105 results about "Hybrid approach" patented technology

The present disclosure generally relates to a two-stage system for detecting malicious nodes in Wireless Sensor Networks (WSNs), enhancing network security and resilience. The system employs a distributed approach, leveraging Cluster Heads (CHs) and a central server for efficient and accurate detection. Initially, sensor nodes are monitored for comprehensive node and network metrics, statistically ranked by significance in identifying malicious behavior. CHs perform a resource-aware first-stage detection based on their resource weight, filtering potential threats locally. Results are then aggregated at a server for a second-stage analysis using a hybrid Machine Learning (ML) and Deep Learning (DL) approach. This advanced analysis, combined with statistically relevant metrics, significantly improves detection accuracy. By integrating resource-conscious CH operation with powerful server-side ML / DL, this system offers a scalable, energy-efficient, and highly effective solution for securing WSNs against malicious node attacks, surpassing traditional detection methods in both speed and precision.

A system and method for automatically extracting data from tables and graphs in scientific literature, particularly in life sciences and healthcare, is presented. The invention employs a hybrid approach combining computer vision, natural language processing (NLP), and a large language model (LLM)-based data extraction module. A neural network enhances accuracy by providing contextual information. The system performs table structure detection, optical character recognition (OCR), Vision Transformers (ViTs) for text recognition, and graph-to- table conversion. An LLM refines the extracted data using advanced prompt engineering. A user interface enables data review, validation, and iterative refinement. The system employs a novel two-stage approach: de-rendering tables and graphs into a machine-readable format, followed by interpretation and user-defined mapping. A knowledge graph enhances extraction by resolving ambiguities and inferring relationships. Designed for scalability and continuous improvement, the invention significantly enhances data extraction efficiency and accuracy, accelerating research and knowledge discovery.

The invention discloses a motion track construction method based on a visual large model, and relates to the technical field of computer vision, and the method comprises the steps: employing a target tracking algorithm based on Kalman filtering, and generating a smooth time sequence synchronization track from an original laser radar point cloud and an original video stream; dividing the trajectory into a series of candidate kinematic segments which are complete in kinematics and semantics by applying a hybrid method combining a minimum description length principle and visual language model semantic verification; performing preliminary semantic annotation on the segments by utilizing a visual language model to generate an initial annotation track; performing logic consistency refining on the initial labeling track until the initial labeling track is converged into a refined labeling track; and formatting the refined trajectory into a standard structured semantic trajectory representation character string. According to the method, a semantic gap between low-dimensional physical observation and high-dimensional driving intention is bridged, and ideal input is provided for understanding and prediction tasks of downstream complex scenes.

A computing platform provides code vulnerability detection and evaluation. The computing platform may use a quantum graph categorizer along with quantum polymorphism execution channels to reduce false positives and provide optimized real-time vulnerability evaluation of code and code segments. The computing platform may use a hybrid approach comprising a mix of static and dynamic vulnerability validation and real-time parallel technique execution.

An optimal route can be determined for delivering an item through a logistics system where vehicles have multiple stops when traveling along routes. The optimal route is determined using a hybrid system employing a classical computing device and a quantum annealer. The classical device reduces the search space that allows the quantum annealer to determine a more optimal solution. In an aspect, a routing graph comprising nodes and edges is populated from routes of vehicles, where the nodes identify locations and the edges represent routing data. At least a portion of the nodes and edges is removed to form a refined routing graph. For an origin-destination input, the refined routing graph can be filtered according to a first set of routing constraints to form a reduced route search space. A quantum annealer is invoked according to an objective and a different second set of routing constraints.

A method and an apparatus for signaling high-level syntax of a hybrid INR approach. One or more syntax elements providing for reconstructing at least one part of an image or a 3D object using at least one hybrid Implicit Neural Representation network are obtained. The at least one hybrid Implicit Neural Representation network comprises at least one mapping transformation that maps input coordinates of the image or 3D object to a group of features and an INR network that takes as input the group of features and outputs values of the image or 3D object at the input coordinates. The obtained one or more syntax elements indicate parameters relating to the at least one mapping transformation and are signaled in a bitstream.

A hybrid approach is employed to determine an optimal packing arrangement of cargo blocks within containers loaded onto a vehicle. Cargo block data is accessed, where the cargo blocks are to be arranged into containers for transport by the vehicle having a payload area. Each cargo block is assigned to the containers subject to constraints on the cargo and the containers. A quantum annealer is invoked to individually solve an optimization problem for subsections of the payload area, where the quantum annealer determines an optimal packing arrangement of cargo blocks within the containers for each subsection of the payload area.

