability-driven parallel search and fusion. Furthermore, a nearest-neighbor-aware soft-label supervised training strategy is proposed to train the gating neural network model in the query stage. This invention abandons the traditional partitioning paradigm that relies solely on geometric metrics. Through an innovative index construction process and query mechanism, with topology awareness and load balancing at its core, it achieves high recall, low latency, and high scalability in multimedia similarity search.

This invention relates to the field of network communication technology, specifically to a flow control optimization algorithm for an FPGA-based intelligent conferencing system. It constructs a central control unit (CCU) with the FPGA as its core, and uses a cascaded CAM to segment the 48-bit MAC address into six 8-bit sub-keys for parallel searching, outputting the matching result in one clock cycle. A system cycle of 1-5ms is set, and data is transmitted in time slots according to the order of SOC frames, real-time data frames, SOA frames, non-real-time data frames, and TAIL frames, isolating real-time and non-real-time data. A dynamically maintained MAC address table based on an aging mechanism is established to achieve adaptive topology changes for slave devices. Dynamic priority polling arbitration is used to fairly allocate shared storage resources across multiple PHY interfaces. This invention achieves zero-conflict, high-bandwidth utilization network data transmission at a rate of 1000Mbit / s, significantly reducing FPGA resource consumption, improving the real-time performance and reliability of the conferencing system, and is suitable for multi-slave distributed intelligent conferencing scenarios.

The present application relates to a kind of multifunctional storage computing unit based on ferroelectric transistor, array and its fault-tolerant method.The unit is composed of two N-type FeFET and two NMOS transistor, by configuring peripheral control signal / circuit can realize three kinds of working mode: memory mode supports the non-volatile storage and read-write operation of data;Memory logic computing mode supports the Boolean logic operation to the storage word;Three-state content addressable memory mode supports high-speed parallel search and "irrelevant" bit matching operation.Further, the present application proposes a kind of anti-radiation reinforced three-state content addressable memory unit.The unit is based on the original multifunctional unit, access single-cycle storage module DICE and four storage transistors, realizes single node flip recovery.The present application integrates storage, computing and search function in single compact unit structure, improves energy efficiency and functional density, and the anti-radiation reinforced unit has good soft error tolerance ability and process robustness.

This invention relates to a method and system for collaborative operation between an unmanned surface vessel (USV) and an unmanned aerial vehicle (UAV), belonging to the field of unmanned system control technology. The method includes: the USV autonomously cruises in a fan-shaped or parallel search mode, detecting targets and calculating their coordinates using side-scan sonar and radar; the USV generates a mission data packet and sends it to the UAV, adjusting its attitude for UAV takeoff and landing; the UAV flies to the target's airspace, confirms the target's attributes and status using an infrared pod or zoom camera, and autonomously drops a lifebuoy and provides acoustic and optical guidance; the USV dynamically optimizes its trajectory based on the transmitted real-time target coordinates and approaches the target, dropping rescue equipment via an electrically operated drop device upon arrival. This invention constructs a closed-loop collaborative process of "detection-confirmation-initial rescue-final disposal," effectively solving the problems of poor adaptability to harsh sea conditions, insufficient collaborative depth, and low reliability of long-range communication in existing technologies.

The application relates to the technical field of robots, and provides a robot path planning method and device, electronic equipment and a storage medium, the method comprising: in the path planning process of a robot, starting multiple parallel path planning threads, and assigning different deterministic sampling sequences to each path planning thread; in each path planning thread, performing path search based on the deterministic sampling sequence; in response to any path planning thread successfully planning an initial path, terminating all path planning threads, and performing backtracking optimization on the initial path to obtain a motion path of the robot. The application effectively overcomes the problems of inconsistent planning results and poor real-time performance of traditional random sampling by using deterministic sampling sequences and multi-thread parallel search; the path quality is improved while the efficiency is ensured through backtracking optimization. Under the premise that the environment does not change, the consistency of the path planning result is achieved, and the optimality and calculation efficiency of the path planning result are improved.

