72 results about "Continuous training" patented technology

Described is a system for performing a set of machine learning model training operations that include: accessing media content items associated with interaction functions initiated by users of an interaction system, generating training data including labels for the media content items, extracting features from a media content item of the media content items, identifying additional media content items to include in the training data based on the extracted features from the media content item, processing the training data using a machine learning model to generate a media content item output; and updating one or more parameters of the machine learning model based on the media content item output. The system checks whether retraining criteria has been met, and repeats the set of machine learning model training operations to retrain the machine learning model.

The invention belongs to but not limited to the technical field of communication, and particularly relates to an air-to-ground network optimization method and system based on MEC and digital twinning, and the method comprises the steps: carrying out the learning and decision-making of an intelligent agent through employing a main strategy network and a commentator network of deep reinforcement learning; initializing parameters of a strategy network and an evaluation network, and setting training related hyper-parameters; the intelligent agent interacts with the environment according to the current strategy, executes an action and updates a state; an experience playback buffer area is used for storing experience, and updating is carried out when the capacity is reached so as to keep the latest learning experience; performing parameter updating on the main strategy network and the commentator network by adopting loss function calculation and a gradient descent method; soft updating of strategy network and value network parameters in the target network is carried out regularly to ensure learning stability; training is continued until the strategy converges, and then task allocation and trajectory planning are completed by applying the strategy.

Described is a system for performing a set of machine learning model training operations that include: accessing media content items associated with interaction functions initiated by users of an interaction system, generating training data including labels for the media content items, extracting features from a media content item of the media content items, identifying additional media content items to include in the training data based on the extracted features from the media content item, processing the training data using a machine learning model to generate a media content item output; and updating one or more parameters of the machine learning model based on the media content item output. The system checks whether retraining criteria has been met, and repeats the set of machine learning model training operations to retrain the machine learning model.

The invention relates to the technical field of visual health management, in particular to a ciliary muscle adjustment training system and method based on eye movement tracking, and the method comprises the steps: collecting eye movement data of a fixation point track, an adjustment amplitude change rate and pupil response time in real time, and combining personalized parameters such as the age of a user, diopter and a training target; inputting a ciliary muscle training dynamic planning algorithm module, dynamically generating a stage scheme containing a training action type, a single-group duration, a stimulation interval and a focal length adjustment strategy, and adaptively adjusting the strategy according to real-time adjustment stability and a historical ability curve; through multi-dimensional matching analysis of a pre-stored standard eye movement template, a training specification score is calculated from the track goodness of fit, speed uniformity, amplitude standard-reaching rate and binocular coordination, visual fatigue parameters are obtained based on continuous training duration, pupil fluctuation and retina reflex change, a scoring result is dynamically corrected, and a visual fatigue evaluation result is obtained. Personalized, self-adaptive and quantitative safety control of the training process is realized.

The invention relates to an intelligent agent continuous training and effect evaluation closed-loop method and system fusing work order feedback, and a medium, and relates to the technical field of deep learning. The agent continuous training and effect evaluation closed-loop method comprises the following steps: receiving and analyzing work order feedback information, and recording the number of error information and information storage duration in combination with a preset structure sample label; dynamically judging the number of error information and the information storage duration, triggering a retraining rule of an original agent model, distributing candidate information weights, and screening a target sample set; directionally training and optimizing the original agent model according to a retraining rule in combination with the target sample set, and generating a new agent model; constructing a historical work order problem set, comparing and verifying a new agent model in combination with an original agent model, generating an agent evaluation table, and correcting a candidate information weight; by accurately focusing the weak link, the continuous and efficient enhancement of the model capability is realized, so that the answer accuracy of the intelligent agent is remarkably improved, and the AI illusion is effectively inhibited.