In various examples, parking area detection for autonomous and / or semi-autonomous systems and applications is described herein. Systems and methods described herein may use a hybrid method to more accurately determine geometries of the parking areas within environments. For instance, sensor data (e.g., image data, etc.) may be processed using one or more edge detection techniques to determine edge features (e.g., two-dimensional pixels, three-dimensional points, etc.) associated with an environment that includes a parking area. A predicted geometry associated with the parking area, as determined using one or more machine learning models, may then be used to filter the edge features in order to identify a portion of the edge features that represent the parking area. One or more optimization techniques may then be used to determine the final geometry associated with the parking area based at least on the filtered edge features.

A high-position drilling and coal mining face gas extraction-treatment dynamic collaborative prediction method belongs to the technical field of coal mine gas extraction and treatment, and comprises the following steps: firstly, screening out physical parameters influencing a goaf gas extraction effect and an upper corner gas treatment effect according to a porous medium multi-field coupling theory and a field actual working condition; corresponding parameters are collected in real time to form a time sequence data set; and secondly, multivariable input and multivariable collaborative prediction are carried out by adopting a mixed LSTM-Transform method, and an online learning module is adopted, so that the dynamic capture capability of a complex multivariable time sequence mode in an underground coal mine is remarkably improved, and extraction-treatment dynamic collaborative prediction is carried out on high-position drilling and coal mining surface gas. According to the method, the gas extraction effect and the accuracy, stability and real-time performance of upper corner gas result prediction can be effectively improved, meanwhile, optimization of gas extraction pressure parameters can be achieved, and reference is provided for the high-position drilling gas extraction technology.

An electronic copy of an existing invoice is used to extract text therefrom. The extracted text is processed using a hybrid approach-combining fuzzy matching and natural language processing-to map portions of the extracted text to a computer application's specific standard fields. The initial stage is fuzzy matching to map portions of the extracted text to a dictionary of standard fields. For the unmapped portions of the text, natural language processing models such as a fine-tuned DistilBERT model is invoked to determine a second stage mappings. Mappings from the two stages are combined and duplicates are removed to generate a final mapping. The final mapping and the geometrical information of the extracted text is used to generate an electronic template of the invoice. The electronic template can be used to generate future invoices with the same / similar formatting or look and feel of the existing invoice.

There is provided a computer-implemented method for automated gap analysis that performs the steps of: generating a job competencies dataset, generating a course competencies dataset; generating a missing competencies dataset based on the job competencies dataset and the course competencies dataset, and outputting a recommended course dataset. The method identifies job competencies missing in course competencies, and recommends courses in which to include the missing competencies. The approach to competency identification comprises either a hybrid approach composed of rule-based matching and / or similarity matching, or a pre-trained large language model (LLM).

The invention discloses a wind power prediction method and system based on deep fusion of physical knowledge, a medium and a terminal, relates to the technical field of wind power generation, and mainly aims to solve the problem of low prediction result accuracy caused by low reliability of a prediction model due to a shallow fusion level in an existing hybrid method. Comprising the steps of obtaining current operation data and current environment data of a target wind power plant; determining current wind field physical data based on the current operation data and the current environment data; performing deep fusion processing on the current operation data, the current environment data and the current wind field physical data to obtain a current fusion feature vector; current wind field state parameters are calculated according to the current wind field physical data, wind power value prediction operation is carried out according to the current fusion feature vector and the current wind field state parameters, and a current predicted wind power value is obtained; and performing operation state monitoring operation and power grid dispatching operation by using the current predicted wind power value, and formulating an operation maintenance strategy.

Methods, systems, and computer-readable media are used in graph-based machine learning of analog integrated circuits (ICs). In a first aspect, functional device pairs in transistor-level circuits are detected and classified using a hybrid approach combining a subgraph isomorphism algorithm, such as VF2, with a trained relational-graph convolutional network (RGCN). The VF2 algorithm identifies candidate pairs and initial categories, while the RGCN filters false positives using link prediction scores, preserving category labels for validated pairs. In a second aspect, a relational GraphSAGE model performs multi-class link prediction on a heterogeneous graph with netlist and functional relation edge types, labeling device pairs into analog primitive categories without technology-dependent features and merging overlapping pairs into larger functional groups. In a third aspect, circuit performance is predicted using a hierarchy-aware graph neural network comprising an edge-conditioned convolution (ECC) layer and multiple Circuit graph isomorphism network layers corresponding to hierarchy levels of device groupings.

