317 results about "Model configuration" patented technology

The present invention discloses a system and method for performing artificial intelligence (AI) safety red-teaming with integrated policy fuzzing and adversarial prompting to systematically identify, characterize, and mitigate unsafe or non-compliant behaviors in AI models. The disclosed invention automates the process of generating, executing, and analyzing adversarial test cases through coordinated functional units comprising a policy fuzzing unit, an adversarial prompting unit, an execution sandbox, a telemetry processing unit, a scoring and triage processor, and a cryptographic provenance processor. The system applies grammar-driven and reinforcement-based fuzzing techniques to vary policy descriptors, model configuration parameters, and instruction hierarchies, while a learned adversarial prompt generator synthesizes contextually coherent adversarial prompts optimized for maximum policy violation likelihood. The generated prompts and policy vectors are executed in an isolated, instrumented sandbox that records input-output interactions, timing characteristics, and intermediate representations.

Methods and apparatus for leveraging transfer learning of one Wireless Transmit / Receive Unit (WTRU) to benefit another WTRU are provided. One method may include the WTRU receiving AI / ML model configuration information indicating one or more AI / ML models available from the network node, a profile associated with the AI / ML models, and a training convergence threshold. Based at least on the profile(s), the WTRU determining that the one or more AI / ML models are not suitable for use by the WTRU, and sending first information indicating that the one or more AI / ML models are not suitable for the WTRU and / or that the WTRU will be training a local AI / ML model. The method may then include training the local AI / ML model according to the convergence threshold, receiving a request to transfer AI / ML model parameters, and sending an indication of the AI / ML model parameters associated with the trained local AI / ML model to the network node.

The invention relates to the technical field of artificial intelligence chips, and discloses a model performance estimation method and device and computer equipment, and the method comprises the steps: determining model configuration data and candidate distributed strategies of a target model; converting the original calculation graph corresponding to the single-card deployment state based on the model configuration data and the strategy configuration data of the candidate distributed strategies to obtain a distributed overhead calculation graph corresponding to the multi-card deployment state; and performing performance estimation on the candidate distributed strategy according to the basic overhead and the additional overhead in the distributed overhead calculation graph to obtain strategy performance data of the target model under the candidate distributed strategy, thereby realizing conversion of a single-card model into a multi-card model. And based on the multi-card model, simulation calculation of the multi-card interconnection mode is realized on the premise of limited hardware resources, so that the influence of a communication operator and a topological structure corresponding to the multi-card interconnection mode in the overall operation of the model is reflected, and the upper limit of the model performance can be accurately evaluated in a simulator verification stage before silicon is applied.

A method and related systems for dynamically adjusting on-device configurations for models based on detected context switches is disclosed. The system can use an on-device language model to determine context parameters in a first mode and then uses the context parameters to determine prediction models or prediction model parameters to use for inputs provided in a second mode.

The invention provides a high-strength aluminum alloy selective laser melting forming thermal stress prediction method based on deep learning. The method comprises the steps of 1, finite element model establishment and simulation, wherein a heat transfer model and a thermal coupling model are established; an SLM process is used as a simulation object, a Gaussian model is adopted to define a laser heat source, and thermal stress distribution under different process parameters and different sample sizes is simulated; wherein the process parameters comprise laser power, scanning speed and hatch spacing; step 2, data processing and deep learning model training; comprising the steps of data preprocessing, neural network model architecture determination, adversarial network part generation and model configuration and training. 3, evaluating and optimizing the model; and 4, thermal stress prediction and process optimization. By learning a complex mode in finite element simulation data through a deep learning model, efficient and accurate thermal stress prediction is achieved, and the problems that traditional finite element analysis is complex in calculation and large in resource consumption are solved.

The invention belongs to the technical field of artificial intelligence engineering, and particularly discloses a model training and deploying method and system based on Jenkins and version control, a terminal and a medium, and the method comprises the steps that a submission event of a GitLab code warehouse is monitored through Jenkins, difference data information including a training script, a data set version and model configuration is extracted in combination with DVC metadata, and the submission event of the GitLab code warehouse is obtained; and a training task is dynamically generated, and computing resources are automatically allocated to execute a training process. Through the process, full-process automation and closed-loop management of model training, verification, deployment and monitoring are achieved, development efficiency, deployment safety and model traceability are improved, and the method is suitable for intelligent model development scenes with high-frequency iteration and sensitive performance.