A system identifying anomalies in an image of an object is first trained using first sets of images corresponding to first anomaly types for the object. A model of the object is formed in a latent space. A label for each anomalous image is used to calculate vectors containing means and standard deviations for each first anomaly types. The means and standard deviations are used to calculate a log-likelihood loss for each first anomaly type. The system is retrained using second sets of images corresponding to second anomaly types for the object. The vectors are supplemented using labels for each second anomaly types. A statistically sufficient sample of information in the means and standard deviations vectors is supplied to the latent space. A log-likelihood loss for each of the first and second anomaly types is calculated based on their respective mean and standard deviation.

The invention discloses a weak supervision time sequence action positioning method and system for a self-supervision pseudo label optimization network, and the method comprises the steps: dividing all segments in a training video into a plurality of class clusters, and obtaining atomic action features representing global semantics; taking the atomic action features representing global semantics as action representation, mapping the video clip-level features extracted by the feature extractor into query vectors, and participating in attention calculation together with the atomic action features to generate enhanced fragment representation containing global context information; the method comprises the following steps: identifying an inconsistent region by utilizing a high-discrimination characteristic expressed by an enhanced fragment, executing pseudo-label correction by collecting pseudo-label results generated in two continuous training periods, judging whether to expand a label range and generate a pseudo-single-frame label yfp by adopting an affinity search strategy and utilizing characteristic similarity, and optimizing through consistency loss. And obtaining a positioning result. According to the method, the marking cost is remarkably reduced, background interference is effectively reduced, and the integrity and precision of action positioning are improved.

The invention relates to the technical field of power system power distribution networks, in particular to a multi-source energy storage power distribution network voltage regulation and control method based on a safety reinforcement learning algorithm. Firstly, a power distribution network voltage regulation and control model is constructed, power distribution network nodes are regarded as state nodes, the electrical connection relation forms a state space, and regulation and control operation is used as an action space. A state transition probability matrix is established by analyzing a physical relationship, and a reward function is defined in combination with security operation constraints. Then, a near-end strategy optimization algorithm is selected, the preprocessed operation state data of the power distribution network are used as input, and the intelligent agent and the environment are subjected to interactive training; during training, a reward function is utilized to guide an intelligent agent to learn a safe and effective regulation and control strategy. The average reward and the voltage qualified rate are monitored through continuous training, and the model can adapt to dynamic changes of the power distribution network after convergence. The problems that a traditional regulation and control method is difficult to adapt to complex changes of the power distribution network and does not comprehensively consider various factors are solved.

The invention belongs to the technical field of water area rescue practical training, and provides a torrent water channel and urban inland inundation combined practical training system which comprises a torrent water pump room, a water distribution pool, a training water channel, a training pool and an urban inland inundation training area. The training water channel communicates with the water distribution pool and the training pool. The urban inland inundation training area is communicated with the water distribution pool and the training pool. According to the system, a torrent water pump house, a water distribution pool, a training water channel, a training water pool and an urban inland inundation training area are integrated into an organic whole, and a parallel water channel is designed, so that a single system can provide two differentiated training environments of torrent rescue training and urban inland inundation rescue training at the same time. Flow regime areas such as a boiling line, a rolling flow, a covering flow, a smile flow, a rugosa flow and a vortex flow are arranged in the training water channel, and the most typical and dangerous water flow form in a natural river is systematically and completely reproduced. Therefore, learning and training personnel can learn and master skills of identifying and coping with various complex water flow forms in a continuous training water channel.

The present invention discloses an underwater computing task offloading method based on an improved deep deterministic policy gradient. First, the state space of the task offloading problem is initialized, the action space is constructed by adding noise to the offloading mode and offloading quantity matrix, and the reward function is designed using the reciprocal of the task offloading energy consumption. Secondly, the deep deterministic policy gradient algorithm is improved, and the adaptive update amplitude method is used in the process of updating the algorithm parameters. The update amplitude is set to a relatively high value in the initial stage and gradually reduced during the training of the algorithm. Finally, the task offloading problem is solved by the improved deep deterministic policy gradient. The Markov decision model of the task offloading problem is input, and continuous training is iterated until the average loss function no longer decreases, and the task offloading mode and offloading quantity are output. The present invention can effectively accelerate the convergence speed of the algorithm and reduce the energy consumption of sensor node task offloading.

