59 results about "Latent vector" patented technology

The present disclosure provides a method for generating a video of a body moving in synchronization with music by applying a first artificial neural network (ANN) to a sequence of samples of an audio waveform of the music to generate a first latent vector describing the waveform and a sequence of coordinates of points of body parts of the body, by applying a first stage of a second ANN to the sequence of coordinates to generate a second latent vector describing movement of the body, by applying a second stage of the second ANN to static images of a person in a plurality of different poses to generate a third latent vector describing an appearance of the person, and by applying a third stage of the second ANN to the first latent vector, the second latent vector, and the third latent vector to generate the video.

A deep-learning neural network can be trained to model a probability distribution of the asset-price trends for a future time period using a training data set, which can include asset-price trends of a plurality of assets over a past time period and a latent vector sampled from a prior distribution associated with the asset-price trends of a plurality of assets. The training data set can represent a time series data. A portfolio optimization can be executed on the modeled probability distribution to estimate expected risks and returns for different portfolio diversification options.

Systems and methods for building a latent item vector and item bias for a new item in a collaborative filtering system are disclosed. The method includes dividing incoming users into intervals with each interval having a learning phase and a selection phase. The learning phase scores each incoming user according to a best estimate latent vector and bias and saves the highest score. In the selection each incoming user is scored and a user exceeding the highest score is selected. The best estimate latent vector and bias is then updated based on the user's vector and bias, and the user's interaction with the item. The updated best estimate latent vector is then used in further intervals for learning and selecting users.

Disclosed herein is an evaluation predicting device including: an estimating section configured to define a plurality of first latent vectors, a plurality of second latent vectors, evaluation values, a plurality of first feature vectors, a plurality of second feature vectors, a first projection matrix, and a second projection matrix, express the first latent vectors and the second latent vectors, and perform Bayesian estimation with the first feature vectors, the second feature vectors, and a known the evaluation value as learning data, and calculate a posterior distribution of a parameter group including the first latent vectors, the second latent vectors, the first projection matrix, and the second projection matrix; and a predicting section configured to calculate a distribution of an unknown the evaluation value on a basis of the posterior distribution of the parameter group.

A computing device for handling anomaly detection, comprises an encoder, for receiving an input image, to generate a first latent vector comprising a semantic latent vector and a visual appearance latent vector according to the input image and at least one first parameter of the encoder; and a training module, coupled to the encoder, for receiving the input image and the first latent vector, to update the at least one first parameter according to the input image and the first latent vector and a loss function.

The invention brings forward a method for intelligent visual operation on product pictures according to client preferences. The main content of the method comprises the following steps: obtaining the product images, projecting the images to a manifold, controlling a latent vector, performing editing transfer and obtaining processed pictures. The process is as follows: the product pictures are obtained first of all, the images are projected to the manifold, the latent vector is then controlled, a system provides three constraints, i.e., coloring, sketching and deformation for editing the pictures, next, the editing transfer is performed, including a motion color flow algorithm and transfer editing to original images, and finally, the processed pictures are obtained. The method is simple and convenient, is easy to operate and enables clients without design capabilities to perform picture operation; trade companies can have a better understanding of the preferences of the clients so as to adjust design of their own brands; and the clients can also perform search directly through modified product pictures, time and labor are saved, and the efficiency is high.

A method and apparatus for automatic recognition of similarities between perturbations in a network, the apparatus includes a memory unit for storing a first data array of multiple perturbation data snapshots each recorded in response to a perturbation observed in the network; a generation unit adapted to generate by machine learning a data model of perturbations trained on the first data array, wherein the trained data model provides a latent vector representation for each of the perturbations; a recording unit adapted to record a perturbation data snapshot if a perturbation is observed during operation of said network and adapted to provide a corresponding second data array for the recorded perturbation data snapshot; and a processing unit adapted to derive a latent vector representation for the observed perturbation from the second data array using the trained data model of perturbations.

A method for recommending an item includes generating a metadata latent vector and an item latent vector based on an item database and a usage history database of items for a specific user; predicting a user latent vector from user information obtained from the specific user based on the usage history database and one of the metadata latent vector and the item latent vector; and generating a recommendation list by extracting at least one recommendation item from the item database based on the user latent vector and the item latent vector.