The invention discloses an enterprise data analysis system based on natural language interaction, and relates to the field of enterprise data analysis, the system comprises seven core function modules: a dialogue management module supports a user to create a new dialogue and automatically jumps to an initial page; the report interaction module realizes query support and switching functions of a real-time report and a T + 1 report through a floating button; the file import module allows uploading of Word, PDF and JPG format files to supplement question and answer contexts; the model configuration module supports dynamic switching of different AI service interfaces; the interaction auxiliary module provides a quick question button and a dialogue interruption function; the report preview module integrates Excel and PDF plug-ins to realize detail data preview; the session sharing module generates a sharing link and sets an access permission rule; dynamic model configuration optimizes response speed and accuracy; the operation process is simplified by a suspension button and a quick questioning function; the session sharing mechanism supports cross-team cooperation; and the real-time collaborative design with the T + 1 report meets the multi-dimensional data analysis requirement.

The invention relates to the technical field of document intelligent review, in particular to a multi-mode digital publishing intelligent review system and method based on a large model, and the system comprises a sample construction module, a multi-mode analysis module, a structure recognition module, a term verification module and an annotation output module. According to the method, a model configuration data set is generated by analyzing a subject-object combination relationship in a text and jointly screening part-of-speech, sentence pattern and semantic structure, the accuracy of semantic recognition and structure judgment is improved, and local features of semantic offset and image-text expression separation are recognized by combining a comparison relationship between an image action path and a text behavior object; a paragraph logic structure and a theme connection mode are analyzed based on semantic offset information, the problems of content dislocation and theme jump between paragraphs are effectively revealed, matching change tracks and part-of-speech continuation fluctuation in term cross-paragraph contexts are recognized, label overlapping and coverage redundancy conditions are analyzed from sentence group hierarchy, and the term cross-paragraph contexts are obtained. And forming a label combination suggestion and constructing an annotation structure record.

The invention discloses an engineering field-oriented AI agent construction and management and control system. The system comprises a data acquisition module, a data preprocessing module, a vector knowledge base module, a large model module, a model configuration strategy module and an agent construction module. The data acquisition module is used for acquiring related data of an engineering project; the data preprocessing module is used for processing original engineering data; the vector knowledge base module is used for performing semantic analysis and feature extraction on the processed data; the large model module is used for deep understanding, semantic analysis and intelligent reasoning of engineering project data; the model configuration strategy module is used for screening out a large model most suitable for a certain task; and the agent construction module is used for constructing an agent. The AI capacity can be managed in a unified mode, model selection and data processing are optimized, repeated construction is effectively avoided, the safety risk is reduced, and the intelligent agent performance and the application effect are improved.

The invention discloses a medical ethical analysis report generation method based on a multi-model game, and the method comprises the steps: analyzing a medical ethical case, and constructing a structured element set and a role model configuration set; implementing a multi-model game, dynamically evolving the demonstration intensity through an excited state attenuation mechanism, and obtaining an evolution demonstration intensity matrix, game track data and a resonance suppression record; analyzing the evolution argumentation intensity matrix and game trajectory data, mining viewpoint fusion and crystallizing structural divergence, and obtaining a fusion viewpoint set, a crystallization divergence set and an ethical tension map; and integrating and generating a medical ethical analysis report. According to the invention, the dynamic authenticity of the medical ethical analysis process and the objectivity of the result are improved.

The invention discloses a differential chlorophyll fluorescence inversion method, system and device and a storage medium, and is applied to the field of vegetation remote sensing and physiological ecological monitoring, and the method comprises the steps: obtaining vegetation state monitoring data of a target region; on the basis of vegetation state monitoring data, asynchronous leaf-changing type evergreen arbors and synchronous deciduous type arbors are identified; carrying out SIF simulation on the asynchronous leaf-changing type evergreen arbors and the synchronous deciduous type arbors by adopting corresponding models respectively; according to the phenological phase function and the stress response function, generating correction factors corresponding to the asynchronous leaf-changing evergreen arbors and the synchronous deciduous arbors; the obtained initial SIF value is optimized through a corresponding correction factor; fusing the optimized SIF values to generate an SIF time sequence product; according to the method, differential model configuration and physical mechanism driven online correction are coupled, so that the problems of model mismatching and physiological mechanism deficiency in heterogeneous vegetation SIF inversion are solved, and an SIF time sequence product which is high in precision and clear in physiological and ecological significance is generated.

