255 results about "Training performance" patented technology

The invention relates to the technical field of cloud computing and artificial intelligence, in particular to a disaster recovery drill and performance evaluation system and method for multi-cluster distributed training. In order to solve the problem of disaster recovery and performance evaluation in multi-cluster distributed training, the method comprises the steps of environment analysis and modeling, fault injection and drill control, cross-cluster scheduling and resource monitoring, performance data acquisition and index aggregation, training result evaluation and tuning closed loop and the like. Multi-cluster environment elements are identified, a fault and index mapping relation is constructed, faults are injected by using a fault injection tool, and task scheduling, resource monitoring and performance data acquisition and analysis are realized by means of a multi-cluster management and monitoring tool, so that an adjusting and optimizing strategy is formed, and closed-loop management is realized. According to the method, faults can be dynamically simulated in a multi-cluster environment, the training performance is monitored, the disaster recovery and recovery efficiency is evaluated, the training efficiency and robustness of the system are improved, and the requirements for high availability and high performance in the fields of finance, automatic driving and the like are met.

The invention relates to a gradient aggregation federated learning method based on combination of zero knowledge proof and a block chain technology, and belongs to the field of distributed machine learning. Through combination of zero knowledge proof (zk-SNARK) and an extensible block chain technology, limitations of a traditional method in aspects of data privacy protection and model verification are broken through. According to the method, the correctness of a global model is verified by using zero-knowledge proof, the transparency and data privacy of an aggregation process are ensured, the Chinese remainder theorem is introduced to compress the proof, and the storage burden of block chain nodes is reduced. And the hash value updated by the encryption model is recorded through the block chain, verification is carried out based on a block chain consensus node, and the data integrity and the system transparency are further improved. Besides, a dynamic client selection mechanism is introduced in the method, participants are optimized based on training performance scores, and the training efficiency and the model quality are improved. As a whole, the invention provides a safe, efficient and transparent distributed learning solution.

The invention discloses a model training configuration parameter determination method and device and a storage medium, and relates to the technical field of computers, and the method comprises the steps: dividing configuration items into fixed configuration items and undetermined configuration items, and under the constraint that the fixed configuration items adopt specified configuration, determining the undetermined configuration items; predicting the predicted training performance of the to-be-trained model when the to-be-trained configuration item is trained by adopting different configuration parameters, and selecting a target configuration parameter from multiple configuration parameters of a group of to-be-trained configuration items based on the predicted training performance corresponding to the multiple configuration parameters, when the to-be-trained model is subjected to model training, the target configuration parameters are selected by one group of to-be-determined configuration items, so that the cost required by model training can be remarkably reduced, the resource utilization efficiency is maximized, the resource waste is reduced, the training efficiency is improved, and the problems of high hardware resource demand and high cost in a large model training stage in the related technology are solved.

The embodiment of the invention relates to a method for evaluating model training performance, computing equipment, a storage medium and a computer program product. The method comprises the following steps: determining deployment strategies of a model on a plurality of board cards, wherein the deployment strategies comprise a data parallel strategy, a tensor parallel strategy and an assembly line parallel strategy; determining the shape of the operator at least based on a tensor parallel strategy; determining a cavitation proportion corresponding to the pipeline parallel strategy at least based on the data parallel strategy and the pipeline parallel strategy; determining inter-group attenuation between the parallel data groups at least based on a tensor parallel policy and a pipeline parallel policy; and calculating the end-to-end consumption duration required by model training at least based on the shape of the operator, the cavitation ratio and the inter-group attenuation, so as to evaluate the model training performance based on the end-to-end consumption duration. According to the embodiment of the invention, when the training performance of the model is evaluated, the deployment strategy of the model and the overhead of the calculation operator and the communication operator are comprehensively considered, and the evaluation efficiency is improved.

The invention discloses a smart home energy management method based on deep countermeasure reverse reinforcement learning. The method comprises the following steps: (1) modeling a smart home energy cost minimization problem and designing an environment state and an action corresponding to a Markov decision process; (2) constructing n groups of random parameter historical data based on a group of random parameter historical data and a rolling generation method; (3) solving the minimization problem by using n groups of random parameter historical data and an optimization algorithm and obtaining n expert tracks; (4) training the reward network by the discriminator based on the expert trajectory and the trajectory generated by the generator agent; under the guidance of the reward network, training a generator agent by using an experience tuple and a near-end strategy optimization algorithm; (5) repeating the step (4) until an intelligent agent strategy with stable training performance is obtained; and (6) deploying the intelligent agent strategy obtained by training in an actual environment for operation. Compared with an existing method, the method can effectively reduce energy cost and improve user comfort.

