298 results about "Sample selection" patented technology

The invention discloses an uncertainty and representativeness combined active learning method and system for medical image data annotation, and relates to the technical field of medical image data. The method comprises the following steps of: training a variational auto-encoder infoVAE on an image pool; m samples are randomly extracted from the image pool and labeled, an initial labeled image set is constructed, and a segmentation model is trained on the labeled image set; t rounds of active learning circulation are carried out, and in each round of t active learning circulation, the following steps are specifically executed: screening candidate samples based on a representative method; screening a final sample based on an uncertainty method; updating the labeled data set and the unlabeled data set; retraining the segmentation model on the updated labeled image set, and optimizing model parameters; and obtaining final model parameters. According to the method, by designing a sample selection strategy integrating uncertainty and representativeness, the global performance of the model and the difficult sample segmentation capability are improved.

The invention provides a large model named entity recognition method based on representative sample selection and context enhancement, which comprises a representative sample selection module, an entity knowledge construction module, a dynamic context selection module, a large model calling module and an iterative feedback optimization module, according to representative sample selection, samples with representativeness and information diversity are automatically selected from unlabeled data for labeling through a sample screening strategy based on clustering, entity description integration aims at each entity type, a plurality of high-quality instances are extracted from labeled samples, and standardized entity definition or description prompts are constructed. According to the dynamic context selection, for to-be-recognized text content, a context example most relevant to a target text is dynamically selected from a historical annotation sample or a description set through a semantic similarity retrieval mechanism to serve as auxiliary prompt input, and the adaptability and generalization ability of LLM in a complex or variable scene are improved.

Methods, systems, and apparatuses, including computer programs encoded on computer storage media, for receiving a query relating to a data item that includes multiple data item samples and processing the query and the data item to generate a response to the query. In particular, the described techniques include adaptively selecting a subset of the data item samples using a selection neural network conditioned on features of the data item samples and the query. Then processing the subset and query using a downstream task neural network to generate a response to the query. By adaptively selecting the subset of data item samples according to the query, the described techniques generate responses to queries that are more accurate and require less computation resources than would be the case using other techniques.

The invention discloses a single-point supervised directional target detection method based on adaptive sample allocation and a global-local context enhancement module, which comprises the following steps: inputting an original remote sensing image into a backbone network, extracting to obtain a multi-scale feature map set, inputting a feature map with the highest resolution into a global-local context module to obtain an enhanced feature, and extracting the enhanced feature from the global-local context module; inputting the enhanced features into a projection layer to generate a category probability graph; based on the category probability graph, training and optimizing the category probability graph to obtain a probability histogram set by adopting a sample distribution and optimization strategy combining adaptive positive sample selection, online difficult case mining and a focus loss dynamic weighting mechanism based on prediction confidence; and based on the full training set category probability histogram set, carrying out adaptive pseudo tag threshold calculation, and finally, realizing end-to-end joint training of pseudo tags and enhanced features through a prediction quality guide weighting strategy. According to the method, the target contour precision and the detection robustness are remarkably improved, and high-performance remote sensing image directional target detection is realized under a weak supervision condition.

The invention discloses a remote sensing image sample intelligent acquisition method based on image classification, and relates to the technical field of image acquisition, and the method comprises the following steps: carrying out the feature extraction of a remote sensing image, and obtaining an initial feature; performing classification uncertainty evaluation according to the initial features to obtain uncertainty indexes; performing sample representativeness calculation according to the uncertainty index to obtain a representative index; performing sample diversity analysis according to the representative indexes to obtain diversity indexes; performing comprehensive sample scoring according to the diversity indexes to obtain sample scoring data; and performing sample selection according to the sample scoring data to obtain a final collection sample. According to the method, uncertainty weighted items are introduced, representativeness is calculated in combination with similarity, global distribution and local feature differences are considered, reliability of representativeness calculation is improved, balance of samples on ground feature type distribution is improved, and stability and scientificity of sample screening are improved.

The invention provides an intention perception recommendation method based on knowledge graph and comparative learning fusion, relates to the technical field of artificial intelligence, and aims to optimize node representation by fusing knowledge graph and comparative learning technologies so as to improve the recommendation precision of a recommendation system for long-tail projects. Meanwhile, on the basis of data enhancement, positive sample selection in contrast learning is expanded in combination with user intentions, the ability of the model in the aspect of user intention modeling is further enhanced, and finally the accuracy and correlation of recommendation results are improved. The objective of the invention is to solve the problems of poor long-tail project representation quality, insufficient user intention modeling and limited recommendation precision caused by knowledge graph noise interference in the prior art.

