1224 results about "Negative sample" patented technology

The invention discloses a mineral resource intelligent prediction method and system based on multi-source heterogeneous data fusion and deep learning, and the method comprises the steps: collecting and preprocessing multi-source heterogeneous data, and carrying out the standardization processing to form a structured data set; multi-source heterogeneous data fusion: realizing data layer space registration and feature layer weight dynamic allocation through an attention mechanism multi-modal fusion module, and outputting a high-dimensional metallogenic feature vector; constructing a CNN-LSTM mixed deep learning model and completing initialization training, and outputting an initial mineralization probability graph; and establishing a dynamic updating engine, performing model increment training based on transfer learning, correcting the mineralization probability through positive and negative sample reinforcement learning in combination with a newly added data type, and outputting a time sequence dynamic mineralization probability graph. According to the method, mineralization probability dynamic evaluation and risk quantitative updating are realized, the prediction precision and the model updating efficiency are improved, the method is adaptive to a multi-stage exploration scene, and accurate real-time support is provided for exploration decision making.

The invention discloses an emotion recognition method and system based on cross-modal alignment and matching enhancement. According to the method, firstly, feature extraction is carried out on text, audio and video modalities in a data set, and then a text and audio cross-modal emotion alignment module and a text and video cross-modal emotion alignment module are constructed respectively, so that cross-modal semantic alignment is realized. Constructing an emotion label matching module based on an alignment result, generating modal pairs with similar emotions but different labels by using a difficult negative sample mining strategy, and paying attention to cross-modal emotion consistency through a dichotomy task guide model; performing modal feature fusion on the three modals through a six-layer attention crossing mechanism, finally splicing feature vectors, inputting the spliced feature vectors into a long-sequence context fusion modeling module for deep modal fusion, and capturing cross-modal interaction information; and the fused features are sent to an emotion classification module, and a final emotion category recognition result is output.

A method of emotion recognition in cross-subject EEG signals, belonging to technical field of deep learning, includes the following steps: S1, constructing the extracted DE features into positive and negative samples by using a positive and negative sample generator; S2, sending the DE features of an anchor and the positive and negative samples into the encoder for coding, mapping the DE features to a latent space, performing regression prediction on the encoded anchor samples in the latent space by using an autoregressive model, training the encoder by using a probability supervision contrastive loss function; and S3, connecting the trained encoder to the classifier for fine tuning, and training the classifier through the cross entropy loss function; in this process, the encoder does not perform gradient propagation to complete cross-subject emotion recognition.

The invention discloses an open set domain adaptive image classification method based on a difference prompt learning technology of a pre-training vision-language model. According to the method, high-quality pseudo-open class images are generated, and de-noising text embedding and de-noising visual embedding are obtained by using a differential prompt learning technology, so that class characteristics of a source domain, a target domain and pseudo-open class samples are effectively extracted, and irrelevant noise is inhibited. According to the method, a vision-text comparison loss mechanism, a triple distance comparison loss mechanism and a negative sample penalty mechanism are further designed, a known category and an unknown category are effectively distinguished in a feature space, and the semantic alignment capability of cross-domain similar samples is enhanced. The method can significantly improve the classification accuracy and model robustness in an open set domain adaptation task, has the advantages of simple structure, high calculation efficiency, good generalization performance and the like, and is suitable for image classification, cross-domain transfer learning and other related application scenes.

The invention relates to the field of artificial intelligence cross-modal retrieval, and provides a video text cross-modal retrieval method and system based on spatio-temporal feature fusion. The method comprises the following steps: carrying out key frame sampling and time sequence partitioning on an input video, extracting static visual features through a spatial feature network, and extracting motion features through a time dynamic network; a self-adaptive gating fusion module is adopted to dynamically calculate spatial-temporal feature weights and perform weighted fusion; extracting text semantic features by using a pre-training language model; constructing a double-flow projection network to map video fusion features and text features to a unified measurement space, and optimizing a feature distance by adopting a contrast loss function containing difficult negative sample mining and intra-modal constraint; and outputting a retrieval result according to the cosine similarity sequence. The system comprises four units, wherein the gating fusion module is integrated with an FPGA acceleration circuit. According to the method, mAP (at) 10 is equal to 0.78 in a UCF-101 data set, the time sequence action retrieval accuracy rate is 92.8%, and the single video retrieval delay is 23 milliseconds.

