354 results about "Positive sample" patented technology

The invention discloses a protein generation model optimization method based on deep learning, and relates to the technical field of protein generation. The optimization method comprises the following steps: generating a candidate protein electron density distribution diagram based on the structural characteristics of a target spot by adopting a pre-trained diffusion model; converting the candidate protein electron density distribution diagram into a corresponding amino acid sequence; determining a functional index value corresponding to the amino acid sequence, and constructing a feedback data set; setting a reward threshold value, and marking the feedback data set as a positive sample data set and a negative sample data set according to the reward threshold value; constructing a utility function taking the reward threshold as a reference point, and respectively calculating utility values of the positive sample data set and the negative sample data set; and performing iterative optimization on the diffusion model according to the utility value. By adopting the technology provided by the invention, the dependence on preference paired data of large-scale and high-quality protein sequences can be avoided, and the performance of the protein generation model can be effectively and continuously improved.

The invention provides a drug-disease association prediction method and system, computer equipment and a medium, and belongs to the technical field of computers. The method comprises the following steps: constructing a drug-protein-disease heterogeneous network, and extracting a plurality of element path sub-graphs; inputting the meta-path sub-graph into a multi-scale diffusion graph convolution module, executing learnable multi-step graph diffusion on the basis of graph convolution, synchronously capturing local adjacency and high-order topological information, and generating node embedding; and performing dynamic weighted fusion by utilizing meta-path attention to obtain unified representation. In order to relieve imbalance of positive and negative samples, implementing difficult negative sampling in the embedding space, and constructing a balance training set with the positive samples; medicine-disease features are spliced, a regularization XGBoost classifier is trained, and unknown correlation accurate prediction is achieved. By adopting the method, the drug-disease association prediction precision and efficiency are improved, multi-scale topology and priori knowledge are fused, and a powerful calculation tool is provided for drug relocation.

The invention discloses a multi-modal large language model training method and system, and relates to the technical field of multi-modal large model data processing, and the method comprises the following steps: training a first large model through querying a question set, a positive sample and a hard negative sample; inputting the plurality of test samples into the trained first large model, and generating a plurality of second answers based on a first preset prompt; sorting each test sample based on the plurality of second answers, and retrieving to obtain a previous candidate multi-modal document related to the test question; and training the second large model through the test problem and the corresponding previous multi-modal document. According to the method, the positive samples and the hard negative samples are jointly used for training, the model is forced to capture the fine-grained semantic boundary of correlation judgment through a contrast learning mechanism, the distinguishing capacity of the model for difficult samples is remarkably improved, and the mistaken arrangement phenomenon is avoided.

Embodiments described herein provide a method for training a neural network based language model (LM). The method includes receiving, via a data interface, a training dataset including pairs of user queries and ground-truth responses; generating, via the LM, a plurality of responses based on a query from the training dataset; identifying, from the plurality of responses, a first response having a first similarity metric value below a threshold, based on a similarity metric associated with a corresponding ground-truth response from the training dataset; training the LM using the first response as a negative sample and a second response as a positive sample such that the LM after training is more likely to generate the positive sample and less likely to generate the negative sample; receiving, via a user interface, a query; and generating a response to the query via the trained LM.

The invention relates to the field of network security and the deep learning field, in particular to a satellite internet-oriented Tor website fingerprint identification method, which is implemented by constructing a self-supervised contrast learning framework and is implemented by generating a multi-view sample; extracting a feature matrix of the time sequence granularity; generating a synthesized negative sample based on linear insertion, and constructing a contrast with the positive sample; and track representation is extracted, and joint optimization is carried out in combination with cross entropy loss and mask reconstruction loss. Compared with a traditional website fingerprint identification method, the robustness of the model to satellite inherent noise caused by the Doppler effect can be enhanced. According to the method, multi-scale semantic features are provided for the weak enhancement view and the strong enhancement view, the inter-class distinction degree is improved by synthesizing negative samples, and the misjudgment rate is reduced; the joint loss optimization framework enables the model to give consideration to global statistical characteristics and local burst characteristics, and under the conditions of strong dynamic topology and limited labeling of satellite internet constellations, the accuracy and adaptability of website fingerprint recognition are remarkably improved.