As discussed earlier, labelling techniques that are available for labelling of unlabelled tabular data use some semi supervised models for identification purposes. However, they require sample labeled data for training purposes. Further, the same labelling model / technique cannot be used for all data types. Present disclosure provides method and system for identifying labels of unlabeled column data. The system uses a hybrid approach i.e., it uses language models, regular expressions and known dictionaries for labelling of unlabelled tabular data. For performing labelling, system first classifies received unlabelled tabular data into one or more data buckets. The system then uses appropriate techniques, based on data types, for identification of labels of unlabeled data present in data buckets. Thereafter, system uses feedback mechanism which will impart maturity to system over time. Finally, once system is matured, system can identify labels for all types of data.

A computer-implemented system for predicting financial behavior and adaptive budget optimization based on machine learning, consisting of: a multitude of distributed processing nodes to enable low-latency communication between the nodes; a transaction data ingestion processor configured to establish authenticated connections with a plurality of financial data sources, wherein the ingestion module is further configured to normalize received transaction records into a standardized schema comprising at least a merchant identifier, a transaction category, a timestamp, a transaction amount, and optional geolocation metadata; a preprocessing engine comprising a classification sub-module trained through supervised learning to assign transaction categories based on merchant identifiers and context attributes, and a feature extraction sub-module configured to compute temporal, statistical, and behavioral feature vectors from the normalized transaction data; a prediction control unit comprising a plurality of lightweight neural network architectures, including at least one recurrent neural network (RNN) and at least one attention-based temporal model, the prediction control unit configured to predict short-term and medium-term output trends by sequentially processing the feature vectors; a budget optimization computation unit configured to solve multi-constraint budget allocation problems using a hybrid approach comprising a primary linear programming solver and an additional heuristic optimization technique, wherein the budget optimization computation unit is further configured to dynamically adjust budget allocations based on updated forecasts and user-defined constraints; a security subsystem configured for encryption at rest and in transit, as well as secure key storage in a hardware-based Trusted Platform Module (TPM); and a user interaction interface configured to display budget recommendations and forecasted spending trends through at least one web application, mobile application, or hardware device interface.

A graph database system stores and analyzes connections and relationships between entities using a database structure that represents users, messages, threads, and attachments as nodes with interconnecting edges. The system implements a hybrid approach to electronic communication classification combining self-hosted language models with computer server clustering for pattern matching and suggestion generation. Additionally, the system includes mechanisms for creating and suggesting reusable electronic communication components based on semantic analysis of communication patterns. The system architecture enables relationship analysis, digital content management, and privacy-preserving electronic communication classification.

A method for natural language to SQL (NL2SQL) statements driven by a hybrid approach combining templates and a large language model includes the following steps: constructing a pre-built SQL template library to assist the large language model in quickly matching syntactically correct SQL template skeletons; constructing a database schema information knowledge base to assist the large language model in accurately extracting entities; constructing a large-model natural language parsing and entity extraction engine to extract entity information; constructing a large-model-based template matching engine to match corresponding target SQL templates from the template library using the large language model; and constructing a large-model-based template filling engine to fill the extracted entity information into the corresponding placeholders of the selected target SQL template, generating the final executable SQL statement. By using pre-built SQL templates to ensure syntactic correctness and utilizing the large language model to parse natural language and extract entities to ensure semantic flexibility, this method solves the syntax errors and illusion problems that may occur when generating SQL using pure LLM, significantly improving the practicality and accuracy of NL2SQL technology.

The invention provides a behavior tree dynamic construction method, system and device and a medium, and the method comprises the steps: obtaining original data from a voice sensor, an image sensor and a touch sensor, carrying out the normalization and event abstract semantic processing of the original data, and obtaining a semantic event object; analyzing a corresponding relation type according to the event object, and constructing an event semantic relation graph which is constructed by adopting a predefined rule and large model mixed mode; defining a mapping rule from an event semantic relation graph to behavior tree nodes, and converting a relation type in the semantic relation graph into a behavior tree node type, including setting sequence nodes corresponding to a causal relation and setting condition nodes corresponding to a condition relation; analyzing the event semantic relation graph into a behavior tree structure according to a mapping rule, and constructing a behavior tree; and recording an execution result in the operation process of the behavior tree, and adjusting the event confidence coefficient according to the execution result to optimize the dynamic construction of the behavior tree in the next round. According to the method, the expandable behavior tree meeting the personalized expression of the accompanying robot can be created, so that the accompanying experience of the robot is improved.