A long-term continuous learning method based on task core memory management and consolidation aims to enable a model to sequentially learn from a large number of task sequences, new knowledge is obtained, information of previous learning is reserved, and the method is similar to a human learning mode. The method comprises the following steps: 1) task input and instruction fine tuning; 2) performing difference analysis on the model parameters of the current task and the previous task, identifying a task core memory unit, calculating an adaptive weight based on task prototype similarity, and dynamically updating the memory unit; 3) constructing an experience playback buffer area through a difficult sample selection strategy and a difference sample selection strategy; and 4) utilizing the joint loss function training model to keep the memory of the historical tasks while learning the new tasks. According to the method, the problem of disastrous forgetting in long-term continuous learning is mainly solved, and the performance of the model in a long-term sequence task is remarkably improved.

The invention discloses a multi-component catalyst active site prediction system and method fused with quantum embedding, and relates to the technical field of catalysis and material informatics, and the system comprises a structure and site enumeration module which generates candidate sites; the adaptive quantum embedding calculation module obtains key reaction microcosmic parameters; the unified site fingerprint and feature engineering module constructs and fuses standard site fingerprints; the physical consistency machine learning module predicts adsorption energy and other parameters and uncertainty thereof; the active learning and sample selection module selects a high-value sample optimization model; the microdynamics evaluation module calculates index values such as activity; and the multi-objective optimization and sorting module generates an optimization sorting list. According to the method, the unification of calculation precision and efficiency is realized, the problem of non-unification of locus characterization is solved, the model interpretability and extrapolation reliability are improved, and the comprehensive evaluation and optimization sorting of multi-target performance are completed.

The invention discloses a big language model knowledge extraction method and device based on difficulty perception, and aims to solve the problems that an existing knowledge extraction method is high in training cost and lack of pertinence in sample selection. According to the method, the difficulty of a distillation difficulty score evaluation sample is introduced, a difficult sample with learning value for a student model is identified, a distillation data set is dynamically adjusted in combination with a hierarchical data updating strategy, the difficult sample is preferentially reserved, a simple sample is removed, and meanwhile, data diversity is kept. Besides, a bidirectional difference loss function is provided, KL divergence and inverse KL divergence are combined, the optimization process is stabilized, more attention is paid to difficult samples, and gradient explosion or disappearance is avoided. Experimental results show that the performance of a student model is effectively improved, the method even exceeds a teacher model under some conditions, meanwhile, the training cost is remarkably reduced, and the method is suitable for task-independent instruction following and specific tasks and has wide application prospects.

The invention belongs to the field of food quality evaluation, and provides a delicate flavor synergistic effect evaluation method based on an electroencephalogram monitoring technology. Psychological or physiological factors such as subjective judgment, forced selection or scoring, individual umami cognition and umami sensitivity difference of sensory officers may affect the accuracy of sensory experiment results; the invention discloses methods such as umami sample selection and strength evaluation, electroencephalogram signal acquisition experiment normal form, quantitative evaluation and statistical analysis, feature extraction and response difference analysis, and the like, and the response mechanism of typical umami compounds in the human brain is explored by taking the response of people to umami signals as a research object and taking electroencephalogram signal detection analysis as a main method. A new theoretical basis is provided for quantitative evaluation based on the human palatable taste intensity, and the method is worthy of wide popularization.

The invention provides a photovoltaic scene generation method based on an improved generative adversarial network, and the method is realized based on an improved generative adversarial network model, and the construction of the model comprises the following steps: 1) collecting annual hourly photovoltaic irradiance historical data, and carrying out the normalization processing; 2) carrying out weather feature extraction on the collected photovoltaic irradiance historical data; 3) taking the weather-related characteristics as input and the photovoltaic irradiance data as output, training to obtain an improved generative adversarial network model, generating photovoltaic power data by a generator of the model according to noise input, and evaluating the distribution difference between the generated data and real data by an evaluator; the generator ensures that the generated data accords with the physical law of photovoltaic power generation through dynamic gate function constraint, the sample selection probability can be optimized and adjusted through weighted sampling, and the learning ability of the generator for scarce samples is enhanced. Through the method, accurate and diversified photovoltaic power scenes can be generated, and reliable data support is provided for a power system.