The invention discloses similar object group expansion model training and similar object group expansion methods and devices, and belongs to the technical field of similar object group expansion, and the training method comprises the steps: obtaining a training set and a target time point corresponding to each training sample in the training set; a positive sample in the training set is a seed user, a target time point corresponding to the positive sample represents a time point of the seed user meeting a preset behavior condition, and a target time point corresponding to the negative sample is consistent with the target time point corresponding to the positive sample in distribution; dividing the user behavior sequence of the training sample before the corresponding target time point into a plurality of subsequences by using different preset time intervals, respectively extracting sequence features from the plurality of subsequences, and generating user features of the training sample based on a splicing result of the sequence features; and training a preset model based on the user features of the training samples to obtain a trained similar object group expansion model. Therefore, the extraction efficiency of the long-term and short-term behavior characteristics can be optimized.

The invention discloses a dynamic tool selection system and method based on multi-stage training, and belongs to the field of natural language processing. The method comprises the following steps: acquiring a labeling data set in a specific vertical field through a high-confidence training sample construction module, preprocessing the labeling data set to obtain different types of complex problem sets, generating a positive sample tool and a negative sample tool for each complex problem, and performing reasoning to obtain a high-quality training sample set; a progressive training module is used for multi-stage training of a model, and the model comprises a supervised fine-tuning training stage, a comparative learning stage and a reinforcement learning stage; and performing multi-dimensional evaluation on the trained model by taking the accuracy rate, the recall rate and the F1 score as evaluation indexes through a multi-dimensional evaluation module. Through the synergistic effect of the modules, the question answering ability of the model in the vertical field is effectively improved, particularly, a remarkable breakthrough is made in the aspect of tool selection accuracy, and a reliable technical scheme is provided for application of an intelligent question answering system in the professional field.

The invention provides a large model training method and device, a function recommendation method and device, equipment and a medium, and relates to the technical field of artificial intelligence, in particular to the technical fields of natural language processing, deep learning, large language models, information recommendation, intelligent agents, intelligent agents and the like. According to the implementation scheme, first sample data are obtained, the first sample data comprise sample dialogue information and corresponding positive sample labels and negative sample labels, and the positive sample labels indicate target function recommendation results determined according to the sample dialogue information; the negative sample label is a function recommendation result which is inconsistent with the positive sample label and is obtained by inputting sample dialogue information into an initial large model to execute sampling operation; and training the initial large model based on the first sample data to obtain a target large model.

The present disclosure provides a training method and apparatus for an image-text matching model, a device and a storage medium. The method includes: acquiring a positive sample and a negative sample; where the positive sample includes text and an image, the text in the positive sample is used to describe content of the image in the positive sample; the negative sample includes text and an image, the text in the negative sample describes content that is inconsistent with content of the image in the negative sample; training the image-text matching model by using the acquired positive sample and the acquired negative sample based on a manner of contrastive learning; where the image-text matching model is used to predict, for an input image and input text, whether the input text is used to describe content of the input image.

The invention relates to the technical field of knowledge base generation, and particularly discloses an education knowledge base content generation method and system based on AI big data. The method comprises the following steps: comparing user answer content with answer information to obtain a matching degree, calibrating a positive sample and a negative sample to train an evaluation model, and determining a reference standard score according to a quantitative relationship between the positive sample and the negative sample; and screening the plurality of candidate knowledge contents based on the reference standard score and the tolerance threshold to calibrate the candidate knowledge contents as stable knowledge contents or to-be-corrected knowledge contents, and outputting the stable knowledge contents or the to-be-corrected knowledge contents. According to the method, the evaluation system based on the real answer data of the user and the evaluation standard reflecting the actual application effect of the knowledge content are constructed, and the evaluation standard is automatically calibrated according to data distribution, so that objective screening and continuous iteration of the knowledge content are realized, and the accuracy and reliability of the knowledge base content are improved.

