15 results about "Discriminative model" patented technology

A single-channel speech enhancement method and device based on a mean-reverting Schrodinger bridge. The method comprises: forming clean speech samples and noisy speech samples into a set of sample pairs; performing preprocessing and Fourier transform on speech samples to obtain spectral complex matrix pairs; constructing a discriminative model, estimating a spectral complex matrix of a clean speech, adjusting model parameters on the basis of a difference between the estimated spectral complex matrix of the clean speech and a spectral complex matrix of a real clean speech, and using the discriminative model corresponding to an optimal parameter as a target discriminative model; constructing a score-based model, estimating a reverse optimal shift score in a reverse generation process, adjusting model parameters on the basis of a difference between the estimated reverse optimal shift score and a real reverse optimal shift score, and using the score-based model corresponding to an optimal parameter as a target score-based model; and providing a noisy speech, using a trained target score-based model to parameterize a reverse optimal shift score of a mean-reverting Schrodinger bridge, and performing the reverse generation process to generate a clean speech. The method involves lower computation costs and achieves a better speech enhancement effect.

The present technology relates to a signal processing apparatus and a method that enable improvement in the robustness of emotion estimation against noise. A signal processing apparatus extracts, on the basis of a measured biological signal, a physiological measure contributing to an emotion as a feature amount, outputs, with respect to time-series data about the feature amount, time-series data about a prediction label of an emotion status by a discriminative model built in advance, and outputs an emotion estimation result on the basis of a result of performing weighted summation of the prediction label with prediction label reliability that is reliability of the prediction label. The present technology can be applied to an emotion estimation processing system.

This invention discloses a method for positive / negative identification in medical dialogues based on cue learning, relating to the field of medical dialogues. The method includes the following steps: designing a cue template based on the characteristics of cue learning and positive / negative identification in medical dialogues. The cue template includes the original input text x and a special token; copying the entity [Ent] mentioned in the original input text x that needs to be predicted into the cue text, adding conjunctions designed according to the task characteristics before and after the entity [Ent], and adding the identifier [MASK] as a special token to be predicted, thus completing the cue template design; designing a vocabulary and using the vocabulary and cue template to train a pre-trained language model to obtain a cue learning-based discriminative model. This invention solves the problems of existing discriminative methods failing to fully utilize the prior knowledge of the pre-trained language model, ignoring the independent relationship between context and entities, and consuming large amounts of computer resources.

The present application relates to the technical field of power operation scheduling, and provides a power system operation personnel scheduling optimization method, system, device and medium, which comprises obtaining power operation task data, scheduling constraint rules, historical scheduling data and personnel skill data; feature extraction is performed on the historical scheduling data to obtain a case feature data set, and an expert preference discrimination model is constructed according to the case feature data set; according to the power operation task data, task period decomposition is performed to obtain a target shift combination with the minimum total task operation cost as the optimization objective; and according to the target shift combination, the scheduling constraint rules, the personnel skill data and the expert preference discrimination model, personnel allocation optimization is performed based on a cultural gene algorithm to obtain a target scheduling scheme. The present application can ensure the scheduling optimization quality and efficiency, take into account the scheduling scheme feasibility and personnel allocation efficiency, and improve the practicality and reliability of the scheduling scheme based on the two-stage optimization mechanism of task period decomposition first and then personnel allocation.

This application discloses a generative model-based question-answering method for library reference services in educational settings. The method includes: acquiring and parsing reference request text to obtain structured parsing results containing clues such as task type and integrity intent; constructing a learner profile by combining a set of course policy parameters, calculating a comprehensive risk score, and determining the risk level; subsequently determining the highest permissible output level based on the risk level, course policy, and task completion rate; constructing a constraint-based inference input and generating candidate responses; finally, obtaining a score through a submitability discrimination model; if the score meets the standard, rewriting the candidate responses to generate compliant responses. This method achieves a balance between compliance and pedagogical relevance, effectively mitigating the risk of academic misconduct and improving the quality of reference services.