The invention provides a self-adaptive block chain water basin monitoring method and system based on an intelligent analysis algorithm, and the method comprises the steps: obtaining newly-added sensor monitoring data of a source domain and a target domain of a water basin monitoring station, extracting a monitoring data sequence through a sliding time window and data fusion method, and generating a newly-added sample set; taking the newly added sample set as a training sample set; constructing a self-adaptive water basin analysis network model fusing GCN and LSTM, training based on the training sample set, and calculating the overall loss of the model; if the overall loss is greater than a threshold value, extracting a part of samples from the training sample set according to a data importance sampling algorithm, storing the samples into an experience learning library, and combining the samples with a next round of newly added samples to form a new round of training sample set for continuous training; otherwise, finishing training to obtain a water basin monitoring model; and monitoring the water basin condition of the target domain based on the model to obtain a monitoring result. According to the invention, accuracy and real-time performance of water quality parameter, water level change and water flow speed monitoring can be improved, and data safety is improved.

Provided in the present application are an apparatus and method for training and evaluating an AI model. The method comprises: creating a process task template applied to a containerized application management platform, wherein the process task template comprises: process configuration information associated with a training and evaluation task of an AI model, and a stage flag corresponding to each step in the training and evaluation task; and in response to receiving a start command for a continuous training and evaluation task: querying the process task template, splitting the training and evaluation task into stages on the basis of the process configuration information and the stage flag, so as to construct a task in a training stage, performing product escape analysis on the basis of the task in the training stage, so as to generate a product escape analysis result associated with the task in the training stage, starting the task in the training stage on the basis of a product escape result, and when the task in the training stage reaches a preset training stage node, starting the concurrent running of one or more tasks in an evaluation stage without interrupting the task in the training stage.

Provided is a continuous training method in a limited mobile environment. The method for continuously training a deep learning model according to an embodiment of the present invention involves: fine-tuning a pre-trained deep learning model with a first data set; fixing some weights in the fine-tuned deep learning model in the first fine-tuning step, and pruning the remaining weights; and loading pre-trained weights on the pruned weights. Accordingly, effective continuous training is possible in a fixed small model structure without a neural network addition process in a resource-limited mobile device environment.

Described is a system for performing a set of machine learning model training operations that include: accessing media content items associated with interaction functions initiated by users of an interaction system, generating training data including labels for the media content items, extracting features from a media content item of the media content items, identifying additional media content items to include in the training data based on the extracted features from the media content item, processing the training data using a machine learning model to generate a media content item output; and updating one or more parameters of the machine learning model based on the media content item output. The system checks whether retraining criteria has been met, and repeats the set of machine learning model training operations to retrain the machine learning model.

The invention provides a large language model staged pre-training method and system. The method comprises the following steps: training a Transform model by using a basic data set, optimizing a negative logarithm likelihood target, and adopting an AdamW optimizer and a cosine attenuation learning rate; continuing training by using the universal knowledge data set based on the first-stage parameters; and weighting the professional data in the training field by adopting an oversampling strategy. By structuring a training target and a data type, the model can efficiently learn language basis, general knowledge and professional skills in stages. Experiments show that according to the method, the training efficiency of the model in the basic stage is improved by 40%, the overall training time is shortened by 30%, and meanwhile the accuracy rate of tasks in the professional field is 15%-20% higher than that of traditional end-to-end training. Finally, model parameters are evaluated through professional ability, and both universal language understanding and domain specialty are achieved.