The invention provides a fine-grained image clustering method and system based on FG-CLIP and language enhancement, and belongs to the field of image clustering. A pre-training text generation model is utilized to generate diversified coarse-grained text descriptions, then a text fine-grained module and a text screening module are utilized to perform fine processing on a text, and two enhanced views are fused to form consistent text representation, so that the comprehensiveness of text information is enhanced. Then, a neighbor set is constructed according to the feature similarity through a random neighbor sample selection module, attention to samples in the same cluster is improved, and the selection range of positive samples is expanded; and finally, inputting the image features, the text features and neighbor samples thereof into each modal cluster-level mapping head for dimension mapping, and realizing alignment and joint training of the two modal features through random neighbor contrast loss. Under the assistance of the text information, the fine-grained clustering targets can be accurately distinguished, so that the accuracy and robustness of fine-grained image clustering are effectively improved.

The invention discloses a sample screening prompt generation method based on knowledge point graph guidance, and belongs to the field of computer natural language processing. The method is divided into two stages: in the first stage, knowledge points are extracted and a knowledge point graph is constructed by analyzing book contents of Common Course Standard Invention, and data are aligned to the graph; in the second stage, input query is mapped to corresponding knowledge points through knowledge point association, samples with the knowledge point overlapping degree being 0 are removed through a sample initial screening model, deep knowledge semantic features are extracted through a self-attention neural network, an optimal sample is screened out from a training set in combination with a knowledge point similarity and diversity comprehensive selection model, and the deep knowledge semantic features are extracted; and inputting a large language model to generate an answer after combining with the query. According to the method, the context learning performance of a large language model is remarkably improved, the problem that sample selection is interfered by surface language characteristics is solved, the knowledge point dimension information is introduced, so that the understanding of the model on tasks is more accurate, the generation illusion is reduced, and experimental verification is achieved.

The invention discloses a positive and negative unmarked image classification method and system based on depth metric learning, and relates to the technical field of sample classification, and the method comprises the steps: obtaining a marked positive sample image and an unmarked image, inputting the marked positive sample image and the unmarked image into a depth metric learning model, adjusting the marked positive sample image in a depth feature space, and obtaining the unmarked image; minimizing the distance between the marked positive sample image and the center of the positive sample image, aligning the unmarked image and the corresponding enhanced view in a representation space through self-consistent metric learning, and outputting related feature representation; using related feature representation to select sample images which are in multiple proportion to the positive sample images as reliable negative samples; and taking the related feature representation as input, using a positive sample image and a reliable negative sample as supervision signals to train a binary classifier, and classifying positive and negative unmarked images to be detected. According to the method, dependence on heuristic negative sample selection is avoided, and high-quality feature representation with discrimination on positive and negative samples can be learned.

The invention discloses a target detection online learning dynamic sample selection method and system, computer equipment and a storage medium. The method comprises the steps that classification uncertainty and positioning uncertainty of samples to be screened are obtained through Monte Carlo Dropout sampling; constructing multi-dimensional feature vectors including classification uncertainty, positioning uncertainty, knowledge gap matching degree and the like; constructing a dynamic weight learning network based on an attention mechanism, and calculating the weight of each feature in combination with a model verification set performance index; and performing multi-dimensional value scoring on the samples according to the feature weights, and screening out an optimal sample subset in combination with calculation power limitation so as to complete online updating of the model. According to the method, the sample value is comprehensively evaluated through multi-dimensional feature fusion, different scene requirements are adapted by utilizing dynamic weights, and resource consumption and updating effects are balanced in combination with computing power perception sampling, so that the adaptability and detection precision of the model in a dynamic scene are effectively improved, and meanwhile, the dependence on manual annotation is reduced.