The application discloses a kind of coupling environmental factor and body surface spectral characteristic's early disease warning method of prawn, belong to the water product cultivation monitoring technical field based on machine learning;First, the group hyperspectral image of prawn group is obtained in actual cultivation environment, and multi-dimensional environmental factors are collected synchronously, and the basic data set for model training is constructed.Afterwards, a small amount of labeled group hyperspectral sample is used to fine-tune the designed generative enhancement network, so that it is adapted to the distribution of underwater hyperspectral imaging features, and then generates diversified virtual group spectral data.Subsequently, the original data and the generated expansion data are jointly input into the designed group-level disease warning model for training, to realize the discrimination of early disease risk of prawn group.After model training is completed, it is deployed in the online monitoring system of cultivation pond, realizes the real-time input of group spectrum and environmental factors, dynamic inference and risk warning output, to provide continuous, low-interference early disease warning support for cultivation management.

This invention provides a method and system for rapid detection of emotions induced by VR gory scenes based on electroencephalography (EEG), belonging to the field of EEG emotion detection technology. Subjects wear virtual reality devices connected to an 8-channel EEG acquisition device to obtain raw EEG signals with scene time stamps. Time-windowed EEG data is obtained through preprocessing and time window segmentation. Low-frequency energy features are calculated in the 1-10Hz frequency band, and phase coupling features between EEG channels are extracted based on a 2Hz narrowband segmentation. The coupling relationship is distinguished into contralateral long-distance and ipsilateral short-distance coupling based on scalp spatial location, and their intensity features are statistically analyzed. The low-frequency energy features and the long- and short-distance coupling strengths are constructed as composite discriminative features and input into a discriminative model to achieve rapid detection of gory scenes, improving the timeliness, objectivity, and accuracy of emotion recognition.

The application discloses a processing method and device of a reasoning task, a storage medium and an electronic device. It relates to the field of financial technology. The method comprises the following steps: receiving a target reasoning task input by a target object; preprocessing the target reasoning task to obtain a vector representation corresponding to the target reasoning task; performing reasoning processing on the vector representation by using a target reasoning model to obtain a reasoning result corresponding to the target reasoning task, wherein the target reasoning model is obtained by performing generative adversarial reinforcement learning training on an initial reasoning model and an initial discrimination model according to a sample reasoning task set. Through the application, the problem of low reasoning accuracy of the reasoning model in the related art when processing the reasoning task is solved.

The application discloses a deep learning signal denoising method based on a condition-guided diffusion model, relates to the fields of artificial intelligence deep learning and signal processing technology, and core deep learning technology is used to realize signal denoising. The method comprises the following steps: defining a forward diffusion process and key parameters, laying a theoretical foundation for deep learning model training and reasoning; constructing a condition-guided time series U-Net deep learning model, predicting noise to be removed according to a noisy state, a time step and an original noisy signal; iteratively optimizing and training the model by minimizing L1 loss of prediction and real noise; and relying on the trained model, performing reverse denoising through iterative deep learning reasoning to recover a pure signal from the noisy signal. The application reconstructs denoising as a diffusion model generation process, introduces a condition-guided mechanism, solves the oversmoothing problem of traditional and conventional deep learning discriminative models, realizes high-fidelity recovery of signal details, and has strong robustness to complex noise with non-stationary and low signal-to-noise ratio.

The present application relates to the technical field of target intention discrimination, and provides a group target intention discrimination method and device based on a combination strategy deep learning, wherein the overall structure and processing flow of a discrimination model are redesigned, cluster division processing is additionally added, a discrimination flow from a local to a whole is realized, the model can process a scene where multiple targets exist at the same time, and an intention discrimination network based on Patching technology and an attention mechanism is designed according to the characteristics of ground target time sequence data, the internal relationship between different characteristic dimensions and time dimensions is captured through base models of different Patching divisions, the base models are fused at a decision-making level through the attention mechanism, the discrimination results of different perspectives are taken into account, the discrimination accuracy and robustness of the model are improved, and the discrimination model can process a group target scene.

