117 results about "Small data sets" patented technology

The invention relates to a voltage sag source positioning method based on measured data. The method comprises the following steps: S1, acquiring monitoring data; s2, performing sag-fault matching based on time sequence matching; the voltage sag of one monitoring point is matched with the same bus fault and the external fault according to the time sequence of events, and the time difference between each pair of voltage sag and fault in the matching result is calculated; s3, dynamically adjusting a window to calculate a quantile detection abnormal value; s4, performing secondary matching and iteration termination based on the reduced data set; and S5, repeating the steps S2-S4 for other monitoring points to obtain all matching data in the region, and realizing accurate positioning of the voltage sag source. According to the method, accurate and efficient fault positioning and matching can be carried out on the voltage sag event.

The invention discloses a potential landslide identification method and system based on a graph neural network, and the method comprises the steps: constructing a multi-level feature fusion graph attention network model which comprises a node layer, a subnet layer and a global layer, and employs a dual-branch encoder-decoder structure; the method comprises the following steps of: pre-training a dual-branch encoder-decoder structure model until convergence by adopting a combined data set of a data set 1 of a large sample and a data set 2 of a small sample to form a pre-training model; based on auxiliary set node features obtained by inputting the data set 1 and the data set 2 into the pre-training model, training and verifying the graph attention network model until model convergence training is completed; and inputting the pictures needing to be identified in the data set 2 into the trained picture attention network model to obtain a prediction result of the potential landslide. Through the technical scheme of the invention, effective extraction of the target features in the small data set 2 is realized, and the classification performance and generalization ability of the model are improved.

The invention discloses an end-to-end intelligent direction of arrival estimation method based on subarray sampling, and belongs to the technical field of wireless communication. A sub-array sampling technology is adopted to replace traditional full-array sampling, and signal processing is completed by using fewer radio frequency links, so that the hardware cost is reduced; an end-to-end learning framework is adopted, so that a network can automatically capture features and rules according to a data set, and excessive dependence on a model is avoided; data features are preliminarily captured through a deep neural network, a pre-estimated value is given, then an iteration process of a maximum likelihood estimation method is converted into a deep expansion network, initialization is performed by using output of the deep neural network, and the effect of the method is improved while convergence is accelerated; a large number of data sets are generated through a sub-array sampling signal mathematical model, the network is trained, and the problem that overfitting is prone to occurring under a small data set is avoided. The method is suitable for the technical field of wireless communication, and realizes direction of arrival estimation with lower complexity and higher accuracy in a low-snapshot scene.

The invention relates to a sensitive data identification and desensitization method and system based on an AI large model, and belongs to the technical field of data security processing and artificial intelligence. The method comprises the following steps: receiving a multi-source data processing request, establishing a sensitive data identification framework, analyzing elements of sensitive data, and extracting a sensitive data feature vector through a natural language processing algorithm; classifying the multi-source data, dynamically adjusting an identification threshold value of the sensitive data identification framework through a sensitive data feature vector, and generating a self-adaptive sensitive data identification strategy; establishing a terminal desensitization verification model, performing desensitization verification on the identified sensitive data, performing mask replacement or generalization processing according to privacy requirements, and outputting a desensitization retry processing scheme; the sensitive data is subjected to block desensitization, a large data set is divided into small data sets for independent desensitization, replaceable entities in information are removed according to data attributes, and the desensitized data is further subjected to secondary desensitization.

The invention relates to the technical field of radar signal processing, in particular to a ground penetrating radar reinforcing steel bar clutter suppression method based on physical constraints, and mainly solves the problem that a data set is difficult to obtain in an existing ground penetrating radar reinforcing steel bar clutter removal method based on deep learning. The method is an improved method based on the CUT network, the CUT network structure and a comparative learning mechanism determine that the requirement of the network for the data size of a data set is low, a waveform smoothness constraint is added on this basis, the clutter removal effect and generalization ability are improved by introducing physical prior, the physical constraint serves as a regularization item, and the regularization efficiency is improved. The problem that a CUT network is prone to model collapse under a small data set is solved. Finally, the improved model is compared with other models through different evaluation indexes, and the result shows that the improved model has more advantages.