The present application relates to the field of VPP multi-source collaborative scheduling, and discloses a virtual power plant optimization scheduling method based on information gap decision theory. Historical behavior data of EV, wind power and photovoltaic power generation data are preprocessed; EV travel load data is predicted by using a random sampling algorithm; wind and light output is predicted by using a deep learning time series network; a VPP model with the minimum operation cost as the target is established and solved, serving as a deterministic model of IGDT; a dichotomy method combined with cost margin and multi-start parallel search mechanism is used to find the optimal uncertainty level parameter and the corresponding scheduling scheme under the IGDT risk aversion model; a dichotomy method optimized by introducing chaos mapping and energy level decision mechanism is used to find the optimal uncertainty level parameter and the corresponding scheduling scheme under the IGDT opportunity pursuit model.

Embodiments of the present application provide a task execution method and device, a storage medium and a computer program product. In the embodiments of the present application, the traditional kernel sampling scheme is abandoned, and a new kernel sampling scheme is proposed. In the embodiments of the present application, on the basis of introducing a candidate boundary value, a parallel search method can be used to filter out a part of target confidence degrees that may meet the kernel sampling condition in the second candidate set. In this way, by adjusting the candidate boundary value in a loop, the target confidence degree that can meet the kernel sampling condition can be found, and then the target element can be sampled from the first candidate set. In each round of filtering the target confidence degree, the parallel search method can be used, and therefore, the processing efficiency of the internal part of this link can be effectively improved, and the execution efficiency of the kernel sampling task can be improved.

A semiconductor device (SoC) (1) and computer-implemented method for deterministic artificial intelligence supervision. It incorporates a secure port (8) in an inert state to ensure subordination by injecting a base truth. An AECM module (3) audits the integrity of physical memory using hash trees on the hot path, repairing tensors on the fly with no latency. An AECM module (4) vetoes semantic hallucinations by calculating spatial divergences in hardware and consolidating neuroplasticity in parallel search associative memories (4a) without retraining software. A forensic module (5) extracts the causal triad and cross-references it with a logical topological map, recording the exact node and layer responsible for the failure in immutable storage to provide full explainability.

The application relates to a kind of double-branch flow face forgery detection methods based on double domain, comprising the following steps: constructing face forgery video dataset;The video data in face forgery video dataset is preprocessed, and preprocessed face image is obtained;Based on preprocessed face image, data set is divided, and training set, verification set and test set are respectively constructed;A double-branch network model is constructed, the double-branch network model is trained using the training set, and the optimal model is obtained;Using the optimal model, the frequency domain feature and the space domain feature of the preprocessed face image are extracted, and the frequency domain feature and the space domain feature are interacted, and the face forgery detection result is obtained.The application starts from the space domain and the frequency domain of image, searches the forgery trace in double domain in parallel, and interacts and fuses the frequency domain information and the space domain information, captures the correlation between different modal information.Compared with the prior art, the application has the advantages of high network generalization performance, good detection performance in the environment where video quality is compressed, etc.

This invention proposes an optimization method and system for energy storage systems based on a multi-objective ant colony optimization algorithm, belonging to the field of grid regulation technology for energy storage systems. It aims to efficiently solve the coordinated optimization problem of energy source, grid, load, and storage. The core of this method lies in constructing multiple heterogeneous heuristic information sets with different optimization biases, and dividing the ant colony into different groups. Each group searches for a scheduling scheme in parallel based on a specific heuristic information. This mechanism breaks through the limitations of traditional single search, greatly improves the diversity of global search, and can generate higher-quality and more widely distributed trade-off solution sets, thereby achieving optimal coordination between economic and environmental goals.

This invention discloses a method for generating unique identifiers based on RAM tags. The first step involves receiving commands from higher-level software via a command processing module. The second step involves identifying and processing the command type using the command processing module. An initialization module and a control logic module receive and process the commands issued after parsing and identification by the command processing module. These processing operations include creating a unique identifier, destroying the unique identifier, and initializing the RAM. This invention utilizes the characteristics of the RAM to simultaneously achieve serial and parallel searching, processing multiple tags at once and shortening the delay in generating unique identifiers. During processing, a temporary 24-bit counter is divided into two parts: one part indexes the RAM to obtain multiple tag outputs, and the other part indexes these tag outputs in parallel, enabling simultaneous checking of available positions for multiple tag outputs. Furthermore, through accumulator and scrambling processing, the generated identifiers are randomized, avoiding the reuse of recently destroyed identifiers.