According to the large language model reasoning optimization method and device, the storage medium and the computer equipment provided by the invention, the multi-dimensional monitoring data is acquired, and then the multi-dimensional monitoring data is input into the pre-trained environment decision network to obtain the optimal model configuration. The current resource state is evaluated according to the multi-dimensional monitoring data, and the optimal execution path is determined according to the resource state and the optimal model configuration. And finally, adjusting the target model based on the optimal model configuration, and performing model reasoning according to the optimal execution path in the reasoning process of the target model. In the process, the resource state is evaluated and the optimal model configuration is determined by collecting the multi-dimensional detection data, and then the optimal execution path is dynamically determined, so that the model structure and the calculation strategy can be dynamically adjusted according to the state during actual operation, and the response delay is effectively reduced and the resource utilization efficiency is improved while the reasoning precision is guaranteed.

The invention provides a multi-dimensional and dynamic digital assessment system and assessment method. Comprising a support module (comprising an index knowledge base, an algorithm operation base and an evaluation model base), a multi-source index fusion center (for constructing a three-dimensional data model and integrating a multi-dimensional data resource pool), a dynamic rule configuration engine (for realizing flexible configuration through a visual interface, a model adaptation mechanism and double-track verification) and a core business module (for managing assessment indexes, evaluating indexes and evaluating indexes). And evaluating the model and plan). The support module lays a basic resource layer, a fusion center constructs a data base, a configuration engine realizes flexible verification, the business module drives full-process execution, and full-process digital management from index creation, model configuration to examination execution is realized. The problems of iteration lag, multi-scene adaptation redundancy, weak data verification and the like of a traditional system are solved, index legality is guaranteed through static grammar detection and dynamic simulation operation, data traceability is achieved through a temporary database and full-link recording, and the dynamic response capacity, scene compatibility and data reliability of the assessment system are improved.

The invention relates to a production line vehicle type configuration transmission and verification method, and the method comprises the steps: obtaining vehicle type configuration data corresponding to a work-in-process when an intelligent transportation trolley enters an initial station of a production line; the vehicle model configuration data are written into the flexible conveying bracket provided with the work-in-process product; the intelligent transportation trolley is controlled to carry the flexible conveying bracket to enter a plurality of target stations matched with the vehicle type configuration data; when the intelligent transportation trolley arrives at any target station, vehicle type configuration data are read from the flexible transportation bracket; verifying the configuration state of the current station based on the vehicle type configuration data; if verification is not passed, the target station is controlled to execute configuration switching operation, and verification is executed again after configuration switching is completed; and if the verification is passed, controlling the target station to execute the corresponding production operation. By adopting the method and the device, the consistency of the vehicle type configuration data acquired by the station and the actual vehicle type configuration data of the product in process can be improved, and the processing abnormity caused by wrong station configuration is reduced.

The invention relates to a causal structure discovery method based on evolutionary neural architecture search and reinforcement learning, and the method comprises the steps: obtaining to-be-detected observation data, inputting the to-be-detected observation data into a causal structure detection model, and obtaining a directed acyclic graph composed of highest confidence coefficient edges; the causal structure detection model is obtained by training an Actor-Critic model by using a training set; the Actor-Critic model comprises an actor network model and a commentator network model; in the training process, the model configuration of the Actor-Critic model is optimized based on an evolutionary algorithm, and the optimal configuration is obtained; and inputting the observation data in the training set into the actor network model under the optimal configuration to obtain a candidate directed acyclic graph, and evaluating the candidate directed acyclic graph by the commentator network model under the optimal configuration to assist in strategy optimization to obtain a candidate directed acyclic graph. And strengthening the actor network model under the optimization strategy through a reinforcement learning algorithm to tend to output an optimal causal graph structure.