The invention provides a training-free text-image generation method based on a diffusion model, and relates to the technical field of computer graphic processing and artificial intelligence. The method comprises the following steps: extracting semantic phrases and layout information in an input text by utilizing a natural language model, inputting the input text, the semantic phrases and the layout information as additional conditions into a diffusion model, and extracting cross attention maps of different time steps; a positive and negative sample concept and a foreground and background concept based on an object are constructed, a new loss function is calculated on a cross attention map for semantic information and layout information, the loss function combines semantic loss, regional loss and original loss of a diffusion model and is used for updating a potential space image, and the image is generated through iterative denoising and a decoder. According to the method, additional training is not needed, image generation output based on the diffusion model better meets text requirements, and a better text and image alignment effect is achieved.

The invention belongs to the technical field of industrial operation and maintenance term processing, and provides a big language model training method based on a knowledge graph, and the method comprises the steps: carrying out the cooperation with operation and maintenance personnel in advance to set a fault dictionary, collecting fault instances, carrying out the classification and preprocessing, extracting causal trigger words of the fault instances, and setting a text dependency relationship rule. Splitting and extracting a triple of the fault instance, and importing to form a knowledge graph; mapping the instance attribute into a node feature vector, outputting an embedded representation of a node, and fusing the embedded representation with the multi-modal feature vector to form a fused feature embedded representation; replacing semantic and degree words with fault instances in the knowledge graph to generate positive samples, randomly replacing subjects, objects and causal relationships to generate negative samples, inputting the negative samples into a semantic encoder, mapping the negative samples to the same metric space, and calculating similarity to obtain a comparison loss value to adjust graph convolutional network parameters; and after the fusion feature embedding representation retrieval of the fault instance is carried out and a subgraph is generated, maintenance suggestions are given step by step along a causal chain.

The invention belongs to the technical field of vulnerability detection, and particularly relates to a source code vulnerability detection method and system based on semantic comparative learning. Comprising the steps of obtaining source codes, preprocessing the source codes, and generating positive and negative samples paired with the source codes; improving the CodeBERT model to obtain an encoder, and building a momentum contrast learning model comprising a vulnerability classification branch, a representation consistency branch and a contrast learning branch based on the obtained encoder; training the momentum contrast learning model comprising the three branches by adopting a contrast learning method; and based on the trained momentum contrast learning model, performing vulnerability detection on a to-be-analyzed code. According to the method, a semantic contrast learning-driven source code vulnerability detection framework is designed and realized, and the problems of limitation of traditional static analysis and insufficient generalization ability of an existing deep learning model are solved by learning representation which is robust to code semantic changes and sensitive to vulnerability modes.

The invention provides a code retrieval method and device for AI programming and related equipment, and the method comprises the steps: constructing a double-layer embedded vector model which comprises a basic semantic coding layer for parameter freezing and a plurality of independent lightweight adaptation layers which coexist; monitoring implicit feedback data generated by interaction of a user in the target code library to extract positive and negative sample pairs, and mining difficult negative samples based on a current model to form a training sample set; when the cumulative number of the samples reaches a preset trigger threshold value, optimizing and updating at least one lightweight adaptation layer parameter by utilizing a preset comparison loss function, and applying elastic weight constraint to key parameters of the lightweight adaptation layer parameter to obtain an updated double-layer embedded vector model; and performing multi-dimensional performance evaluation on the updated model, and when an evaluation result meets a preset publishing condition, deploying the model to a code retrieval service. The problems that in the prior art, cross-domain is difficult, the adaptation cost is high, and catastrophic forgetting is likely to happen are effectively solved, and zero-labeling self-evolution of the model is achieved.

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 bone tumor fine-grained classification model training and classification method and device, and the method comprises the steps: constructing a multi-modal positive sample containing global / local positive lateral X-ray images, lesion attributes and patient information, and removing a false negative part in combination with text semantic similarity to construct a high-quality negative sample; global / local image features are extracted through double image encoders, lesion attribute keywords are converted into'entity-translation-existence 'triples based on a medical knowledge base, and basic information of a patient and global / local semantic features of lesion attributes are extracted through a text encoder; infoNCE contrast loss is constructed for global images and global semantics based on contrast learning, global image-text feature alignment and local image-text feature alignment are realized in combination with a local mutual information loss and classification loss training model calculated based on a DV variational formula, medical term semantics are deeply combined, the training stability is improved, and the training efficiency is improved. The accuracy and robustness of bone tumor subtype classification are remarkably improved, and reliable support is provided for clinical precise diagnosis.