The invention discloses a method and device for improving the fine tuning effect of a large model based on an iterative data enhancement strategy. The method comprises the following steps: manually generating a batch of seed data with different distributions according to requirements; enhancing artificially generated seed data based on a rule or a large model; checking the quality of the enhanced data; finely adjusting the model by using the checked data, and reasoning training data by using the trained model; performing comprehensive evaluation on the reasoning result in various aspects; whether the evaluated indexes meet requirements or not is judged; if the error sample does not meet the requirement, enhancing the error sample again, and iterating repeatedly; and if the evaluation result meets the requirement, ending the whole process. According to the method, the threshold of fine adjustment of the large model under the small data set is remarkably reduced, and the model is excellent in performance in a data scarce scene by generating the targeted synthetic data.

The invention discloses a dielectric constant prediction method and system and a network training method and system. The method comprises the following steps: acquiring material characteristics of a to-be-tested material; constructing a plurality of descriptors according to the material characteristics of the to-be-tested material; screening the plurality of descriptors through a genetic algorithm, determining M main descriptors, and determining the descriptors except the main descriptors as other descriptors; and inputting the main descriptor and other descriptors into a preset dielectric constant prediction model, and outputting dielectric constant prediction data. The dielectric constant of the microwave high-frequency material is efficiently and accurately predicted through the machine learning technology. Compared with a traditional neural network model and descriptor selection method, the prediction precision, the model efficiency and the generalization ability are remarkably improved, and the problem that high-dimensional descriptors are not matched with small data sets is solved.

The invention discloses a method for predicting the tensile strength of additive manufacturing 316L stainless steel based on machine learning. The method comprises the following steps: firstly, establishing a machine learning data set which takes four process parameters of laser power, scanning speed, scanning spacing and powder laying layer thickness as input and takes the tensile strength of SLM (Selective Laser Melting) molding 316L stainless steel as output; then preprocessing and normalizing data in the machine learning data set; constructing and training a tensile strength prediction model composed of a convolutional neural network and a random forest regression device; and finally, obtaining a tensile strength prediction result by using the trained tensile strength prediction model. According to the method, high-precision prediction of small data set samples can be realized.

The first token prediction of a large language model is bottlenecked by compute and second token predictions onwards are bottlenecked by memory bandwidth. Inferences can be made more efficient through activation sparsity. An activation tensor is pruned using an importance threshold value. The mode of the activation tensor is centered in a lossless manner using an estimated mode value to improve activation sparsity further. Pruning and mode-centering mechanisms can be inserted into a neural network strategically and post-training to implement sparsification. A two-stage greedy grid search algorithm is implemented to determine the calibrated importance threshold values of various pruners and the estimated mode values using a small dataset. A modified neural network with pruning and lossless mode-centering can be deployed onto hardware.

The invention discloses a two-stage sparse method based on a small sample rapid sparse large-scale language model, and the method comprises the following steps: S1, mask initialization: based on a plurality of importance indexes for measuring parameters, employing a plurality of mask initialization methods to initialize a mask of a language model; s2, reordering the masks: reordering the masks based on the small sample calibration set and the straight-through estimator; s3, mask adjustment: based on the small sample calibration set and the reordered mask matrix in the step S2, taking an original model as a teacher model, taking a sparse model as a student model, realizing a block-level self-distillation process of the large language model in a time dimension before and after sparsification, and obtaining the sparse model; the method consumes less computing power, uses a smaller data set, occupies less GPU memory while realizing considerable performance, does not influence the reasoning speed of the sparse model, and has remarkable practical value and application prospect.