A method for performing continual learning on a classifier, in a client, capable of classifying images by using a continual learning server is provided. The method includes steps of: a continual learning server (a) inputting first hard images from a first classifier of a client into an Adversarial Autoencoder, to allow an encoder to output latent vectors from the first hard images, allow a decoder to output reconstructed images from the latent vectors, and allow a discriminator and a second classifier to output attribute and classification information to determine second hard images to be stored in a first training data set, and generating augmented images to be stored in a second training data set by adjusting the latent vectors of the reconstructed images determined not as the second hard images; (b) continual learning a third classifier corresponding to the first classifier; and (c) transmitting updated parameters to the client.

A method and device for image generation and colorization are provided. The method includes: displaying a drawing board interface; obtaining semantic labels of an image to be generated based on user input on the drawing board interface, each semantic label indicating a content of a region in the image to be generated; obtaining a color feature of the image to be generated; and automatically generating the image using a generative adversarial network (GAN) model according to the semantic labels and the color feature. The color feature is a latent vector input to the GAN model.

Methods and compositions to impute or predict genotype, haplotype, molecular phenotype, agronomic phenotypes, and / or coancestry are provided. Methods and compositions provided include using latent space to generate latent space representations or latent vectors that are independent of underlying genotypic or phenotypic data. The methods may include generating a universal latent space representation by encoding discrete or continuous variables derived from genotypic or phenotypic data into latent vectors through a machine learning-based encoder framework. Provided herein are universal methods of parametrically representing genotypic or phenotypic data obtained from one or more populations or sample sets to impute or predict a genotype or phenotype of interest.

Systems and methods for building a latent item vector and item bias for a new item in a collaborative filtering system are disclosed. The method includes dividing incoming users into intervals with each interval having a learning phase and a selection phase. The learning phase scores each incoming user according to a best estimate latent vector and bias and saves the highest score. In the selection each incoming user is scored and a user exceeding the highest score is selected. The best estimate latent vector and bias is then updated based on the user's vector and bias, and the user's interaction with the item. The updated best estimate latent vector is then used in further intervals for learning and selecting users.

A computer-implemented method for securing software installation through deep graph learning includes extracting a new software installation graph (SIG) corresponding to a new software installation based on installation data associated with the new software installation, using at least two node embedding models to generate a first vector representation by embedding the nodes of the new SIG and inferring any embeddings for out-of-vocabulary (OOV) words corresponding to unseen pathnames, utilizing a deep graph autoencoder to reconstruct nodes of the new SIG from latent vector representations encoded by the graph LSTM, wherein reconstruction losses resulting from a difference of a second vector representation generated by the deep graph autoencoder and the first vector representation represent anomaly scores for each node, and performing anomaly detection by comparing an overall anomaly score of the anomaly scores to a threshold of normal software installation.

Methods and compositions to impute or predict genotype, haplotype, molecular phenotype, agronomic phenotypes, and / or coancestry are provided. Methods and compositions provided include using latent space to generate latent space representations or latent vectors that are independent of underlying genotypic or phenotypic data. The methods may include generating a universal latent space representation by encoding discrete or continuous variables derived from genotypic or phenotypic data into latent vectors through a machine learning-based encoder framework. Provided herein are universal methods of parametrically representing genotypic or phenotypic data obtained from one or more populations or sample sets to impute or predict a genotype or phenotype of interest.

The invention discloses a method and a device for generating a text image, and belongs to the field of artificial intelligence (AI). The method comprises the steps of obtaining n to-be-processed text images, determining n prediction latent vectors corresponding to the n to-be-processed text images in a generative adversarial network (GAN) model, determining a target latent vector according to the n prediction latent vectors, and generating a target text image from the target latent vector through the GAN model. According to the method, a text image of a specific style can be generated in a deep learning mode.