The invention discloses an electricity consumption anomaly detection method based on a graph structure, and relates to the technical field of electricity consumption anomaly detection, and the method comprises the steps: obtaining historical electricity consumption data, environment variable data and power grid physical topology information of a power grid region, and carrying out the standardization processing of the historical electricity consumption data, the environment variable data and the power grid physical topology information, a standardized time sequence-environment data set is obtained; based on the standardized time sequence-environment data set, identifying a causal association relationship between electricity consumption and environment variables through a causal discovery algorithm, and constructing a causal perception heterogeneous graph containing region nodes and environment factor nodes; based on a causal perception heterogeneous graph, introducing a plurality of region nodes of which hyperedge connection is influenced by the same environmental event to form a dynamic hypergraph structure for representing a many-to-many environment-region coupling relationship; based on the dynamic hypergraph structure, a graph comparison learning task is constructed, positive and negative sample pairs are generated by applying disturbance to node features and the graph structure, and unsupervised pre-training is completed;

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 relates to the technical field of federated learning security, and discloses a federated learning poisoning defense method based on time-frequency spectrogram and comparative learning, which comprises the following steps: receiving model update uploaded by each client, grouping and vectorizing parameters according to model layers, and generating a time-frequency spectrogram by applying short-time Fourier transform to parameter vectors of each layer; based on the time-frequency spectrogram, constructing a positive sample pair through data enhancement, carrying out difficult negative sample mining, and training an encoder by using a contrast loss function to extract an embedded vector with high discriminant power; and performing unsupervised clustering on the embedded vector by using a DBSCAN clustering algorithm, judging the maximum cluster as a benign client, performing final judgment in combination with historical malicious records, and only aggregating model parameters of the benign client to update a global model. According to the invention, high-precision detection of attack features can be realized, and a more universal, more efficient and more practical federal learning poisoning attack defense method is realized.

The invention provides an energy storage device damage diagnosis and interaction system based on penetration vision and a large model, and relates to the technical field of energy storage device damage diagnos.The energy storage device damage diagnosis and interaction system comprises a penetration type 3D sensing module, a periodic topology analysis module, a feature mapping and retrieval module and a large model reasoning and interaction module, and the energy storage device is scanned and reconstructed; the method comprises the following steps: constructing a time-space decoupling periodic Transform network, introducing a periodic mask matrix to force the network to pay attention to a repeatability rule of an internal structure of a battery, and identifying internal tiny deformation and structural damage by calculating topological consistency under the condition that a large amount of negative sample training is not needed; visual defect features are mapped into text embedding by utilizing a feature projection technology, a diagnostic report containing physical cause analysis and maintenance suggestions is generated by combining a retrieval enhancement generation technology and a large language model, and a user is supported to perform interactive questions and answers in a natural language. The method can solve the problems that in the prior art, three-dimensional deformation is difficult to quantify, small samples are difficult to train, and intelligent decision-making ability is lacked.

The invention requests to protect an emotion analysis method based on a text-guided comparative learning and gating fusion mechanism, which is used for solving the problems of insufficient modal feature representation discrimination and insufficient modal information fusion in multi-modal emotion analysis. Firstly, features of three modes are extracted in a feature extraction module, then a text mode is selected as a dominant mode, a text-guided weighted comparison learning strategy is designed, the strategy screens positive and negative sample pairs through the text mode, and weights are dynamically distributed for hard pairs and easy pairs, so that the model keeps universal features while focusing on the hard pairs, and the model is optimized. Therefore, the discrimination capability of modal feature representation is improved. Secondly, in a feature fusion module, a text-guided gating fusion mechanism is provided, weights of text modals and cross-modal fusion features are adaptively adjusted through a gating unit, and a self-attention mechanism is used to suppress fusion noise, so that the modals are fused more fully. Experiments on public data sets CMU-MOSI and CMU-MOSEI show that compared with a reference method, the method provided by the invention has advanced performance on multiple indexes such as F1, ACC-2 and the like.

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.