The invention relates to a geological disaster prediction method and system based on an optimized PU-XGBoost model, and belongs to the technical field of geological disaster monitoring and risk assessment, and the method comprises the steps: obtaining multi-source data including InSAR deformation monitoring data and historical geological disaster point data, taking the historical geological disaster points and InSAR significant deformation points in the InSAR deformation monitoring data as a positive sample set P, and carrying out the prediction of the InSAR significant deformation points in the historical geological disaster points; a PU-Learning strategy of a Spy sample is introduced to construct a reliable negative sample system, an XGBoost hyper-parameter is optimized and searched in combination with introduction of a BTO optimization algorithm, an optimal hyper-parameter combination is obtained, and an optimal model is obtained through training of a positive sample set P and a negative sample set RN; and performing geological disaster susceptibility prediction and hierarchical mapping on the research area based on the optimal model, thereby realizing high-precision and high-reliability evaluation of the geological disaster susceptibility of the research area.

The application discloses a resource consumption abnormal object identification method and related device, which is applied to the field of artificial intelligence. By obtaining unlabeled samples and positive samples, determining target feature dimensions, inputting the unlabeled samples and the positive samples into a semi-supervised learning framework to iteratively train a classification model, obtaining multiple identification feature values corresponding to the unlabeled samples output by the classification model in the iterative training process, and determining resource consumption abnormal objects, the resource consumption abnormal object identification process under a small amount of positive samples is realized. In the training process, the positive samples in the unlabeled samples are continuously mined heuristically for supplementation and added to the next round of iteration, effectively solving the sample imbalance problem in the identification scene and improving the accuracy of resource consumption abnormal object identification.

The embodiment of the invention provides a training set generation method, a model training method and a distribution duration prediction method. The method relates to the field of artificial intelligence, and comprises the following steps: based on historical delivery data corresponding to each historical version of a target model and first delivery data corresponding to a current version, respectively screening positive sample data corresponding to each historical sample in the historical delivery data from the first delivery data; based on the positive sample data corresponding to each historical sample, screening a target historical sample from each historical sample; and generating a target training set based on the target historical sample and the first distribution data. According to the embodiment of the invention, the historical samples supported by sufficient positive samples are screened to enrich the training set, the generated target training set can provide a stable distribution rule of cross-version difference, the richness and data quality of the training set are improved, and the problem of difficulty in effectively introducing long-period historical data in related technologies is solved. And thus, the technical problem of insufficient richness of the training set is solved.

The application discloses a CSI-based position-independent human activity continuous learning recognition method, which comprises the following steps: 1, collecting CSI action sample data; 2, pre-processing the CSI action sample data; 3, constructing positive samples by randomly scaling the pre-processed samples in the time dimension; 4, constructing a multivariate time graph neural network and extracting CSI action sample features; 5, calculating the similarity between the sample feature values and the positive samples and the feature values of the remaining samples, obtaining a comparison loss, and optimizing the feature extraction network; 6, freezing the feature extraction network, sending the features obtained from the input samples into a classifier for training to obtain a classification model. When the application continuously learns new action categories, the user does not need to retrain the feature extraction network, and the new and old action recognition in any position in the room can be realized by providing limited position new category samples to train the classifier, and the practicability is relatively high.

This invention relates to the field of agricultural intelligent equipment and machine vision technology, and discloses a method for weed detection and growth point localization based on shared features. The method constructs a collaborative dataset containing weed detection labels and growth point labels, utilizes a shared feature extraction network to extract multi-scale features, and sets detection task branches and growth point localization branches in a unified feature space to achieve joint modeling of weed detection and growth point localization. Then, adaptive positive sample selection is performed based on the joint cost between candidate sample points and labeled growth points, and joint training is conducted using detection loss, growth point localization classification loss, and growth point localization regression loss. During the inference stage, the method outputs weed detection results and growth point localization results, and generates target point information. This method can reduce redundant calculations and error cascading in the multi-stage processing, and improve the accuracy of growth point localization and the stability of target point output in complex farmland scenarios.