The invention provides a training method and device for a semiconductor device performance prediction model and related equipment, and the method comprises the steps: determining key parameters and performance parameters of a semiconductor device, the key parameters comprise key size parameters and / or process parameters, and the performance parameters comprise electrical characteristic parameters of the semiconductor device; based on the multiple groups of key parameter values and the corresponding performance parameter values, establishing a sample set; the key parameter values are disturbed, and the correlation degree of the key parameters and the performance parameters is determined; wherein after the key parameter value is disturbed, the change condition of the corresponding performance parameter value is used for representing the correlation degree of the key parameter and the performance parameter; initializing the weight of a preset neural network model based on the performance prediction target and the correlation degree of the key parameter and the performance parameter; and inputting at least part of samples in the sample set into a weight-initialized preset neural network model for training to obtain a semiconductor device performance prediction model.

The invention provides a federated learning differential privacy method based on Rayleigh divergence and adaptive noise adjustment, and aims to balance data privacy protection and model training performance and improve model accuracy and convergence speed of federated learning on the premise of protecting user data privacy. And the contradiction between privacy protection and model performance in the existing federated learning is solved. The method comprises the following steps: step 1, constructing a privacy loss quantification model based on Rayleigh divergence; 2, deducing a tight upper bound of a Gaussian noise standard deviation; 3, initializing noise parameters of the federated learning system; 4, the client side executes local model training and noise adding; 5, updating the aggregation model of the central server and evaluating the performance; step 6, implementing a self-adaptive noise adjustment decision based on model performance; and step 7, iterating federal learning training until convergence or completion.

The invention relates to a set communication method and device, computer equipment, a readable storage medium and a program product. The method comprises the following steps: acquiring network state information of each intelligent calculation center in an intelligent calculation network; performing decision processing on the network state information through a pre-trained reinforcement learning model to obtain a set communication strategy for each intelligent calculation center; controlling data interaction among the intelligent calculation centers according to the set communication strategy; wherein the pre-trained reinforcement learning model is obtained by taking network state information as a state space and defining training performance of model training of the intelligent computing network as reward function training. By adopting the method, the set communication efficiency can be improved, the network congestion can be reduced, and the training rate of cross-computing-center model training can be further improved.

The invention discloses a large model training resource configuration optimization method for a cloud heterogeneous GPU cluster. The method comprises heterogeneous GPU resource configuration and resource perception stage deployment. A cost tree pruning mechanism is provided for resource configuration, a plan with low cost is searched preferentially, a configuration scheme is expressed as a tree path, meanwhile, a multi-dimensional pruning method is adopted to reduce unnecessary branches, and efficient GPU configuration is achieved. According to stage deployment, through defining a Compute-Memory Ratio (Compute-Memory Ratio, CMR), the requirements of each stage on the GPU computing power and memory capacity are considered in a combined manner, the communication and computing efficiency in the training process is optimized, and the large model training time is remarkably shortened. And finally, a large model training resource configuration optimization method Espresso oriented to the cloud heterogeneous GPU cluster is provided in combination with resource configuration and stage deployment strategies, and the user cost is reduced to the maximum extent while the training performance is ensured.

This application provides a method for sending training data of an AI model, which is applied to a scenario in which a reporting network element of the training data sends the training data to a training network element of the AI model. After obtaining the training data, the reporting network element classifies the training data into reference data and non-reference data, and indicates the training data by sending first-type information and second-type information to the training network element. The first-type information includes full information of the reference data, and the second-type information includes incremental information of the non-reference data relative to reference data corresponding to the non-reference data. Thus to help reduce overheads of sending the training data, thereby helping reduce air interface overheads in an AI model training or updating process, reduce a transmission delay, and improve model training performance.