The present invention provides a method and device for planning gait trajectories for a lower-limb exoskeleton based on self-oscillation. The method comprises continuously collecting joint angle information and plantar pressure information from a plurality of walking groups of a wearer, preliminarily dividing the walking process into multiple gait phases based on the plantar pressure information, using the preliminarily acquired gait phases and corresponding joint angle information as a data set, building a convolutional neural network (CNN) model, using the data set as training data, and continuously training the CNN model to accurately acquire the starting point of each gait phase. Each gait phase is accurately acquired from the starting point, establishing an adaptive oscillator model, inputting the data set into the adaptive oscillator model, outputting a tracking phase, and performing gait trajectory planning for the lower-limb exoskeleton based on the gait phases acquired by the CNN convolutional neural network model. The predicted output trajectory is dynamically adjusted in real time online. The present invention can adapt to the requirement of non-constant walking speed and achieve dynamic gait trajectory planning.

The invention discloses a fireman physical sign monitoring and early warning method based on a multivariate state estimation technology, relates to the technical field of fireman body abnormal state monitoring and early warning, and is realized through a vital sign acquisition device and a digital individual soldier. Comprising the following steps that S1, heart rate, respiration, body temperature, perspiration and blood oxygen data of firemen in ordinary states are collected through a vital sign collecting device, and the ordinary states comprise a daily activity state, a training state and a state after continuous training for one hour; according to the method, the physical state difference between individual firemen is comprehensively considered, the vital sign state of the individual firemen can be evaluated more accurately, early warning can be given out in advance when the body of the firemen is in a potential danger state, the digital individual soldier calculates and analyzes the data, the influence caused by network abnormity on a rescue site is avoided, and the rescue efficiency is improved. At the same time, the system can send out the alarm information of the abnormal state of the firefighter at the first time, and can send the alarm information to the digital individual terminals on other teammates and the on-site headquarters.

The invention provides a progressive respiratory training device which comprises a shell, an air inlet and an air outlet are formed in the two ends of the shell respectively, an impeller is further rotatably arranged in the shell, a propelling disc is slidably arranged in the shell, the propelling disc is connected with the impeller through a resistance mechanism, and a driving plate is hinged to the interior of the air outlet through a torsional spring. The driving plate is connected with the propelling disc through a propelling mechanism so that rotation of the driving plate can be converted into driving of the propelling disc to slide by a specified distance, and a reset mechanism connected with the propelling disc is arranged on the shell. According to the device, the training intensity can be automatically and gradually improved, manual adjustment of a patient or a medical worker is not needed, the labor intensity of the worker is greatly reduced, meanwhile, training interruption caused by manual adjustment is avoided, the training continuity is guaranteed, and when the breathing intensity of the patient cannot control the rotation driving angle of the driving plate, the driving angle is adjusted. Therefore, the strength can be kept for continuous training, the maximum training strength can be quickly achieved, and the training effect is improved.

A system and method for configuration-driven continuous training of large language models in autonomous medical coding normalize multi-source electronic health data into provenance-preserving, specialty-aware schemas and provide schema-mapped context to a configurable multi-stage reasoning pipeline to generate candidate code bundles including evaluation and management levels, procedure codes, modifiers, diagnosis codes, and linkages. A ruleset-based validation layer enforces bundling, modifier applicability, and payer policies, and emits machine-readable explanations with rule identifiers, evidence references, and pass / fail indicators. A telemetry-enabled user interface captures approvals, rejections, edits, and rationales, which are consolidated with validation results into provenance-tracked training examples via a trust- or quality-weighted consensus. Signals from reasoning outputs, validation checks, user interactions, and downstream outcomes are fused to produce continuous learning signals that update prompt templates and output ranking parameters through configuration rollouts-without redeploying base models. The approach improves accuracy, robustness, explainability, and auditability for scalable EHR integration.

This invention discloses an image risk classification method and apparatus based on YOLOv7. The method crawls basic NFT information through the API provided by OpenSea, obtains and partitions an open-source dataset, and uses the partitioned training and validation sets to train the model. After improving the YOLOv7 algorithm, the backbone network of the model extracts and fuses features from the images to obtain image feature maps; the head network in the model fuses features from the feature maps and uses a new GAP-FCN classification structure to obtain classification results; the loss is calculated using the obtained classification results and preset labels, and the network weights are updated in reverse. Through continuous training with multiple batches of samples, a trained model is obtained; the images from the partitioned test set or the images to be predicted are input into the trained model, and the input, extraction, and fusion are repeated to obtain the image risk classification results. The data augmentation technique used in this invention expands the dataset, increases the number of training samples, and improves data utilization.