The invention discloses a power distribution network voltage control method and a related device, and the method comprises the steps: carrying out the precise sensing of a multi-scale voltage situation in a space-time and physical mixed manner, namely predicting the future node voltage and a corresponding multi-scale node voltage out-of-limit risk index, obtaining a basic control strategy through a main channel, and carrying out the prediction of a multi-scale node voltage. A corresponding robust compensation strategy is obtained by adopting an anti-interference channel, the high-dimensional problem solving efficiency is improved in combination with a Red-beak-Baijie optimization algorithm, all distributed energy sources are equivalent to a quantized virtual impedance network in dual-channel MPC optimization and power distribution network node voltage control, the dimension optimization bottleneck is broken through, and the power distribution network node voltage control efficiency is improved. Finally, a global prediction-hierarchical decision-edge execution full-chain control system can be formed, and the method is suitable for voltage stability control in a high-proportion distributed energy access scene.

The invention discloses a sound field model construction method based on a mixture of a parabolic equation theory and a finite element method, and belongs to the technical field of sound field propagation. According to the method, a sound field is divided into a far-field region and a near-field region, a parabolic equation is established in the far-field region, a PINN model is constructed, an initial sound field result calculated by the parabolic equation is corrected and optimized by using the PINN model in each step of calculation by the parabolic equation, and a final sound field result is obtained. The method comprises the following steps: performing grid division in a near-field region, establishing a wave equation in each grid unit, constructing a unit stiffness matrix and a unit mass matrix, finally assembling a global stiffness matrix and a global mass matrix, and further solving detailed information of a three-dimensional sound field; and sound field information at the junction is processed through a multi-physics field coupling boundary condition at the junction. According to the method, a proper calculation method can be selected according to the physical characteristics of different areas, and the propagation rules of sound waves in different scales and environments can be better reflected.

In one implementation, a shifted-window transformer with sorted grouping is used for point cloud compression. The points are first sorted according to a specified order and then the sorted points are grouped into windows with an equal number of points. All resulting windows are operated upon by self-attention to obtain initial attention features, followed by shifting the windows and another self-attention. We utilize the proposed transformer for lossless compression of point clouds via the octree representation and lossy compression via feature coding. Downsampling can be used to obtain features at different resolutions, which can be combined in a multi-scale solution. In addition, in a hybrid approach, the features can be divided into two parts: one for the window attention and the other for convolutions. The same sorting strategy can be used throughout the proposed architecture, or the network can switch between different sorting strategies every few attention layers.

In one implementation, a shifted-window transformer with sorted grouping is used for point cloud compression. The points are first sorted according to a specified order and then the sorted points are grouped into windows with an equal number of points. All resulting windows are operated upon by self-attention to obtain initial attention features, followed by shifting the windows and another self-attention. We utilize the proposed transformer for lossless compression of point clouds via the octree representation and lossy compression via feature coding. Downsampling can be used to obtain features at different resolutions, which can be combined in a multi-scale solution. In addition, in a hybrid approach, the features can be divided into two parts: one for the window attention and the other for convolutions. The same sorting strategy can be used throughout the proposed architecture, or the network can switch between different sorting strategies every few attention layers.

This invention introduces an advanced framework for channel estimation in IRS-enabled wireless communication systems, leveraging a novel concept of "angular domain suppression" in the angular domain By strategically segmenting large antenna arrays and classifying scattering clusters into ordinary and controlled types, the method allows efficient decomposition of the channel into Type-S, Type-L, and Type-M channels. Passive beamforming at the IRS and active beamforming at the base station enable precise manipulation of signal reflections to control support sets. The process incorporates compressive sensing, compressive sensing, MMSE, LS, or machine learning techniques and metric-driven estimation, minimizing pilot overhead and computational complexity. The framework ensures resource-efficient estimation by terminating when predefined performance metrics are achieved. Furthermore, the system optimizes channel modeling through a hybrid approach, integrating ordinary scatterers and IRS elements for improved signal accuracy. This innovation provides a robust solution for enhancing communication reliability and efficiency in modern IRS-enabled networks.