The invention provides a semi-supervised surface defect instance segmentation method based on pseudo label enhancement. The method comprises the following steps: acquiring a surface defect image data set; training an initial instance segmentation network by using a small amount of labeled defect image data; inputting the marked defect image data into a student instance segmentation network to train the student instance segmentation network; inputting the defect image data without labels into the teacher instance segmentation network and the student instance segmentation network, and respectively obtaining a mask and features of each defect instance; inputting the data into a cross-supervision comparative learning module and a general distribution fusion module; constructing a contrast learning loss function, obtaining a fused pseudo tag, and training a student instance segmentation network by using the pseudo tag; and updating the parameters of the teacher instance segmentation network by using the parameters of the student instance segmentation network. According to the method, challenges of small data set scale and defect boundary fuzziness can be handled, complex category feature distribution can be estimated through the general distribution fusion module, and the quality of mask false labels from a teacher model is improved.

The invention provides a reverse design method of a composite function metasurface based on transfer learning, relates to the field of metamaterial design and optimization, and solves the problem of low efficiency of a traditional design method of a metasurface with a complex structure. Comprising the following steps: generating a data set containing various metasurface structure parameters and corresponding spectral responses; building a deep prediction neural network taking the unit structure parameters as input and the spectral response as output; the trained network is combined with a genetic particle swarm optimization algorithm, metasurface phase distribution, transmissivity and other requirements determined in advance are used as objective functions, and optimal parameters are searched through multiple iterations; and migrating universal features learned on the large-scale data set by the source domain network model through a parameter fine tuning method, applying the universal features to a new small data set target domain, repeating the previous process until all parameters are determined, and completing the design. According to the method, the complex metasurface design efficiency can be improved, the neural network performance is ensured, meanwhile, the requirement of the deep learning method for the data collective quantity is reduced, the robustness of the model is improved, and a low-cost practical scheme is provided for metasurface inverse design.

The present invention relates to the field of organic polymers. Specifically disclosed is a method for predicting the properties of grafted polypropylene for a high-voltage power cable. In the method of the present invention, a property prediction model for grafted polypropylene is constructed on the basis of a multi-layer perceptron model and a random forest model. The model solves the defect that a conventional machine learning model mainly relies on a large amount of data and is not suitable for a small data set. Experimental data is used to quantitatively establish a relationship between a grafted structure and mechanical properties, the amount of data required is small, the data is easy to obtain, and the prediction accuracy is high, thereby providing a reliable grafted group for a grafted polypropylene material used for a high-voltage power cable.

By accurately predicting the industrial aging process (IAP), such as the slow deactivation of catalysts in chemical plants, maintenance events can be scheduled further in advance, ensuring the cost - effectiveness and reliable operation of the plant. So far, these degradation progressions have typically been described by mechanical models or simple empirical prediction models. To accurately predict the IAP, data - driven models are proposed, comparing some traditional stateless models (linear and kernel ridge regression, and feed - forward neural networks) with more complex state - recursive neural networks (echo state network and long - short - term memory network). In addition, variants of the stateful models are discussed. In particular, stateful models that use mechanical pre - knowledge about the degradation dynamics (hybrid models). Stateful models and their variants may be more suitable for generating near - perfect predictions when trained on a sufficiently large dataset, while hybrid models may be more suitable for better generalization in the case of smaller datasets under changing conditions.

The present invention discloses a classification method and system for the invasive evolution of lung adenocarcinoma based on a state space network, which relates to a classification method for lung adenocarcinoma infiltration. The purpose is to solve the technical problems of poor classification accuracy and low classification efficiency of the IA type classification of GGO imaging features in the prior art. The constructed GGO invasiveness classification model includes a teacher encoder, a student encoder, and a fine-tuning encoder, and the teacher encoder, the student encoder, and the fine-tuning encoder are all SSM state space networks; during training, a large sample data set without GGO classification labels is used to train the teacher encoder and the student encoder, and the output of the teacher encoder is used as the pseudo-label available to the student encoder, and the parameter update of the teacher encoder is guided by the exponential moving average of the output of the student encoder; a small data set with GGO classification labels is used to fine-tune and train the fine-tuning encoder. It can effectively improve the classification accuracy and classification efficiency of the IA type of GGO imaging features.