The invention discloses a method for migrating a traditional electric power SCADA system to a new energy system, and relates to the technical field of electric power system migrating, and the method comprises the steps: building an instance object of each equipment model according to the type of electric power equipment and a four-remote configuration point table, and carrying out database model instance configuration and batch implementation through a model configuration tool, forming a model instance set configuration taking a partition or equipment as a unit; generating a power system calculation formula script by using a calculation formula configuration simulation tool, and recording the information relationship of each virtual point through a standardized data format to realize export and import of calculation script configuration; importing and exporting partition pictures by using a graphic configuration tool to realize automatic configuration from an original engineering project to a new project; and according to the equipment type, classifying the matched configuration image templates, and realizing image static loading and template dynamic configuration. The method can change few codes, smoothly upgrade to adapt to various scenes, and reduce the engineering workload.

The invention discloses an asynchronous intelligent control method for a real-time simulation system. The asynchronous intelligent control method is based on a simulation environment construction module, an asynchronous feedback acquisition module and an intelligent control decision module. The simulation environment construction module is used for loading nodes, channels and model configuration files and defining simulation operation parameters; the module supports node starting, stopping and parameter adjusting operation, and an executable interface is provided for subsequent intelligent control; the asynchronous feedback acquisition module is used for periodically acquiring node states, link quality, throughput rate and time delay indexes through a probe module in a simulation running process, and writing the node states, the link quality, the throughput rate and the time delay indexes into a state cache in an asynchronous mode; the intelligent control decision module is used for carrying out analysis and strategy reasoning on the asynchronously acquired state by utilizing a reinforcement learning or deep reinforcement learning algorithm; the intelligent agent generates an action decision according to the current state, and outputs a node parameter adjustment instruction, an interface control instruction, a power configuration instruction or a flow scheduling control instruction to the environment construction module; the decision-making process can be continuously carried out in simulation operation, and dynamic self-adaptive optimization is achieved.

Various aspects of the present disclosure relate to quasi co-location (QCL) indication for AI / ML-based model configuration. An apparatus, such as a UE, receives, from a network entity, an artificial intelligence (AI)-based configuration corresponding to signal transmission and / or signal reception by the UE, where the AI-based configuration is associated with one or more AI models, and an AI model is associated with a dataset that is configured with a set of condition parameters. The UE measures a set of report parameters responsive to a signal received from the network entity, and the set of report parameters are associated with the set of condition parameters of the dataset of the AI model. The UE transmits, to the network entity, one or more feedback parameters based at least in part on the set of report parameters, and the one or more feedback parameters usable by the network entity to select the AI model.

The invention provides an embedded operating environment-oriented AI model cross-platform deployment method and system. The method comprises the following steps: generating a model configuration file conforming to a heterogeneous model configuration information standard format to realize model decoupling; the running state of the heterogeneous processor is collected in real time and abstracted as a reasoning resource node, and static or dynamic deployment is executed according to the priority and the running state for efficient resource scheduling; bottom layer differences are shielded by calling a unified packaged standardized inference engine service interface, and model inference is completed; and starting an AI model file dynamic updating process in application software operation, and dynamically switching the AI model file under the condition that the system does not need to be restarted through legality, content change and check value consistency check. According to the method, the cross-platform portability, the heterogeneous resource utilization efficiency and the system reliability of the AI model application software are remarkably improved.

The invention discloses a supply chain demand prediction system based on federated learning, and belongs to the field of federated learning, and the system comprises the steps: constructing a local data processing module in each node based on a federated learning method, carrying out the preprocessing of original data through employing an encryption technology, and carrying out the operation of a local model training process on each node according to a data feature set, analyzing the feature set through a preset training parameter to obtain a parameter update value of the local model; collecting a parameter update value from each node through a global coordination mechanism, and fusing parameters of multiple nodes by adopting an aggregation algorithm to obtain a preliminary update result of a global model; according to the corrected global parameter set, an updating instruction is distributed to each node, the parameter set is adapted through a local module, and local model configuration of each node is obtained; and for local model configuration, executing a prediction task at each node, and processing real-time input of market environment data to obtain a local demand prediction result of each node.