The invention relates to the technical field of intelligent training, in particular to an intelligent shooting examination training system and method.The intelligent shooting examination training system comprises a posture recognition module, a track monitoring module, a hit analysis module, a rhythm comparison module and a label output module. According to the method, on the basis of action images in the process from gun holding to firing of a trainee, synchronous angle tracks of shoulders and elbows are recognized in real time, comparison is conducted in combination with a standard action combination, the matching difference in each time period is accurately screened and quantified, and the action control stability is dynamically evaluated by analyzing laser track coordinate points and direction changes in the aiming period; structured scores are formed by using a hit point sequence and combining path ductility and hit area density, action unit starting and ending time in a training process is compared with rhythm arrangement one by one, advanced and delayed phenomena are identified, performance scores are flexibly adjusted according to multi-dimensional index weights, and accurate matching between training performance evaluation and actual ability requirements is realized.

The invention relates to the field of power system optimization scheduling, in particular to an active power distribution network toughness improving method based on multi-agent deep reinforcement learning. The method comprises the following steps: (1) constructing a toughness improvement problem model: taking a distribution network internal key load recovery amount and a mobile energy storage pre-distribution cost as target functions, wherein the model follows a power distribution system safety operation constraint condition; (2) constructing a Markov decision process: defining Markov decision variables, and establishing a state space, an action space and a corresponding reward function; the strategy solving link is allocated to a plurality of agents to improve the training performance; (3) constructing an intelligent agent-environment interaction interface, and establishing a power flow analysis and topology check module to realize communication between the environment and the intelligent agent; and (4) model training and online application: combining an improved CTCE-MADQN algorithm with a strategy solving process, training the model to obtain a mobile energy storage configuration and microgrid group optimal division strategy, and improving the toughness of the power distribution network.

The invention provides a virtual reality simulation training-based outside-related alarm on-site disposal system, which comprises a virtual scene construction unit, a data management and analysis unit, a course design module, a resource library and an evaluation analysis system, and is characterized in that a scene editing system is constructed in the virtual scene construction unit; a landform generator and a building and facility editor are arranged in the scene editing system, and an event trigger and a behavior editor are further arranged in the virtual scene construction unit. The data management and analysis unit is used for recording data such as operation behaviors, decision-making processes and interaction conditions with virtual roles of police officers in each training, various complex, extreme or rare police affair scenes can be easily simulated, students can fully exercise the ability to cope with dangerous situations under the risk-free condition, and the training efficiency is improved. The training performance of the trainee can be accurately evaluated, the trainee can timely know the training effect of the trainee, and the instructor can conveniently adjust the teaching strategy according to the condition of the trainee.

The invention relates to the technical field of unmanned aerial vehicle training, and discloses an unmanned aerial vehicle training system and method based on multi-modal analysis and a dynamic scene, and the method comprises the steps: obtaining multi-source data, carrying out the preprocessing, obtaining multi-modal data, and inputting the multi-modal data into an intelligent analysis module; receiving the multi-modal data, analyzing the multi-modal data to obtain scene modeling information, and unifying the format of the scene modeling information; carrying out scene coverage simulation based on the scene modeling information, calculating a scene coverage rate, carrying out dynamic adjustment on the scene modeling information based on the scene coverage rate, updating the scene coverage simulation, calculating a training adaptation degree by fusing the hardware performance of the unmanned aerial vehicle after updating, and judging whether training is allowed or not; if yes, obtaining training performance data, calculating training performance values and sorting; training data of the trainees are obtained based on scene coverage simulation, the training proficiency is calculated, the training levels of the trainees are dynamically adjusted, training demand values of the trainees are calculated according to historical training data and sorted for analysis, and the training scene and the unmanned aerial vehicle are matched.

The invention relates to the technical field of practical training teaching, and discloses an intelligent practical training system based on a computer neural network, and the system comprises a knowledge explanation unit which employs a neural network model to provide personalized theoretical knowledge explanation; the skill operation simulation unit is used for constructing a simulation model to realize virtualized practice of skill operation; and the practical training evaluation unit is used for accurately evaluating the practical training performance of the user by using a deep learning technology. Meanwhile, the system is provided with a data acquisition unit which is responsible for acquiring user learning behaviors, operation simulation and practical training result data and transmitting the data into each model to acquire an output result. And finally, an intelligent feedback unit provides personalized practical training feedback and suggestions for the user through a defined intelligent feedback algorithm according to a model output result. According to the system, theoretical knowledge explanation and skill operation simulation are integrated, intelligentization and individuation of practical teaching are realized, and the practical teaching effect is effectively improved.