In various examples, parking area detection for autonomous and / or semi-autonomous systems and applications is described herein. The systems and methods described herein may use a hybrid approach to more accurately determine the geometry of a parking area within an environment. For example, sensor data (e.g., image data, etc.) may be processed using one or more edge detection techniques to determine edge features (e.g., two-dimensional pixels, three-dimensional points, etc.) associated with an environment including a parking area. The edge features may then be filtered using predicted geometries associated with the parking area, as determined using one or more machine learning models, to identify a portion of the edge features representative of the parking area. One or more optimization techniques may then be used to determine a final geometry associated with the parking area based at least on the filtered edge features.

Hybrid classical-quantum systems for improved vehicle routing use a combination of genetic algorithms, simulated annealing, and quantum annealing. The hybrid method allows further exploration of the solution space of a capacitated vehicle routing problem (CVRP), thus identifying improved routing results. Nodes (i.e., locations) are initially clustered, and the node clusters are processed via a genetic algorithm to determine potential solutions. The potential solutions are evolved using simulated annealing to generate evolved solutions, each having evolved node clusters. An evolved solution is selected. Each of the evolved node clusters for the selected evolved solution is separately annealed by a quantum annealer to determine the optimal route through the evolved node clusters.

The present invention discloses a method for positioning a uniformly moving target under conditions where transmitters are asynchronous and positions are unknown. Aiming at multi-base positioning scenarios, the method utilizes time delay information at consecutive moments to replace a traditional hybrid method of time delay and Doppler information, thereby reducing system hardware requirements. Aiming at a distance measurement model, the method utilizes numerical characteristics between measurement values ​​to eliminate some coupling variables, then separates unknown variables, and transforms the positioning problem into a constrained weighted least squares problem. By performing a first-order Taylor expansion on the objective function of the problem, a constrained weighted least squares problem with bias reduction capability is obtained. The problem is then transformed into a convex semi-definite programming problem. Finally, optimal estimates of the target's initial coordinates and moving speed are obtained. The method has the advantages of utilizing distance measurement values ​​observed at consecutive time points to estimate the target's position and speed when transmitters are asynchronous and positions are unknown, and has low equipment cost, low computational complexity, and a significant bias reduction capability.

A hybrid approach is employed to determine an optimal packing arrangement of cargo blocks within containers loaded onto a vehicle. Cargo block data is accessed, where the cargo blocks are to be arranged into containers for transport by the vehicle having a payload area. Each cargo block is assigned to the containers subject to constraints on the cargo and the containers. A quantum annealer is invoked to individually solve an optimization problem for subsections of the payload area, where the quantum annealer determines an optimal packing arrangement of cargo blocks within the containers for each subsection of the payload area.

This invention provides a multi-task learning resource optimization method based on the lottery hypothesis, belonging to the field of multi-task learning and neural network optimization technology. This method employs a sparse expert hybrid approach based on the lottery hypothesis to effectively prune neural network model parameters, thereby significantly reducing the computational cost and storage overhead of the neural network model. The sparse expert network structure retains high sensitivity to key task information. The introduction of a softmax-based router and mask matrix allows each task to select and activate suitable expert network components based on the features of its input data. Cross-training and iterative amplitude pruning optimize the omnipotent expert matrix, further enhancing the model's generalization ability in handling complex tasks and different data distributions. Task-specific layers independently process and optimize the features of each task, reducing mutual interference between tasks and improving the accuracy of feature extraction and the efficiency of task processing.

A hybrid method for green intelligent manufacturing (GiM) combines the carbon reduction and the energy saving into the intelligent manufacturing based Industry 4.1 cloud platform. GiM assists companies to achieve the goal of net zero transition and help them advance to Industry 4.2 as soon as possible by simultaneously taking carbon footprint and energy issues into account. GiM collects large volumes of essential data (including carbon footprint) via cyber physical agents (CPAs), and sends them to two critical services of carbon management and intelligent energy management system (iEMS) deployed on the cloud platform. The two critical services optimize the energy dispatch schedule by strictly following the requirements of energy saving, carbon reduction, and net zero. Then, the state of zero defects of intelligent manufacturing achieved in Industry 4.1 can be upgraded to net zero of GiM in Industry 4.2.

A method and an apparatus for signaling high-level syntax of a hybrid INR approach. One or more syntax elements providing for reconstructing at least one part of an image or a 3D object using at least one hybrid Implicit Neural Representation network are obtained. The at least one hybrid Implicit Neural Representation network comprises at least one mapping transformation that maps input coordinates of the image or 3D object to a group of features and an INR network that takes as input the group of features and outputs values of the image or 3D object at the input coordinates. The obtained one or more syntax elements indicate parameters relating to the at least one mapping transformation and are signaled in a bitstream.