Disclosed are systems and methods for building and using a machine-learning model to facilitate intelligent selection of treatment strategies for subjects suffering from hemophilia. Sensor data (e.g., that characterizes movement of or a physiological characteristic of particular a subject) can be used as an input to a machine-learning model workflow. The sensor data may affect a model selection, model configuration, model result, pre-processing and / or post-processing. A result of the workflow may inform or influence a treatment selection, treatment schedule, treatment dosage and / or activity recommendation for the particular subject.

The invention discloses a direct-current cable terminal structure optimization method and device based on a double-layer integrated stacking optimization proxy model and a medium. The method comprises the following steps: (1) carrying out trapezoid-chamfer parametric modeling on an outer side curve of an epoxy sleeve, and constructing a training sample by Latin hypercube sampling; (2) constructing a'base learner-meta learner 'double-layer integrated stacking agent model, and synchronously optimizing model configuration and hyper-parameters by using an improved sparrow search algorithm; (3) driving ISSA (International Standard Standard Architecture) to quickly optimize by using the trained proxy model to obtain an optimal geometric curve; and (4) through finite element-stream theory-thermal shock joint verification, it is confirmed that the maximum electric field intensity of the optimized terminal is reduced, the initial discharge voltage is increased, and the mechanical reliability meets the long-term operation requirement. The method solves the problem that a traditional agent model is insufficient in precision and insufficient in insulation margin of a gas-solid interface of a 550kV direct-current cable terminal, and is widely applied to localization design of + / -550kV GIS cable terminals.

The invention discloses a throat mucosa intelligent state evaluation and inflammation early warning management system and method. The method comprises the following steps: synchronously acquiring a throat mucosa image, breath sound / cough sound / sounding signals, airflow data and body temperature or environment relative humidity data, evaluating image definition, shielding degree and coverage rate, and outputting acquisition guide superposition information and triggering re-acquisition when the image does not reach the standard; reliability is calculated for the multi-modal features, weight fusion output state vectors and confidence coefficients are determined in combination with quality gating, and cross-modal consistency verification is executed to trigger re-sampling or strategy adjustment; generating a risk index based on the change of the sliding time window, and outputting an early warning level by adopting hysteretic classification; and writing the threshold, the weight and the sampling and guiding strategy into a parameter library and a fusion model configuration area to form a closed loop. According to the system, improper electroencephalogram fusion and multi-terminal collaborative optimization can be selected, and daily scene evaluation stability and early warning reliability are improved.

Methods and systems for determining a recommended ensemble model configuration is disclosed. Method performed by server system includes access a validation dataset and generating one or more ensemble model configurations. Each ensemble model configuration includes a subset of base models. Operations are performed iteratively for each ensemble model configuration till predefined criteria are met. Operations include determining, by the subset of base models, a set of predictions, computing, one or more prediction losses, computing a pairwise diversity loss metric for the subset of base models, and fine-tuning the subset of base models on backpropagating the one or more prediction losses and the pairwise diversity loss metric. Method includes determining the recommended ensemble model configuration based on each ensemble model configuration including the subset of fine-tuned base models.

The invention discloses an intelligent modeling and checking calculation method and system for a bridge component based on Dynamo, and belongs to the field of bridge engineering information modeling and structural performance visualization analysis, and the method comprises the steps: presetting a component family sketch and a parameter control rule in Revit; a three-dimensional modeling process based on geometric lofting is constructed in the Dynamo; bidirectional synchronization between parameters and geometry is realized by using a Python script; calling a bridge member standard database to carry out model configuration comparison and replacement; structuring the nested family parameters and outputting XML format information; docking a structure analysis platform through a model-checking calculation interface and automatically establishing a finite element model; executing multi-working-condition checking calculation, and returning a structural performance index; a visualization result is embedded in the Dynamo in a thermodynamic diagram mode; closed-loop feedback and real-time updating of structural design and checking calculation are achieved, and a full-process parameter-driven bridge component modeling and performance analysis integrated system is constructed. The method has the beneficial effects that the deep linkage between the BIM modeling platform Revit and the parameterized modeling platform Dynamo is realized.