According to the mechanical arm control method and system based on the self-adaptive force field, the motion state of a patient is monitored, the motion state is compared with a demonstration space trajectory, and training performance information of the patient is obtained. Based on the training performance information and the constructed adaptive force field model, auxiliary force applied to the patient is obtained through calculation, wherein the auxiliary force comprises normal force, tangential force and viscous force. And the applied auxiliary torque is calculated based on the tail end posture of the mechanical arm and the target posture, the joint control torque is obtained according to the auxiliary force and the auxiliary torque, and the dynamic torque and the null space control torque of the robot dynamic model are obtained. And the joint control torque, the dynamic torque and the null space control torque are combined to obtain the comprehensive control torque for controlling the mechanical arm. Therefore, the self-adaptive adjustment of the auxiliary force is realized based on the position, speed and direction of the patient, and the time degree of freedom can be obtained in the motion of the self-adaptive force field, so that the patient obtains more freedom of active motion under the assistance of the space force field.

The invention discloses an efficient distributed machine learning training system based on a server-free architecture, and the system comprises a modeling module which is used for obtaining an input workload for modeling; the resource configuration module is configured with an optimized mathematical model and used for obtaining optimal parameter configuration based on the optimized mathematical model on the basis that the modeling module completes modeling; optimizing the relation between double-layer parameters and system-level parameters of mathematical model modeling server-free distributed training and training delay, cost and convergence efficiency; and the K-REDUCE training framework is used for triggering the server-free function, and the server-free function carries out distributed training according to the K-REDUCE framework. According to the method, a mathematical model is constructed to describe the relationship between parameters and performance, and optimal parameter configuration is searched by adopting a heuristic search algorithm based on pruning, so that the training performance is effectively improved, the economic cost is reduced, system resources are dynamically adjusted, and the optimal training performance is realized.

The invention discloses an iterative supervision fine tuning data generation method and system, and relates to artificial intelligence and data processing technologies, and the method comprises the steps: generating a basic training data pair based on a pre-training language model, and the basic training data pair comprises an instruction-response pair; performing model training based on the basic training data, and determining training performance feedback data according to the performance of the model on the verification set and the task index; analyzing a model performance index according to the determined training performance feedback data; predicting and optimizing an optimization strategy of training data and training parameters based on the model performance index and the basic training data pair; and performing model training according to the optimized training data and the training parameters so as to execute AI model training platform data generation by using a strategy output by the model after training. Model performance feedback, automatic task mining and sample optimization mechanisms are introduced, training data are dynamically adjusted and expanded, and closed-loop iteration and performance-oriented optimization of data generation are achieved.

The invention discloses a DIKWP-driven individualized brain-map interactive feedback system, which is used for neural rehabilitation and brain-computer interface training. The system obtains brain activity data of a patient through electroencephalogram acquisition equipment, constructs a personal brain-semantic map in combination with cognitive evaluation, and establishes a mapping relation between semantic units and brain region responses. In the training process, the DIKWP semantic analysis module performs multi-layer analysis on indexes such as reaction time, accuracy and intention achievement, generates multi-mode instant feedback such as visual sense, auditory sense or tactile sense, and performs directional reinforcement on a weak semantic domain. The system has a dynamic target optimization capability, and the difficulty can be automatically adjusted according to training performance; when attention distraction, emotion abnormity or semantic deviation is detected, a safety intervention mechanism is automatically triggered, and training effectiveness and safety are guaranteed. According to the method, closed-loop individualized rehabilitation interaction is realized, the adaptation degree and efficiency of brain-computer interface training are remarkably improved, and the method is suitable for various rehabilitation scenes such as languages, movement and cognition and has a good industrial application prospect.