The invention relates to a safety knowledge fusion method based on distributed continuous learning and intelligent sample driving, and belongs to the technical field of machine learning. The core of the method is that an edge node not only executes a local model, but also evaluates the performance of the model on local data, so that a representative data sample which is most valuable for global model optimization is intelligently screened out. The confidentiality of the samples is ensured through asymmetric encryption (such as ECIES), and the completeness and source authenticity of the samples are ensured through digital signature (such as ECDSA) in combination with a hash function (such as SHA-256), and then the samples are uploaded to a cloud server. And the cloud server verifies, decrypts and aggregates the received samples for training and updating the global model. The updated global model is then safely distributed back to the edge node. According to the method, an intelligent sample selection mechanism is introduced at an edge end, and a cryptographic tool is combined, so that the efficiency of distributed learning, the model performance and the security of data transmission and fusion are effectively improved.

The invention relates to the technical field of artificial intelligence, and discloses a self-supervised learning-based electrocardiosignal anomaly detection system, which comprises an electrocardiosignal acquisition module, a data preprocessing module, a feature coding module, a self-supervised comparative learning module, an anomaly score calculation module and an anomaly judgment module which are in communication connection, the electrocardiosignal acquisition module is used for acquiring an original electrocardiosignal and transmitting the original electrocardiosignal to the data preprocessing module; the data preprocessing module is used for conducting denoising, baseline drift correction and standardization processing on original electrocardiosignals and transmitting the processed signals to the feature coding module. According to the invention, through a dynamic negative sample selection unit in the self-supervised contrast learning module, depending on technologies such as feature queue maintenance, feature distance calculation, clustering analysis and the like, samples which have significant difference from positive sample feature distribution and are different in category are screened as negative samples, so that the model can accurately learn real similarity between similar electrocardiosignal samples; and feature learning confusion is avoided.

The invention discloses an intelligent decision-making method and system for intraoperative stress injury based on machine learning, and relates to the technical field of electric digital data processing. The intelligent decision-making method for the intraoperative stress injury based on machine learning comprises the following steps: S1, selecting and analyzing a sample; s2, performing multi-collinearity evaluation; s3, evaluating information loss; and S4, intelligent decision making. According to the method, sample selection adjustment is carried out through sample selection analysis data, then principal component dimension reduction is carried out based on multi-collinearity evaluation data, finally potential factors are introduced step by step based on information loss evaluation data, and whether risk weights of high-risk factors are obtained or not is judged based on multi-collinearity evaluation values after the potential factors are introduced. And intelligent decision-making of the intraoperative stress injury is performed in combination with the high-risk factors, so that the screening accuracy of the high-risk factors in the intelligent decision-making of the intraoperative stress injury is improved, and the problem of low screening accuracy of the high-risk factors in the intelligent decision-making of the intraoperative stress injury in the prior art is solved.

The invention discloses a sample selection device for natural resource engineering surveying and mapping, and relates to the technical field of engineering surveying and mapping. The bottom plate is arranged at the lower end of the sampling pipe in a sleeving manner and is used for ensuring that the sampling pipe is always in a vertical state during sampling, and the depth-keeping sampling mechanism is arranged at the bottom of the movable groove and is used for sampling soil with the required depth. The tip end of the bottom of the sampling pipe is inserted into a soil sampling area firstly, the sampling pipe is rotated to vertically drill into soil, after the through groove is drilled into the soil needing to be sampled by a depth, a soil sample can be taken out more conveniently by utilizing the depth-keeping sampling mechanism, and meanwhile, depth-keeping sampling can be carried out on the soil needing to be sampled.

The invention discloses an electric power field training set dynamic construction method and system based on BERT and reinforcement learning. The method comprises the following steps: S1, constructing a terminology library and a knowledge graph in the power field; s2, performing power field adaptation on the BERT model by using a terminology library and a knowledge graph to obtain a field adaptation model; s3, designing a strategy network based on reinforcement learning so as to dynamically select unlabeled samples and generate pseudo labels; and S4, based on the domain adaptation model and the strategy network, dynamically constructing and optimizing a training set through an iteration process. According to the method, the training set quality and the model performance are jointly improved, manual labeling is avoided, sample selection can be adaptively adjusted, the labeling cost is low, and the model accuracy is high. According to the method, the pseudo labels are automatically generated through the reinforcement learning strategy network, manual labeling requirements are reduced, labeling efficiency and model performance are improved, the quality of a training set can be optimized, and adaptability is enhanced.