This invention discloses a mechanical equipment fault monitoring system and method based on multidimensional data-driven approaches, belonging to the field of fault prediction technology. The method specifically includes: collecting multidimensional data of the mechanical equipment, preprocessing the collected multidimensional data, extracting features from the preprocessed multidimensional data, introducing a time-frequency locality adjustment mechanism to obtain first local weight coefficients, weighting the first local weight coefficients using a likelihood function to obtain second local weight coefficients, weighting and fusing the extracted features of the preprocessed multidimensional data, training a fault prediction model using common fault data of the mechanical equipment, and predicting faults of the mechanical equipment based on the trained fault prediction model and the features of the weighted and fused multidimensional data. This invention, through continuous training and optimization of the model, adapts to the operating characteristics of the equipment under different working conditions, automatically adjusts the fault prediction model, and possesses strong adaptive capabilities.

The invention discloses a continuous training device for a two-way memory alloy wire. The continuous training device comprises a pay-off wheel, a stretching wheel and a plurality of heating and cooling assemblies which are sequentially arranged in the advancing direction of the wire. The pay-off wheel is connected with a first motor, the stretching wheel is connected with a second motor, and the advancing speed of a wire on the stretching wheel is larger than that of the wire on the pay-off wheel; the heating and cooling assembly comprises a hollow roller and a constant tension wheel which are arranged in the wire advancing direction. Wherein a resistance wire is arranged in the hollow roller, and the constant tension wheel is arranged in the cooling box; according to the invention, the hollow roller and the constant tension wheel form the heating and cooling assembly, so that the wire can continuously advance between the two parts, and meanwhile, the wire is subjected to cold and hot variable temperature circulation, so that the cold and hot circulation training of the wire is realized, and the training efficiency is improved.

The application relates to the technical field of large model training strategies, in particular to a distributed fault tolerance and stability guarantee method for large model training, which comprises the following steps: acquiring model training related data in a large model training process; analyzing the abnormality degree of the model training state in a single training cycle; analyzing the stability change degree of the training state of the model in a continuous training process; and evaluating whether the model parameters of a single training cycle are saved by a checkpoint. The application aims to adjust the saving strategy of the checkpoint by analyzing the model training state, so as to improve the efficiency of breakpoint resuming training of the model.

The present invention provides a tennis transmitter, which relates to the field of transmitters and includes a launching mechanism, a ball-dialing disc, a ball storage bin, and a mounting bracket. The ball-dialing disc and the ball storage bin are installed on the mounting bracket. The ball-dialing disc is arranged at the bottom of the ball storage bin, and the outlet of the ball-dialing disc is communicated with the inlet of the launching mechanism. The ball-dialing disc is used to convey tennis balls from the ball storage bin into the launching mechanism. The launching mechanism includes a link assembly, a pitching assembly, and a limiting assembly. When it is necessary to adjust the launching angle of the launching mechanism, the second baffle in the pitching assembly rotates. At this time, the temporary storage bin rotates around the axis of the rotating shaft. Since the centers of the tennis balls at the outlet of the link assembly intersect with the axis of the rotating shaft when the tennis balls are stacked at the outlet of the link assembly, the link can change but the tennis balls are not squeezed and will not get stuck. The limiting assembly prevents the tennis balls in the temporary storage bin from falling, so that the tennis balls will not fall out when serving is not required, ensuring continuous launching of tennis balls and enabling continuous training for the trainees.