The invention discloses an automatic algorithm feature processing method and system, and relates to the technical field of computer data processing. Data are collected and cleaned, then feature parameters are extracted, test cases are established based on the feature parameters, then a model is trained according to the feature parameters, test cases of online data feature parameters are established, and finally the test cases are input into a target server for testing, so that the whole test process is automatically carried out, and the test efficiency is effectively improved; and the testing process is suitable for artificial intelligence algorithm testing of various processing classes, and the universality is high. According to the automatic algorithm feature processing method and system, machine learning algorithms and deep learning algorithms are integrated, uniform file configuration and configurability are supported, when the data set reaches a certain amount, the deep learning strategy can be triggered, and the defect that the small data set modeling effect is poor is overcome.

The invention discloses a general framework construction method for small data set color image classification. The general framework construction method specifically comprises the steps of 1, preprocessing a sample image set; step 2, performing channel decomposition processing on images in the sample image set obtained after preprocessing in the step 1; step 3, performing Haar transformation on the feature map of each channel obtained in the step 2; step 4, cascading the obtained feature maps into one feature map; 5, a BLOCK module is built, the feature map obtained in the step 4 is partitioned and flattened into column vectors, and then the column vectors are input into the BLOCK module for Stage1 stage processing; step 6, performing Stage2 stage processing; and step 7, performing Stage3 stage processing. According to the method, a large-scale data set with diversified categories and high-quality annotation is not needed, the classification precision of the obtained color image is higher than that of an existing Vision Transform frame, and the precision of the image classification method based on the CNN frame is achieved.

The application belongs to the field of data enhancement application, and particularly relates to a GAN model improvement method for small data set scenarios. The multi-discriminator fusion module is composed of multiple discriminators and a discriminator fusion submodule. Each discriminator is a network composed of multiple convolutional layers and fully connected layers, which receives a batch of pictures in a downsampling manner and converts the pictures into a tensor. The tensor has four dimensions, namely the number of batches of images batch, the number of channels c, the width w of each image and the height h of each image. The multi-discriminator fusion submodule is responsible for coordinating multiple discriminators, receiving the return values of the multiple discriminators, weighting the return values and then transmitting the weighted return values to the generator for back propagation. In order to avoid the occurrence of sample penetration, the weight of the discriminator responsible for enhancement is reduced. In this way, the probability of overfitting of the discriminator can be reduced, and the possibility of sample penetration can be reduced.

An improved multi-target regression algorithm based on target specific features is combined with a non-dominated sorting genetic algorithm to construct a double-layer structure multi-target algorithm, so that the problems of data distortion possibly caused by data enhancement, poor feature engineering generalization ability, negative migration caused by large inter-domain difference and the like are avoided, and the method is suitable for large-scale popularization and application. By capturing the correlation among a plurality of output targets, a high-precision model for a high-dimensional small data set is constructed, and the machine learning assisted hot-rolled ultrahigh-toughness steel component and process design with low data set cost is realized.

The application belongs to the technical field of unmanned aerial vehicle image detection, and in particular to an unmanned aerial vehicle image detection method based on an improved RT-DETR model. The method comprises the following steps: selecting a public unmanned aerial vehicle image dataset, dividing the training set, constructing a small dataset, training and evaluating the improved RT-DETR model on the small dataset, and then implementing the model on the complete dataset to obtain the final data. The application enhances the information of small targets in the fusion process without increasing the number of model parameters. While introducing S2, a new fusion module mainly composed of DSConv is used to replace the fusion module of the benchmark model, which balances the parameter quantity of the model while improving the small target detection accuracy. Compared with the previous algorithm, the improved model has higher accuracy and smaller parameter quantity, and can provide protection for accurate detection of unmanned aerial vehicle images.