This application provides a text correction method, electronic device, and computer-readable storage medium, relating to the field of natural language processing. The method includes: performing parallel error correction processing on the target text using multiple lightweight error correction models to obtain candidate modification suggestions corresponding to each lightweight model; splitting the multiple candidate modification suggestions based on confidence assessment to obtain a first split and a second split; inputting the candidate modification suggestions from the first split into a large discriminative model for judgment to obtain a judgment result on whether to adopt the candidate modification suggestions; and performing text correction on the target text based on the candidate modification suggestions indicated by the judgment result and the candidate modification suggestions from the second split. This application achieves a balance between error correction cost, efficiency, and accuracy under resource-constrained conditions, meeting the real-time error correction needs of high-concurrency scenarios such as news and publishing.

This invention relates to the technical field of illumination environment state perception and feedback control, and in particular to a machine learning-based illumination feedback method, apparatus, device, and medium. The method includes inputting illumination intensity data and attitude data into a trained data validity discrimination model to obtain validity label information; filtering out invalid data from the illumination intensity data based on the validity label information to obtain valid illumination data; performing anomaly detection processing based on the illumination feature information and a preset illumination health threshold range to obtain illumination anomaly detection result information; sending the illumination anomaly warning information to a terminal; receiving parameter configuration instructions returned by the terminal; and updating the monitoring parameters at the acquisition end based on the parameter configuration instructions. This invention effectively solves the problems of insufficient reliability and poor feedback adaptability in existing illumination monitoring systems.

This application discloses a federated learning modeling optimization method, device, readable storage medium, and program product, applied to a first device. The federated learning modeling optimization method includes: acquiring first sample features generated by a feature extraction model and second sample features generated by a feature generation model to be trained; judging the first sample features and second sample features through a feature discrimination model to be trained, iteratively optimizing the feature generation model to be trained, and obtaining a feature generation model; sending the feature generation model and classification model to a second device, so that the second device can iteratively optimize a global feature generation model and a global classification model based on the feature generation models and classification models sent by each first device, so that the first device can construct a target feature extraction model and a target classification model based on the iteratively optimized global feature generation model and global feature classification model. This application solves the technical problem of the risk of leaking the data privacy of participating parties in federated learning methods.

The present application belongs to the field of marine geology and deep-sea mineral resources exploration technology, and relates to a method and system for identifying the genesis of oceanic iron-manganese deposits based on machine learning, which comprises the following steps: constructing a geochemical dataset of rare earth elements plus yttrium and preprocessing; constructing a derived feature dataset and standardizing, and dividing the training set and the test set; training multiple base learners in parallel to obtain posterior probability distribution; splicing the posterior probability distribution and the original features into meta-feature representation; constructing and training a Stacking integrated model; evaluating the Stacking performance through multiple random divisions, determining the optimal genesis identification model and training; determining the reliability of the genesis identification result through consistency index and confidence index; inputting new sample data into the model, outputting the identification result and visualizing the result. The present application breaks through the traditional three-classification framework, realizes the automatic and rapid identification of the genesis type, and improves the accuracy and reliability of the identification of the genesis of oceanic iron-manganese deposits.

This invention belongs to the field of pattern recognition technology and relates to an ultrasonic guided wave pattern recognition method based on generative adversarial networks (GANs) to solve the problem of separation between defect discrimination and region localization in existing ultrasonic guided wave structures, making integrated pattern recognition difficult. The method includes: constructing an ultrasonic guided wave system, dividing the system into regions, and recording information about the regions corresponding to damage; preprocessing the ultrasonic guided wave signals to obtain corresponding model input samples and dividing the dataset; constructing a two-stage cascaded semi-supervised GAN and training the corresponding model to obtain a damage presence discrimination model and a defect region localization model; inputting the preprocessed samples into the corresponding models, ending pattern recognition when a healthy result is output, and outputting a damage result when the localization model outputs the region to which the defect belongs. This invention integrates defect and region localization, achieving coordinated pattern recognition and localization, while maintaining pattern recognition accuracy even in labeled sample scenarios.