The invention discloses a bacteriophage host prediction method based on a weighted integrated convolutional neural network. The method comprises the following steps: obtaining a joint feature vector for an input bacteriophage protein sequence and a host receptor protein sequence; taking the bacteriophage-host pair as a positive sample, randomly selecting non-host bacteria for each bacteriophage to generate a negative sample, dividing the negative sample into M subsets, and combining each subset with all the positive samples to form M balance training subsets; for each balance training subset, training an independent one-dimensional convolutional neural network model; independently training a one-dimensional convolutional neural network sub-model for each balance training subset to obtain a plurality of sub-models with different performances; and distributing weights based on the performance indexes of the sub-models on the verification set, performing weighted fusion on the prediction probabilities of the M sub-models, and outputting a final host interaction probability. The method realizes accurate prediction of phage-host interaction on the premise of only needing basic sequence information, and has the advantages of light weight, high efficiency and wide applicability.

The invention belongs to the technical field of artificial intelligence, and discloses a method for generating a migratable adversarial sample for a vision-language pre-training model based on comparative learning, and the method comprises the steps: carrying out the discrimination of an image-text pair set obtained through the data enhancement of the vision-language pre-training model, and obtaining a positive sample set and a negative sample set; in the positive sample guided adversarial sample generation process, negative samples are introduced to further enrich the diversity of the adversarial samples, so that cross-modal interaction is realized to the greatest extent; in the adversarial sample iteration generation process, through combination of three different types of learning modes of positive sample learning, negative sample learning and contrast learning, a transferable vision-language pre-training model adversarial sample is guided to be generated. Wherein the adversarial samples generated by fusing three learning methods show better mobility in different vision-language pre-training models and downstream tasks of the different vision-language pre-training models.

The invention discloses an unmanned aerial vehicle electric power inspection image segmentation method and system, and relates to the technical field of image segmentation. By means of a labeled sample image set, positive and negative information is fully fused in a confidence map cooperation module to carry out confidence modeling, so that a cooperation confidence map can accurately reflect the difference between a target and a background; the negative sample distribution data effectively depicts the features of background interference, then high-discrimination positive and negative points are screened out in the point selection module, accurate prompt is provided for segmentation, iterative optimization is performed through the noise sensing and refining module, and therefore the segmentation efficiency is improved under the condition that only few labeled samples are needed. According to the method, high-precision, high-robustness and training-free segmentation of the target component in the electric power inspection image is realized, the influence of complex backgrounds and similar interferents is effectively resisted, the method has high robustness, the technical problems of low segmentation precision and insufficient robustness under the condition of few labeled samples in the prior art are fully solved, and the requirements of electric power inspection on automation and precision are met.

The invention belongs to the field of automatic driving algorithms, and particularly relates to a vehicle track prediction method based on implicit map feature expression. Comprising the following steps: step 1, a track-map semantic relation alignment preprocessing mechanism based on a map topological structure; 2, a negative sample generation method based on map clustering preprocessing; 3, constructing a track-map semantic feature comparison learning network; and 4, establishing a universal fusion mechanism for the semantic features of the implicit map. Compared with an existing trajectory prediction method depending on a high-precision map, the method has the advantages that high-precision trajectory prediction can be realized without high-cost map information, and the robustness and generalization of the model in a map missing or distorted scene are improved. Meanwhile, the method can be seamlessly combined with various map-free prediction models, and has universality and expansibility.

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 technical field of artificial intelligence, discloses a multi-modal self-supervision abnormal mode detection method and system, and aims to solve the problems that in the prior art, a large amount of annotated data is relied on, modal fusion is insufficient, the anomaly discrimination ability is weak, and the dynamic environment is difficult to adapt. The method comprises the following steps: synchronously acquiring videos, audios, sensor time sequences and log text data, and carrying out time alignment; extracting spatial-temporal characteristics of each modal through a modal specific encoder; constructing a contrast learning task under a label-free condition, generating positive and negative sample pairs by utilizing data enhancement, and driving model learning discriminative representation; and cross-modal feature alignment and dynamic weighted fusion are realized by adopting an attention mechanism, and joint representation is generated. According to the scheme, efficient anomaly detection without annotation data is realized, the multi-modal fusion representation capability is remarkably improved, the false alarm rate is reduced, the environmental adaptability is enhanced, and the real-time monitoring requirement is met.