The invention discloses an equipment anomaly detection method using Gaussian noise, which does not need a large number of fault samples, adopts a generative adversarial mode based on Gaussian noise to train normal samples to obtain a generator used for mapping a latent vector mapping space, and once an abnormal sample is separated from the latent vector mapping space after being mapped by the generator, anomaly detection can be realized. The generator is only matched with a normal sample, a normal vibration image can be mapped into originally defined distribution pg-N(0, 1) through the generator, but an abnormal sample cannot be matched with the originally defined distribution through the generator, and the generator only uses the normal sample to perform generative adversarial training. Therefore, the abnormal samples in different fault modes can deviate from the latent vector mapping space of the normal samples in different modes, and the abnormal detection model based on the method can realize fault early warning in different fault modes.

The disclosure relates to multi-layer perceptron architecture, that may be used for image generation. A new architecture for image generators, where the color value at each pixel is computed independently given the value of a random latent vector and the coordinate of that pixel is provided. No spatial convolutions or similar operations that propagate information across pixels are involved during the synthesis.

The present disclosure provides a method for generating a video of a body moving in synchronization with music by applying a first artificial neural network (ANN) to a sequence of samples of an audio waveform of the music to generate a first latent vector describing the waveform and a sequence of coordinates of points of body parts of the body, by applying a first stage of a second ANN to the sequence of coordinates to generate a second latent vector describing movement of the body, by applying a second stage of the second ANN to static images of a person in a plurality of different poses to generate a third latent vector describing an appearance of the person, and by applying a third stage of the second ANN to the first latent vector, the second latent vector, and the third latent vector to generate the video.

A computer-implemented method for securing software installation through deep graph learning includes extracting a new software installation graph (SIG) corresponding to a new software installation based on installation data associated with the new software installation, using at least two node embedding models to generate a first vector representation by embedding the nodes of the new SIG and inferring any embeddings for out-of-vocabulary (OOV) words corresponding to unseen pathnames, utilizing a deep graph autoencoder to reconstruct nodes of the new SIG from latent vector representations encoded by the graph LSTM, wherein reconstruction losses resulting from a difference of a second vector representation generated by the deep graph autoencoder and the first vector representation represent anomaly scores for each node, and performing anomaly detection by comparing an overall anomaly score of the anomaly scores to a threshold of normal software installation.

The invention discloses a method for identifying non-Gaussian aquifer parameters by fusing underground water level and natural potential data based on a convolutional neural network, which comprises the following steps: 1, training a CVAE network by using a non-Gaussian aquifer parameter field sample, sampling from standard normal distribution, and initializing an estimation set of a latent vector z; 2, inputting the latent vector z estimation set into a trained CVAE decoder, and reconstructing a corresponding non-Gaussian aquifer parameter field estimation set; 3, during a harmonic water pumping test, based on the reconstructed non-Gaussian aquifer parameter field estimation set, operating a hydrological geophysical forward modeling model to obtain hydraulic water head and natural potential simulation data; 4, iteratively updating the latent vector z estimation set by adopting an ESMDA method in combination with the hydraulic head and natural potential observation data; repeating the steps 2 to 4 until the maximum number of iterations is reached; and 5, for the posterior set of the latent vector z obtained by updating, reconstructing through a CVAE decoder to obtain an estimation result of a non-Gaussian aquifer parameter field.

The invention provides an industrial image inspection method and system, and a computer readable recording medium. The method includes: generating a to-be-tested latent vector of a to-be-tested image;measuring a distance between a training latent vector of a normal image and the to-be-tested latent vector of the to-be-tested image; and judging whether the test image is normal or defected according to the distance between the training latent vector of the normal image and the to-be-tested latent vector of the to-be-tested image.

A deep-learning neural network can be trained to model a probability distribution of the asset-price trends for a future time period using a training data set, which can include asset-price trends of a plurality of assets over a past time period and a latent vector sampled from a prior distribution associated with the asset-price trends of a plurality of assets. The training data set can represent a time series data. A portfolio optimization can be executed on the modeled probability distribution to estimate expected risks and returns for different portfolio diversification options.