The invention discloses an intelligent computing cluster parallel training performance optimization method oriented to large model training, and relates to the technical field of artificial intelligence computing. Hardware feature indexes of heterogeneous computing nodes in an intelligent computing cluster are collected in real time, a hardware topological graph is constructed, and an optimal computing power combination suitable for the hardware topological graph is calculated; the method comprises the following steps: selecting an optimal communication path by adopting a self-adaptive routing algorithm, coding and decoding transmission data in combination with a mixed precision compression technology to realize efficient transmission of the transmission data, constructing a fault prediction model based on a deep learning algorithm, and inputting real-time data in a hardware monitoring log into the fault prediction model to obtain a prediction result. Prediction results are classified and stored according to a preset result classification standard, check points are trained based on the prediction results to achieve rapid recovery, the super-node computing power utilization rate is improved, meanwhile, a communication path can be dynamically adjusted according to the network congestion state, and the fault recovery speed is improved.

The invention discloses a method for establishing a rough-arrangement search model based on LTR, and the method comprises the following steps: S1, collecting candidate total data, and completing the preprocessing operation; s2, the sorting score of each piece of data is calculated and arranged in an ascending order, and the data are divided into equivalent data buckets; s3, randomly extracting data to form a training sample pair, generating a Pair-wise data set, and marking a sorting relationship; s4, executing rough-arrangement search model training and calculating a prediction score; s5, constructing a loss function by adopting a Lambda gradient signal, feeding back a sorting error and iteratively updating model parameters; s6, monitoring a training performance index, stopping training after a stopping condition is met, and deploying the model to an online environment; and S7, storing the trained model file, and defining an input format, a sorting structure and an output interface. According to the method, the sorting accuracy and fine sorting consistency of the rough sorting search model are improved, and the generalization ability of the model is remarkably enhanced.

A table tennis recognition system based on YOLOv8 and an edge computing device deployment method belong to the technical field of computer vision, the table tennis recognition system comprises a camera, an edge computing device and a pitching machine, the camera and the pitching machine are connected with the edge computing device, and the edge computing device is connected with a main control center. The table tennis ball recognition system comprises the following steps: S1, image acquisition; s2, image preprocessing; s3, the YOLOv8 network is improved; s4, training a network model; s5, carrying out RKNN conversion on the model, and carrying out quantization processing; and S6, deploying to Orange Pi5Plus (Orange Pi5Plus), and carrying out multi-thread real-time detection on the table tennis ball. According to the system, a training mode instruction is received according to the specific requirement of a user, and the ball serving machine is controlled to serve balls according to a specified mode after the training mode instruction is processed by orange pie 5Plus; a personalized serving strategy can be customized according to the actual training performance of the user, and the coordinate position of the ball is displayed in real time; according to the invention, high-efficiency and low-cost real-time table tennis ball detection and analysis are realized.

The invention discloses a VR fire-fighting training system, method and device based on Unity, and the system comprises a scene simulation module, a user interaction module, a personalized feedback module and the like, the system simulates a plurality of immersive fire virtual training scenes based on a Unity engine, and deploys a VR fire-fighting training system; receiving a user gesture or a controller instruction by means of VR equipment, and guiding the user to perform interactive training in the immersive fire virtual training scene; in user interaction training, the system records and intelligently evaluates user interaction training performance, and individually adjusts and optimizes an immersive fire virtual training scene according to the user performance. According to the invention, a vivid fire environment is created, a user can feel the temperature, smoke and sound of a fire scene in an immersive manner, a strong crisis feeling is formed, and the highly immersive experience can help the user to master fire-fighting knowledge more quickly and make a correct judgment when a fire occurs in reality.

The invention discloses a rail transit intelligent operation and maintenance practical training system based on cloud computing, which relates to the technical field of operation and maintenance practical training teaching and comprises a practical training integration unit, a data acquisition unit, a practical training analysis unit, a priority analysis unit, a task issuing unit and a period updating unit. According to the invention, the training performance index is utilized to accurately quantify the ability weakness of the trainee, the qualified training items and the unqualified training items are divided in combination with the threshold, and the computing resources are focused on the unqualified items, so that the cloud computing load is reduced; secondly, through dynamic sorting of practical training project priority indexes, it is ensured that weak practical training projects with high fault risk and high cost are preferentially trained, and the practical training resource allocation accuracy is improved; and meanwhile, the maximum release times of a single training item are forcibly constrained through a preset release times threshold value, and each training item in the unqualified set is trained at least once, so that the training trainees are prevented from being overtrained on a single item, and other weak items are prevented from being omitted.