The invention discloses a sequence recommendation method fusing time interval and comparative learning, and provides a sequence recommendation framework TiDuoRec fusing time interval perception and comparative learning in order to solve the core problem that an existing sequence recommendation model generally ignores interaction time interval information and is difficult to accurately capture timeliness differences of user intentions. According to the framework, a relative time interval embedding mechanism is introduced in a Transform architecture in a breakthrough mode, original time differences are converted into learnable discrete embedding through normalization processing and threshold truncation, and a self-attention module can model three-dimensional correlation of project semantics, absolute positions and relative time distances at the same time; on the basis, model-level Dropout enhancement and target object driven supervised positive sample construction are combined, a comparative learning regularization module is designed, the problem of representation degradation of a high-dimensional embedding space is effectively relieved, and the discrimination and generalization ability of sequence representation are remarkably improved.

This specification provides a method, device, medium, and program product for training a generative model and retrieving goods. The method includes: acquiring first sample data including first sample goods information; generating multiple first sample query messages from an initial generative model based on the first sample goods information; determining a first positive sample and a first negative sample from the multiple first sample query messages; wherein the correlation between the first positive sample and the first sample goods information is higher than the correlation between the first negative sample and the first sample goods information, and the number of historical queries corresponding to the first positive sample is greater than the number of historical queries corresponding to the first negative sample; determining a first probability that the initial generative model outputs the first positive sample based on the first sample goods information and a second probability that the initial generative model outputs the first negative sample based on the first sample goods information, and training the initial generative model with the goal of increasing the first probability and decreasing the second probability to obtain a target generative model.

The invention discloses a method for realizing lightweight open world target detection, which comprises the following steps of: generating a prediction frame set for an input image through RPN, and screening out prediction frames of which the intersection-to-union ratio (IoU) meets a predetermined requirement from the prediction frame set to form a potential unknown target candidate set; adopting a density clustering algorithm to screen out a spatially dense candidate region cluster from the center coordinates of the prediction frame in the potential unknown target candidate set, and filtering sparsely distributed background noise to obtain a filtered candidate set; and calculating an IoU matrix of the filtered candidate set and the bounding box set, dividing unknown target sample types according to the IoU matrix, and determining unknown target positive samples, negative samples and neglected samples. The method has the resource adaptability and full-autonomous and high-reliability open world perception capability of the embedded edge device, and provides a practical and efficient open world target detection solution for large-scale landing of scenes such as intelligent cities, industrial monitoring and intelligent traffic.

The embodiment of the application discloses a contrast self-supervised learning training method and system, the method comprises the following steps: a sample data set is calculated with a first data enhancement operator and a data enhancement operator to obtain a data enhancement sample data set; the data enhancement sample data set is input into a first model and a second model to obtain a positive sample feature set and a negative sample feature set; a positive-negative sample combined feature set is obtained according to the original image corresponding to the negative sample feature set and the positive sample feature set; a loss function is calculated according to the positive-negative sample combined feature set; the gradient is calculated based on the loss function and the parameters of the first model are updated, the parameters of the second model are updated based on the updated parameters of the first model, and the cycle is repeated until the loss function meets the set condition. The designed loss function and data enhancement operator are suitable for various contrast self-supervised learning methods and have better generalization and robustness.