The invention provides a training evaluation device and method for an AI model. The method comprises the steps that a process task template applied to a containerized application management platform is created, wherein the process task template comprises process configuration information associated with a training evaluation task of an AI model and stage marks corresponding to all steps in the training evaluation task; and in response to a received starting command of the continuous training evaluation task, querying a process task template, performing stage splitting on the training evaluation task according to the process configuration information and the stage mark to construct a training stage task, performing product escape analysis according to the training stage task, and performing product escape analysis according to the product escape analysis. In one embodiment, the method includes receiving a training stage task, generating a product escape analysis result associated with the training stage task, initiating the training stage task based on the product escape result, and initiating concurrent operation of one or more evaluation stage tasks without interrupting the training stage task when the training stage task reaches a predetermined training stage node.

Provided is a learning difficulty level-based continuous training method for a mobile deep learning model. The continuous training method according to an embodiment of the present invention includes obtaining datasets to be used for continuous training of a previously-trained deep learning model, calculating learning difficulty levels of the datasets, determining a training order of the datasets according to the calculated learning difficulty levels, and continuously training the deep learning model with the datasets according to the determined training order. Accordingly, it is possible to maximize the performance of the continuous training within restricted resources and model environment by determining the training order according to the learning difficulty levels of the datasets and continuously training the deep learning model.

The utility model discloses a knee joint loosening training device which comprises an inclined plate, a positioning ring fixedly connected to the inclined plate, a rotating shaft installed in the positioning ring through a bearing, an L-shaped supporting plate fixedly connected to the outer surface of the rotating shaft, a gear fixedly connected to one end of the rotating shaft, and a reinforcing rod fixedly connected to the outer surface of the positioning ring. And one end of the reinforcing rod is fixedly connected with a limiting frame, a driving rack is slidably connected into the limiting frame, and the gear is in engaged connection with the driving rack. According to the knee joint looseness training device, a servo electric cylinder can drive a fixing plate to move, a driving rack can slide in a limiting frame, the driving rack can drive a gear to rotate, an L-shaped supporting plate can drive the shank of a patient to swing, then the knee joint is trained, the servo electric cylinder is adopted for providing power, the patient does not need to apply force, and the training efficiency is improved. The patient is more labor-saving during training, and continuous training can be carried out.

Described is a system for performing a set of machine learning model training operations that include: accessing media content items associated with interaction functions initiated by users of an interaction system, generating training data including labels for the media content items, extracting features from a media content item of the media content items, identifying additional media content items to include in the training data based on the extracted features from the media content item, processing the training data using a machine learning model to generate a media content item output; and updating one or more parameters of the machine learning model based on the media content item output. The system checks whether retraining criteria has been met, and repeats the set of machine learning model training operations to retrain the machine learning model.

A system, method and device for early detection of lower extremity injuries in patients with neuropathy who may have impaired pain sensation. The system comprises a sensor apparatus with multiple sensors that measure physiological parameters at specific locations on a patient's foot and lower limb. A connected computing entity receives and analyzes sensor data to identify potential injury markers such as temperature anomalies, tissue oxygenation changes, and swelling. Using machine learning algorithms that improve through continuous training, the system determines injury likelihood based on personalized baselines and historical data at a patient level. The computing entity generates an intuitive user interface displaying injury likelihood assessments for healthcare providers. This technology enables early intervention through non-invasive monitoring, potentially preventing serious complications in a vulnerable populations.

The invention discloses a multi-board-card signal synchronization method and device, a storage medium and a computer product, and relates to the technical field of integrated circuits, the method is applied to a signal synchronization card integrated with a training optimization module and a reconfirmation synchronization module which are in communication connection, and the signal synchronization card establishes communication connection with each acquisition card; the training optimization module responds to the signal synchronization requirement of each acquisition card, determines a synchronization training result according to the training byte data sent by each acquisition card, and further generates a control signal according to a stepping intermediate value generated by executing optimization operation for the synchronization state of the multiple board cards according to the synchronization training result; the reconfirmation synchronization module responds to the control signal sent by the training optimization module, determines a synchronization reconfirmation result according to the received intermediate byte data of the continuous training byte data sent by each acquisition card, and determines that all the acquisition cards are in a synchronous acquisition state for a reconfirmation synchronization state according to the synchronization reconfirmation result; the invention aims to improve the synchronization precision among multiple acquisition cards.