The invention discloses a children voice expression error recognition and correction method based on comparative learning, and the method comprises the steps: carrying out the preprocessing of an inputted children voice signal, obtaining a logarithmic Mel spectrum feature sequence, and converting the logarithmic Mel spectrum feature sequence into voice semantic coding features through an improved Transform encoder; on the basis of an age-adaptive positive and negative sample selection mechanism, voice features are optimized by using a comparative learning method, and an enhanced voice representation vector is obtained; constructing a multi-modal fusion network, combining an enhanced voice vector and BERT language model features, realizing adaptive fusion through multi-head cross attention and a gating mechanism, adopting a bidirectional LSTM to design an error positioning module to recognize the position and the type of a pronunciation error, and using a joint loss function to execute end-to-end training; standard correction audio is generated according to the error type, and correction guidance is provided for children. According to the invention, high-precision recognition, positioning and correction of children's speech expression errors are realized, and technical support is provided for children's language development.

The invention discloses a multi-modal cross-domain small sample facial expression recognition method based on relational distillation self-paced learning, and relates to a computer vision technology. Constructing a multi-modal semantic enhancement module, generating semantic descriptions of expressions by using a large language model, performing CLIP coding, and performing alignment and fusion with image visual features in the multi-modal semantic enhancement module to construct a multi-modal prototype; a self-paced learning mechanism based on relational distillation is designed, visual and semantic structural errors are calculated, and progressive training from easy to difficult is realized through a soft and hard mixed sample selection strategy and a mixed sample selection mechanism regulated and controlled by a dynamic threshold value. And cross-domain migration of emotional knowledge from basic expressions to fine-grained composite expressions can be effectively realized. And under the condition that only a small number of labeled samples are provided, rapid adaptation and accurate recognition of new expressions can be realized. The method is remarkably superior to a traditional supervised learning method, has higher practicability and expansibility, and can better meet the requirement for efficient recognition of new expressions in practical application.

The invention discloses a lightweight CAN (Controller Area Network) bus intrusion detection method based on enhanced active learning, which optimizes an annotation strategy in a sample selection process through a label self-adaptive optimization mechanism, an enhanced learning-driven strategy adjustment capability and a resource-controllable dynamic reasoning architecture, can efficiently utilize annotation data, and improves the detection efficiency. The workload and cost of manual annotation are greatly reduced, the overall annotation requirement is reduced, the utilization efficiency of annotation resources is improved, a lightweight inference model set and an accurate inference model set are combined by designing a double-stage inference gating algorithm based on reinforcement learning, inference calculation can be dynamically adjusted according to actual needs, and the annotation efficiency is improved. And uncertain samples are transferred into a precise reasoning model set for further judgment, so that the use of overall computing resources is optimized, the computing burden of an intrusion detection system is reduced, meanwhile, the response speed is increased, the sustainable optimization capability of the model is enhanced, and the detection precision and the adaptive capability are improved.

An example device for decoding video data includes: a memory for storing video data; and a processing system implemented in circuitry and configured to: generating a prediction block for a current block of the video data using a sample position-dependent intra-prediction mode, including, for one or more samples of the prediction block: select a filter for the sample according to a shape of the current block and a position of the sample; and predict the sample using the selected filter; decode a residual block for the current block of the video data; and combine the prediction block with the residual block to decode the current block of the video data.

The invention discloses a class increment target detection method combining sample playback and an attention mechanism, and belongs to the field of target detection, and the method comprises the steps of dynamic sample selection playback, DETR model construction based on attention enhancement, and hybrid playback training. At the end of each incremental learning stage, dynamically replaying the samples, screening representative samples from the current task through a K-Center-Greedy algorithm, and storing the representative samples into a memory bank with fixed capacity; according to the DETR model based on attention enhancement, a channel attention module (SE module) is embedded in a target detection network DETR, and key feature expression is enhanced through feature channel re-calibration; the hybrid replay training is mainly characterized in that during new task training, a historical sample is extracted from a memory bank and mixed with a current sample to serve as a training set to be input, and cross-task knowledge fusion is achieved by jointly optimizing a detection loss function and an attention weight.