The application relates to a kind of open set transmitting source identification method, it is related to wireless communication and signal processing technical field, solve the problem such as existing method under open set scene, it is difficult to accurately exclude unknown device, classification accuracy and robustness are insufficient etc..The method only retains Q component by cutting to fixed length IQ signal, and extracts the absolute value and spectrum amplitude of the Q component, and is fused into three-channel pseudo-image as neural network input.Classification branch is optimized for cross entropy for known training equipment, to realize closed set classification;Contrast branch simultaneously utilizes known and unknown training samples to construct positive samples, known negative samples and open set negative sample mask, reduces the similarity of unknown samples through dynamic suppression boundary mechanism and calculates contrast loss.Furthermore, according to the dynamic adjustment of classification loss weight, the convergence of known category and the dispersion of unknown samples are realized.

The application discloses an industrial defect detection method, device, equipment and readable storage medium, comprising the steps of: acquiring a historical detection model of an original product, a historical training set corresponding to the historical detection model, and a positive sample set of a product to be detected; generating a target training set according to the positive sample set and the historical training set; establishing a comparative detection model, and training the comparative detection model according to the target training set to obtain a plurality of trained detection models; generating a target detection model according to a preset update strategy, the historical detection model and the trained detection model; and detecting the product to be detected by using the target detection model. The application improves the defect detection accuracy of a new product when a product is rapidly changed.

The invention relates to a fool-proof detection method and device based on three classifications, and the method comprises the steps: collecting a plurality of fool-proof region images, and constructing a training data set in a classified manner, which comprises a positive sample, a negative sample and a background data set; and based on the training data set, constructing and training a three-classification image classification model taking the fool-proof structure region image as input and taking the positive sample, the negative sample or the background as output, collecting a current fool-proof region image, and inputting the current fool-proof region image into the trained three-classification image classification model to obtain an image category. The problem that background interference misjudgment easily occurs in the prior art is solved.

The application discloses a cross-natural language code retrieval model training method, a cross-natural language code retrieval method, a device, equipment and a medium, and relates to the technical field of artificial intelligence and software engineering. The cross-natural language code retrieval model training method comprises the following steps: obtaining an original corpus database, and constructing training data according to the original corpus database; performing confusion and inversion on main language codes to obtain main language code samples, wherein the main language code samples comprise main language code positive samples and main language code negative samples; and training an initial model through a gradient inversion layer according to the training data and the main language code samples to obtain a target model. According to the application, the natural language-specific "fingerprint" features in the codes can be removed, the embedding space alignment direction can be unified, the sampling distribution deviation in the training process can be reduced, and the consistency and generalization capability of cross-language code retrieval can be improved.

An ensemble learning based method is for a binary classification on an imbalanced dataset. The imbalanced dataset has a minority class comprising positive samples and a majority class comprising negative samples. The method includes: generatively oversampling the imbalanced dataset by synthetically generating minority class examples, thereby generating a generated dataset; using the generated dataset to generate subsamples, and learning a base classifier on each of the subsamples to determine a plurality of base classifiers; and learning a weighted majority vote classifier by combining outputs of the base classifiers. Each of the base classifiers is assigned a weight in such a way that a diversity between the base classifiers on the positive samples is minimized.

In order to further study the performance of underwater target recognition task by contrastive self-supervised feature learning method, a kind of underwater target recognition method based on double-channel self-supervised acoustic feature learning is proposed. It includes the following steps: (1) a double-channel self-attention audio encoder model is proposed; (2) a double-channel self-attention audio encoder model with dynamic positive sample storage is proposed; (3) the underwater target recognition method based on the double-channel self-attention audio encoder model with dynamic positive sample storage is completed, the spectral feature of the double-channel self-attention audio encoder with dynamic positive sample storage is extracted, and the multi-layer perception machine model and the multi-classification logistic regression model are used to complete the underwater target recognition task. The underwater target recognition method based on double-channel self-supervised acoustic feature learning has good recognition accuracy and convergence speed, can effectively recognize underwater targets in noisy environment, and has strong robustness.