The present application describes methods of using the pre-configured AI / ML (artificial intelligence / machine learning) based model training modes in wireless mobile communication system including base station (e.g., gNB, TN, NTN) and mobile station (e.g., UE). In AI / ML model is applied to radio access network, model training performance can be degraded depending on UE ML condition or capability changes. Therefore, model operation (e.g., model training / inferencing / monitoring / updating) can be set up between network and UE by configuring model training capability elements.

The invention discloses a driving school data management method for driving training contract management, and relates to the technical field of machine learning, and the method comprises the following steps: data collection and student information input; making a training plan according to the student information; performing course tracking and adjustment according to the training plan; performing contract validity period prediction according to the training performance data; fee payment and fee refund management are carried out by using contract validity period prediction data; trainee compliance detection is carried out through contract and payment data; performing contract renewal according to a detection result; and generating an optimization management decision report based on the renewal result. According to the innovative compliance detection method based on the weighted logistic regression model, the model parameters are adjusted in real time, and the weights of the features are optimized, so that a driving school can monitor the compliance of trainees more accurately, potential default behaviors are found in time, the management efficiency is improved, and the default risk is reduced.

The invention discloses a knowledge graph embedding method based on negative sampling and weighting. The method comprises the following steps: acquiring a medicine data set; based on an embedded vector obtained by pre-training and a designed high-quality candidate negative sample set construction strategy, sampling a high-quality candidate negative sample set for all positive triads; according to the knowledge graph embedding model based on translation, calculating scoring function values of the triad and the negative triad corresponding to the triad; carrying out statistics on related frequency data in the data set, and calculating a frequency-based weight of each triple; according to the scoring function value and the weight, calculating the loss L of the triple set to obtain a final loss function of the training framework; optimizing a loss function of the training model by adopting a stochastic gradient descent method; and outputting a potential association relationship between the medicine and the disease and the symptom. According to the method, the negative sampling quality of the model is improved, the adverse effects of long-tail entities and relationships on model training are relieved by weighting the loss function, and the model training performance is improved.

The invention discloses a pulse neural network lightweight method based on nonlinear pruning, and relates to the technical field of artificial intelligence and neural networks. The method comprises the following steps: S1, constructing an NDI-LIF spiking neural network model, and introducing a bilinear product term into an input current in a dynamic updating process of a neuron membrane potential; s2, constructing a nonlinear synaptic pruning mechanism, representing a connection weight as a re-parameterization function composed of a re-parameterization weight and a conversion gain coefficient, and pruning the connection weight based on the re-parameterization function; and S3, training the NSPDI-SNN model to obtain a lightweight pulse neural network. By introducing a nonlinear dendritic integration mechanism and a state-adjustable synaptic pruning mechanism, balance between network sparsity and high performance is realized while the expression ability of the model is enhanced, and a lightweight spiking neural network model with high spatial-temporal expression ability, highly sparse structure and reasonable biological mechanism is constructed. And the generalization ability and the efficient training performance under various tasks are ensured.

To provide a system.SOLUTION: Provided is a personal training system using generative AI, comprising: means for generating a personalized training program based on the user's fitness level and goals; means for providing training guidance to the user based on the generated training program; and means for monitoring the user's training performance and modifying the training program when necessary.SELECTED DRAWING: Figure 1

The invention relates to a heterogeneous hardware recommendation method and device, a medium and electronic equipment. The method comprises the following steps: receiving hardware demand information; for each piece of candidate heterogeneous hardware, target scores corresponding to the candidate heterogeneous hardware in a plurality of preset score dimensions are determined, and the preset score dimensions comprise a hardware capability dimension, a reasoning performance dimension, a training performance dimension and an ecological support dimension; determining a target scoring dimension in the preset scoring dimensions according to the hardware demand information; and determining at least one piece of to-be-recommended first target heterogeneous hardware in the plurality of pieces of candidate heterogeneous hardware according to a target score corresponding to the candidate heterogeneous hardware in the target score dimension, and outputting hardware information of the first target heterogeneous hardware. Therefore, the performance of the heterogeneous hardware in different scoring dimensions and the actual demand of the user are comprehensively considered, and personalized recommendation for the user can be realized.