The invention provides a large model fine tuning method and system for power intelligent question answering, and relates to the technical field of natural language processing, and the method comprises the steps: obtaining power industry multi-source business data, carrying out the data preprocessing based on a preset script, and generating a structured training sample set; loading the pre-training language model, and configuring hyper-parameters; performing iterative training on the pre-training language model based on the training sample set and the hyper-parameters to obtain a fine tuning model, and storing model weight and configuration information; performing performance evaluation on the fine tuning model through the test set, wherein the performance evaluation comprises accuracy, response speed and robustness; and applying the fine tuning model to the intelligent question-answering system, judging an effect improvement condition and continuously performing fine tuning. According to the method, data preprocessing, efficient fine tuning and multi-dimensional performance evaluation are constructed, so that the adaptation efficiency and the application reliability of a large model in a professional scene are remarkably improved.

Embodiments of the present application provide a model construction method, comprising: displaying a model configuration interface configured with a class tree control; in response to the class tree control being selected, displaying an interface used for displaying a class tree, wherein the class tree comprises a plurality of nodes, and each node corresponds to a class and is configured with a child class adding control; in response to a child class adding control of a target node being selected, displaying a first child class adding interface used for receiving information of a class; and on the basis of the received information of the class, displaying an updated class tree on the interface, wherein the updated class tree comprises a newly added node located at the next level below the target node, and a class corresponding to the newly added node inherits from the class corresponding to the target node. According to the technical solution in the embodiments of the present application, the inheritance relationships between classes can be visually displayed, and visual inheritance is achieved, so that the structure of a model is simplified, and duplicate codes are reduced.

The invention discloses a model reasoning method, a model encryption method, electronic equipment and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the following steps: a main control chip receives a model encryption packet sent by a server; the main control chip calls the security chip to decrypt the symmetric key; the security chip decrypts the model encryption packet by using the symmetric key to obtain a model plaintext file, a model configuration file and a signature value, and stores the model plaintext file, the model configuration file and the signature value in a storage device; the master control chip verifies the signature of the model plaintext file and the model configuration file, and verifies the identity label of the terminal equipment under the condition that the signature verification is passed; and loading the model plaintext file for reasoning under the condition that the verification of the identity label of the terminal equipment is passed. The risk that the model is tampered or abused is reduced, the model is effectively protected, the reliability of the model reasoning result is improved, and the method can be used for various application scenes and is high in universality.

The invention discloses a video compression method based on deep learning, particularly relates to the field of video compression, is used for solving the problem that dynamic migration of residual distribution cannot be tracked in real time by entropy modeling in deep learning video compression, and accurately depicts complex distribution of residual in space and time dimensions through a multi-scale self-attention mechanism. The adaptive entropy model configuration on each piece of data is realized by using a content-based mixed density generator, a bidirectional evaluation quantity is sent to a pre-training attention model to dynamically generate a threshold value to drive a high-precision coding template to implement fine compression on a high-frequency region, and meanwhile, entropy parameters are locally contracted based on real-time bit feedback to balance output. And the consistency of a decoding end model is guaranteed through parameter freezing synchronous metadata, so that synchronous optimization of compression efficiency and reconstruction quality is realized, fluctuation is converged, artifact deviation is suppressed, and the distortion limit is approached.

The invention discloses a complex river network flood prediction method based on data dynamic cleaning and an adaptive recurrent neural network. The method comprises the following steps: S1, collecting flood monitoring data in a research area; s2, abnormal value detection and correction are carried out on the data through a speed constraint dynamic cleaning method based on flood classification; s3, preprocessing the cleaned data by using a sliding window mechanism and a data normalization technology, and setting a plurality of prediction period windows for direct multi-step prediction; s4, constructing and optimizing an adaptive recurrent neural network model, wherein the model comprises a gating circulation unit layer and a bottleneck layer; s5, optimizing the hyper-parameters through grid search and an adaptive momentum estimation optimizer, and finally determining model configuration; s6, performing complete training and testing based on the determined model configuration; and S7, flood prediction is carried out through the trained model. The model is combined with actual monitoring data, intelligent river network flood forecasting in different forecasting periods is achieved, and the flood forecasting precision and stability are remarkably improved.