The application discloses an infrared small target detection network training method and device, equipment and a storage medium, and relates to the technical field of target detection. The method comprises the following steps: based on the number of positive samples and a preset proportion coefficient, difficult negative sample sets and easy negative sample sets are selected from an initial negative sample set; the preset proportion coefficient is a coefficient that is preset to improve the balance degree of the number of positive and negative samples; a weight matrix for controlling samples in a back propagation process is generated according to the difficult negative sample sets and the easy negative sample sets; a difficult point mining loss function is determined based on the weight matrix, and the infrared small target detection network is point supervised and trained by using the difficult point mining loss function. Through the above scheme, the difficult negative sample sets and the easy negative sample sets are determined based on the number of positive samples and the preset proportion coefficient, the unbalanced situation of the number of positive and negative samples in the target detection network training process is prevented, the balance degree of the number of positive and negative samples is improved, and therefore the training effect of the target detection network is improved.

This application provides a model training method, training system, and related equipment. The method includes the following steps: acquiring positive sample data, wherein the positive sample data includes input samples and positive labels, the input samples include questions, and the positive labels include the answers corresponding to the questions; inputting the input samples into a first large model to be fine-tuned, obtaining the first negative label corresponding to the input sample; determining the first score of the first negative label based on the relevance of the first negative label to the question and the quality of the response to the first negative label based on a scoring model; inputting the input samples into the first large model; and fine-tuning the first large model based on a loss function to obtain a second large model. The loss function is used to guide the first large model to improve the prediction probability of positive labels and the prediction probability of negative labels with scores greater than a threshold, so that the model can not only learn standard answers that meet user expectations, but also generate more diverse answers, thereby improving the fine-tuning effect of the large model.

The application provides a phase precision debugging method, which comprises the following steps: firstly, processing a debugging board and a positive sample plate for each LTCC combined substrate, obtaining initial phase precision of a beam forming matrix through the debugging board, and then designing a surface phase line of the positive sample plate according to the initial phase precision to realize coarse adjustment of the phase precision; secondly, printing the corrected surface phase line on the surface of the positive sample plate; and finally, cutting the debugging phase line on the surface of the positive sample plate to change the path length of the signal, and then to improve the phase precision and realize fine adjustment of the phase precision. As can be seen, the application can concentrate the phase errors of all components in the system by printing the phase line on the surface of the positive sample plate, and then eliminate or reduce the phase errors by debugging the surface printed phase line, so that the operation is simple and convenient, the reliability of the system is not affected, and other devices and circuits are not needed, which can ensure the overall structure of the system unchanged and solve the phase precision problem through debugging.

The application discloses a kind of view extraction method, electronic equipment and computer readable storage medium.Therein, method includes obtaining the extraction result after the information extraction model in the view extraction device is handled to positive sample training data;According to the extraction result, generate the training negative sample of the aspect word classification model, the sentiment classification model and the category and sentiment pairing identification model in the view extraction device;According to the positive sample training data and the generated training negative sample, the aspect word classification model, the sentiment classification model and the category and sentiment pairing identification model in the view extraction device are trained;Utilize the view extraction device after training to the text to be extracted and process, obtain the view information in the text to be extracted.The scheme provided in the application can improve the accuracy of view extraction.

The invention belongs to the field of big data, and particularly relates to a comparative learning sample construction and prediction method and system for flight arrival time prediction. The method comprises the following steps: acquiring trajectory data and meteorological observation data of each flight of an airport and preprocessing the trajectory data and the meteorological observation data; and constructing a time sequence sample by using the preprocessed time axis aligned trajectory data and flight feature data of the meteorological observation data. Performing sample phase division according to track features, meteorological features and residual arrival time in the end timestamps of the time sequence samples, and enabling the time sequence samples in the same phase to be weak positive samples. Calculating a feature combination distance between the time sequence samples in the same phase; and screening a specified number of nearest neighbors as strong positive samples of the nearest neighbors. And using the data set containing the positive sample label to carry out mixed contrast learning training on the flight prediction model. According to the method, the problems of insufficient prediction precision and generalization of the existing flight arrival time prediction are solved.