The invention discloses a performance self-improving large model recommendation system and method, and the system comprises an initialization module which is used for pre-training a large language model through supervision and fine tuning, and generating an initial recommendation model; the self-optimization module comprises three iteratively executed sub-modules; the sample selection sub-module is used for screening historical data samples of which the information amount is higher than a threshold value on the basis of comparison between the model prediction probability and the preset threshold value; the response fusion sub-module is used for generating K candidate responses for a selected sample and generating a preference data set based on model evaluation; and the DPO optimization sub-module is used for updating model parameters by utilizing the improved DPO loss function and generating an optimized recommendation model. According to the method, the dependence on static preference data in the prior art is broken, the recommendation quality and robustness are improved, and adaptive optimization is realized.

The invention relates to a cross-domain retrieval method. The method comprises the following steps: extracting image data of a source domain and a target domain; identifying a feature sample of the image data through an encoder; constructing a cross-domain relation graph between the source domain and the target domain based on a neighbor principle; identifying an internal structure of the cross-domain relation graph according to a diffusion sample selection technology; mapping the source domain tag features and the target tag features to binary hash codes according to the low-noise sample set; sampling a random walk path on the cross-domain relation graph; carrying out pixel-level mixing on the data mixing pair by using an edge weight in the cross-domain relation graph as a mixing coefficient; performing semantic mixing on the data mixing pair; and executing cross-domain retrieval according to the mixed features. According to the technical scheme, the adaptive process under different noise levels is effectively balanced, and more robust and efficient cross-domain Hash retrieval is achieved.

The invention discloses a power system power flow self-adaptive regulation and control method based on deep transfer learning, and the method comprises the steps: 1, constructing a power system source domain power flow analysis model, and constructing a loss function for a power flow error in combination with the power flow data characteristics of a source domain and a target domain; step 2, designing a network parameter initialization method suitable for power flow analysis, realizing transferable representation from source domain data to a target domain, and constructing an initialized target domain power flow analysis model; 3, introducing a sample selection and incremental learning mechanism, and carrying out dynamic screening and retraining on key power flow samples in a target domain; and 4, carrying out load flow calculation and optimization control on the target power system by adopting the trained target domain load flow analysis model, and realizing self-adaptive optimization of the system operation state through iterative correction and parameter updating. According to the method, rapid calculation and dynamic optimization of the power flow of the power system are realized, and the intelligent level, stability and calculation reliability of power grid operation analysis are remarkably improved.

The invention provides a landslide susceptibility evaluation method based on a deep neural network and considering time sequence InSAR and time sequence rainfall, belongs to the technical field of geological disasters and geographic information, and particularly relates to a landslide susceptibility prediction method based on the deep neural network. The method comprises the following steps: establishing a buffer area through landslide points, selecting a research area to randomly generate non-landslide points, screening related static characteristic factors by using a Pearson's correlation coefficient matrix and a VIF method, obtaining GCP points by using PS-InSAR, introducing the generated points as SBAS-InSAR processing parameters, generating dynamic characteristic factors of surface deformation, and calculating the deformation of the surface deformation. The time sequence average rainfall of the region is obtained through spatial interpolation, the rainfall before earth surface deformation is obtained through data processing codes to serve as another dynamic characteristic factor, finally evaluation is conducted through the constructed ResNetconvLSTMUnet, and the susceptibility index is divided into five grades; according to the method, the dual-time-sequence dynamic factors are creatively fused, the spatial-temporal feature extraction capability is enhanced, the data redundancy is reduced, the sample selection is reasonable, the evaluation precision is high, and scientific support can be provided for landslide disaster prevention and reduction.

The present invention belongs to the technical field of voiceprint recognition and discloses a feature-based intelligent voiceprint recognition method and system; the method comprises: collecting a plurality of voiceprint data samples and preprocessing each voiceprint data sample to obtain a comprehensive voice sample; constructing a voice directed graph model based on the comprehensive voice sample, and identifying a core voice sample set from the voice directed graph model; extracting features from the core voice sample set to obtain #imgabs0# comprehensive feature vectors; obtaining a sample selection strategy using a clustering algorithm based on the comprehensive feature vectors and the core voice sample set; adaptively training a voiceprint recognition model using the sample selection strategy based on the comprehensive feature vectors; and applying the trained voiceprint recognition model to voiceprint recognition after the training is completed, thereby giving full play to the advantages of deep learning in the field of voiceprint recognition, not only improving the speed and efficiency of voiceprint recognition, but also reducing the system's false recognition rate and missed recognition rate.