The application relates to the technical field of archive construction, and provides a text attribute relation extraction method based on a small data set pre-training model, a pre-training model based on an ERNIE structure is publicly fine-tuned to a relation extraction model in a business background through a public pre-training model UIE which can be started by fine-tuning only a small amount of labeled data, and on the basis of the model, Before_pipeline data processing before prediction and After_pipiline data processing after prediction are constructed in combination with business text features to improve result reliability, the problems of lacking a labeled data set, UIE violent sentence splitting leading to errors, and low extraction result accuracy of the UIE are solved, and the problem that an existing traditional relation extraction model depends on a large amount of high-quality data sets manually labeled is solved.

Methods and systems for ultrasound imaging are provided. In one example, a method includes receiving, with a wireless handheld probe assembly, ultrasound signals of a region of interest; generating, within the wireless handheld probe assembly, a plurality of received digital signals based on the received ultrasound signals; generating each of a larger data set and a smaller data set from the plurality of received digital signals; transmitting the smaller data set from the wireless handheld probe assembly to a hub via a lower bandwidth wireless connection; transmitting the larger data set from the wireless handheld probe assembly to the hub via a higher bandwidth wireless connection; generating, at the hub, a low resolution image from the smaller data set and a high resolution image from the larger data set; and transmitting the low resolution image from the hub to a first display and the high resolution image from the hub to an electronic device.

The invention discloses an industrial printed circuit board image defect detection method based on deep learning, which specifically designs a small data set expansion scheme to enhance data diversity. The method comprises the following steps: designing an early stage of a depth separable convolution reconstruction parameterization module to construct a backbone network, designing a non-adjacent window filling block improved multi-head self-attention module to construct a lightweight neural network Transform architecture to construct the backbone network, and designing a small target step length output improved module to construct a target detection head; a defect detection model capable of positioning an industrial printed circuit board image is designed, a small target detection strategy is designed, and the small target defect learning speed and precision of the model are improved; and a label duplicate removal scheme and an adaptive defect screening scheme are designed, so that the accuracy of model prediction is improved. According to the invention, the feature learning ability of an image with a small detection target can be improved, and the defect detection performance of an industrial printed circuit board image can be effectively improved.

The invention discloses a high-performance deep learning attack method for an asymmetric cryptographic system. The high-performance deep learning attack method comprises the following steps of: 1, encrypting an image by using a PTFT-based cryptographic system and an EMD-based cryptographic system respectively; 2, establishing plaintext-ciphertext pair data sets, and dividing each data set into a training set, a verification set and a test set which respectively comprise 600 pairs of plaintext-ciphertext data pairs, 200 pairs of plaintext-ciphertext data pairs and 200 pairs of plaintext-ciphertext data pairs; and a third step of designing a high-performance deep learning attack network RRU-NetPP, and training and debugging the network by using the established very small data set. And 4, attacking the ciphertext encrypted by the PTFT cryptosystem and the EMD cryptosystem by the established RRU-NetPP model, and retrieving the plaintext in real time with high quality. According to the method, the requirements on known plaintext-ciphertext data pairs can be remarkably reduced, the method has a few-sample learning capability, plaintexts can be retrieved from complex ciphertexts encrypted by different asymmetric cryptographic systems in real time in a high-quality manner, and the method has a strong attack capability.

The application discloses a kind of based on mechanism and data driven's sugarcane squeezing process process index prediction method, comprising the following steps: (1) multi-sensor field data acquisition and preprocessing, establish original data set;(2) establish sugarcane material elastic-plastic constitutive model;(3) establish sugarcane squeezing process porous medium control equation;(4) sugarcane squeezing process fluid-solid coupling model and its simulation, and establish new data set;(5) the establishment of physical guide neural network model;(6) the training details of physical guide neural network, obtain based on mechanism and data driven's sugarcane squeezing process process index prediction model.The method of the application is mainly used to predict the key process indexes such as sugarcane juice precipitation amount and sugarcane juice precipitation speed in the sugarcane squeezing production process.The cost of data acquisition in the production process is greatly reduced.High-dimensional system is characterized by small data set;The consistency of network can be greatly improved, and the generalization ability is enhanced.