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 belongs to the technical field of power load prediction, and relates to a power load prediction method based on interpretable multi-mode enhancement, which comprises the following steps: 1, constructing a text representation mode of an original load time sequence through a multi-mode enhancement module; 2, time sequence modal information and text representation information of the load are embedded into a high-dimensional vector space through a two-channel coding module; 3, receiving an embedded vector of a time sequence mode through a multi-mode prediction module, and taking a representation vector of a multi-mode text as input; 4, realizing an interpretable multi-mode alignment module; according to the method, the effects among the multi-modal information are divided into uniqueness, redundancy and collaboration; by constructing a negative sample pair for training, alignment of multi-modal information representation is enhanced, and then the prediction performance of the model is improved.

The invention discloses a transformer insulation state evaluation method based on artificial intelligence multi-parameter fusion perception, and relates to the technical field of electrical equipment detection, and the method comprises the steps: 1, obtaining the multi-modal monitoring data of a transformer, and constructing a dynamic graph structure; wherein the multi-modal monitoring data comprises oil chromatography data, partial discharge data and thermal image data; 2, encoding the dynamic graph structure through an encoder extraction embedding technology, carrying out space-time dimension feature aggregation and evolution modeling based on the encoded dynamic graph structure by utilizing graph convolution and time convolution technologies, inputting preset transformer insulation positive and negative sample pairs into a constructed model, carrying out InfoNCE loss unsupervised learning, and carrying out infoNCE loss unsupervised learning; obtaining a state manifold of a normal transformer, and carrying out multi-modal data space-time alignment and anomaly distinguishing on the state manifold; and step 3, introducing a Bayesian weight after multi-modal data space-time alignment and anomaly distinguishing, and outputting probability distribution of insulation state embedding.

The invention discloses a living cell super-resolution imaging method and system based on self-supervised learning. The method comprises the following steps: S1, obtaining a fluorescence microscopic image through a standard fluorescence microscopic system; s2, for the acquired single noise image, generating positive and negative sample pairs of self-supervised training data through an autonomously designed and optimized self-supervised strategy; s3, constructing a front denoising neural network and a rear deconvolution network, and performing network training; and S4, inputting a newly obtained noise image into the trained front denoising network to obtain a denoised image, and inputting the denoised image into the rear deconvolution network to obtain a final super-resolution reconstructed image. According to the invention, the standard fluorescence microscopic system and the self-supervised denoising processing module are combined, so that high-quality denoising and super-resolution reconstruction of the fluorescence microscopic image can be realized in a low-photon signal scene.

The invention discloses a contrastive learning molecular property prediction method based on an anti-fact and a large language model, and belongs to the technical field of molecular representation learning combining a large language model and a graph neural network, and the method comprises the following steps: generating an original molecular graph through a molecular SMILES character string; generating a skeleton disturbance diagram and a functional group disturbance diagram; ensuring that the generated perturbation diagram is an anti-fact hard negative sample; generating a molecular natural language description through a large language model, and converting the molecular natural language description into a molecular semantic embedding vector by using a small language model; performing comparative learning among the original molecular graph, the positive sample and the anti-fact hard negative sample; original molecular graph embedding and molecular semantic embedding obtained after comparative learning are fused, and downstream molecular property prediction is carried out. According to the method, hard negative samples can be generated through an anti-fact mechanism, the diversity of molecular characterization modes can be ensured through a large language model, and the accuracy of comparative learning during molecular property prediction is remarkably improved.

The invention relates to the technical field of computer vision and deep learning, in particular to a large-area defect image restoration method based on multi-scale feature fusion, and the method comprises the following steps: obtaining a damaged image and a binary mask corresponding to the damaged image; constructing an encoder, extracting multi-level spatial features of the damaged image, and obtaining a potential semantic vector; decoupling the random noise vector into a semantic embedding vector through a mapping network; a multi-scale feature fusion module based on weight distribution is introduced into the generator, context information of different receptive fields is aggregated, a feature map is reconstructed, and a repaired image is generated; a comparative learning mechanism is introduced into the discriminator, and by constructing positive and negative sample pairs, the feature discrimination ability is enhanced and stable training is performed; and performing joint optimization on the network in combination with multiple loss functions to generate a repair result. By introducing a multi-scale feature fusion module and a comparative learning mechanism, global semantic consistency and local texture details of large-area defect image restoration are realized, and the method has good stability and generalization ability.