The embodiment of the present invention discloses a method for predicting the strength of drug inhibitory efficacy based on a small data set, which uses adaptive data expansion, reconstructs feature space, and relabels labels as relative size relationships of inhibitory efficacy, thereby converting the predicted accurate numerical value of inhibitory efficacy into a direct prediction of the relative size of inhibitory efficacy, and converting the accurate value prediction regression task into a simpler classification task. At the same time, the adaptive feature space reconstruction preserves the relative size relationship of the physicochemical properties between compounds, and more importantly, significantly weakens the inherent similarity between compounds with similar inhibitory efficacy. Using the data set processed by the present invention, the machine learning model can more easily distinguish similar compounds, and thereby screen out samples with better inhibitory efficacy.

The application provides a semi-supervised surface defect instance segmentation method based on pseudo label enhancement, comprising the following steps: obtaining a surface defect image dataset; training an initial instance segmentation network with a small amount of labeled defect image data; inputting the labeled defect image data into a student instance segmentation network for training; inputting unlabeled defect image data into a teacher instance segmentation network and the student instance segmentation network to obtain the mask and features of each defect instance; then inputting the mask and features into a cross-supervised contrast learning module and a general distribution fusion module; constructing a contrast learning loss function, obtaining fused pseudo labels, training the student instance segmentation network with the pseudo labels, and updating the parameters of the teacher instance segmentation network using the parameters of the student instance segmentation network. The application can cope with the challenges of small data set size and fuzzy defect boundary, estimate complex class feature distribution through the general distribution fusion module, and improve the quality of the mask pseudo labels from the teacher model.

The present invention discloses a mobile target tracking method based on a fusion network of MHS and GRU, belongs to the technical field of mobile target tracking, and is used for mobile target tracking, including inputting a training set into a fusion network based on MHS and GRU for iterative training, inputting a verification set into a fusion network based on MHS and GRU after each iteration, returning to iterative training if the accuracy requirement is not met, and inputting a test set into a fusion network based on MHS and GRU if the accuracy requirement is met, obtaining a fusion gain FG, and inputting the local filter of each sensor together with the fusion gain FG into a mobile target tracking model to obtain a mobile target tracking result. The present invention automatically corrects and adapts to errors in the process of time registration, effectively reduces error accumulation, and achieves improvements in tracking accuracy, convergence speed, and stability; using a relatively compact RNN, relatively small data sets can be used for training.

The application discloses a kind of multi-dimensional dynamic model under the evaluation method of soil salinization to cultivated land productivity, it is related to the field technical field of agricultural resources and environment, and the multi-source data set of fusing remote sensing, ground monitoring, meteorological and farmer management data is first constructed in the scheme, and the minimum data set of soil physical, chemical, biological properties is screened after fusion preprocessing;Soil quality index (SQI) and cultivated land productivity index (CPI) are calculated again;Subsequently, the multi-dimensional dynamic evaluation model containing key salinization index screening, effect quantification, mechanism model and crop sub-model is constructed;Finally, the productivity influence dynamic update is realized in combination with real-time data of Internet of Things, and the optimal improvement technology mode is simulated and screened by structural equation model simulation management scene. The method is comprehensive in data coverage, accurate in evaluation, dynamic in model and targeted, can reveal the influence mechanism of salinization on cultivated land productivity, provide practical decision support, and help efficient and sustainable use of saline-alkali soil.

The invention provides a marine mammal individual identification method, which belongs to the field of computer image identification, is based on a convolutional neural network model, combines an attention mechanism, multi-scale feature processing and a depth separable convolution technology, and comprises four steps of data acquisition, data enhancement, model construction and model evaluation. Through the application of the method, the facial key features of the marine mammals can be accurately captured, the parameter quantity and the calculation demand of model operation are reduced, and efficient and accurate identification of the marine mammal individuals under the background of a small data set is realized; and the marine mammal individual population identification challenges of high cost, low efficiency, unfriendliness and the like in the traditional method are solved. Therefore, the marine mammal individual identification method has good marine mammal individual identification capability, and can be widely applied to the fields of marine species investigation monitoring and protection management.