A method for detecting fauls of an electrochemical energy storage system by utilizing machine learning comprises the following steps: providing a data set for detecting the faults of the electrochemical energy storage system, wherein the data set comprises one or more sub-data sets, and each sub-data set comprises one or more input objects or object sequences for collecting and / or processing the faults of the electrochemical energy storage system; providing an auto-encoder structure, establishing at least one machine learning algorithm training for at least one influence factor for detecting the fault of the electrochemical energy storage system, and outputting a predicted value of the influence factor; extracting at least one or more current input object data or object sequence data, projecting the object data or object sequence data to corresponding autoencoders of the set of machine learning algorithms, outputting corresponding latent vectors, and outputting predicted values of corresponding impact factors after the latent vectors are used as inputs of corresponding decoders; through comparative analysis of the prediction values, detecting the prediction possibility of the fault of the electrochemical energy storage system.

A Kalman filter word vector learning method based on Diesel process, the method includes: training and preprocessing the corpus, generating an LDS language model system, initializing the system parameters, assuming that the process noise satisfies a normal distribution, defining the aggregation class theta t =(μ t ,∑ t ), μ t For the frequency of word t in the corpus, calculate θ t The prior distribution of Dirichlet, the posterior distribution is calculated by Kalman filter derivation and Gibbs sampling estimation, the candidate clusters are extracted by MCMC sampling algorithm, the selection probability of the candidate clusters is calculated, and the candidate with the highest probability value is selected Choose the cluster as θ t , calculate the estimated value of the minimum mean square error of the clustering, substitute the calculation result into the LDS language model, train the model through the EM algorithm, make the model parameters stable, input the preprocessed corpus into the trained LDS language model, and pass Carl The Mann filter updates the formula in one step to compute the implicit vector representation.

The embodiment of this specification provides a computer-executed method and device for evaluating interaction events, wherein the method includes first obtaining a dynamic interaction graph for reflecting the correlation of interaction events, and then using the two target nodes to be analyzed as the current root nodes , and determine the corresponding two subgraphs in the dynamic interaction graph. Input the two subgraphs into the neural network model to obtain the interaction representation vector corresponding to the interaction of the two target nodes. In the neural network model, the processing layer obtains The hidden vector of each node, the fusion layer uses the compression-transform mechanism to determine the weight estimation of the node according to the hidden vector, and update the input characteristics of the node accordingly, pass it to the next processing layer, and finally obtain the aforementioned interaction representation vector through the output layer . Therefore, based on the interaction representation vector, the interaction event of the interaction between the two target nodes can be evaluated.

A method for guided design generation includes receiving a plurality of image samples for an intended design and generating a first plurality of images by providing a positive latent vector to a first layer out of a plurality of layers and a negative latent vector to remaining layers out of the plurality of layers. Responsive to receiving a first image selection, identifying the first layer out of the plurality of layers generated the first image includes a feature of interest. Determining a feature expression variation for the feature of interest in latent space based on the ranking of a first plurality of generated sample images, wherein the first plurality of generated sample images includes the feature of interest. Responsive to receiving a second plurality of images for a plurality of base features, generating an initial design based on the feature of interest and the plurality of base features.

A computing device for handling anomaly detection, comprises an encoder, for receiving an input image, to generate a first latent vector comprising a semantic latent vector and a visual appearance latent vector according to the input image and at least one first parameter of the encoder; and a training module, coupled to the encoder, for receiving the input image and the first latent vector, to update the at least one first parameter according to the input image and the first latent vector and a loss function.

An electronic device is provided. The electronic device includes an inputter configured to obtain an input sentence in a first language, a memory, and a processor, and the processor is configured to obtain a feature vector corresponding to the input sentence by inputting the input sentence to an encoder model, obtain a first latent vector by inputting the feature vector and a specific integer to an intermediate network, obtain information on a first output sentence in a second language different from the first language by inputting the first latent vector to a decoder model, obtain a second latent vector by inputting the feature vector and the information on the first output sentence to the intermediate network, and obtain information on a second output sentence in the second language by inputting the second latent vector to the decoder model.

The present invention proposes a personalized scenic spot recommendation method based on knowledge graph and user's long-term and short-term preferences, including: preprocessing the tourist's historical scenic spot sequence and performing scenic spot-coding conversion; using node2vec random walk to obtain the scenic spot sequence, using word2vec The Skip-gram model in to obtain the feature vectors of tourists and scenic spots; the feature vectors of scenic spots plus bias will be used as the input of the GRU network, and then use the GRU network to train and output the potential vector of each scenic spot; for each scenic spot Assign different weights, multiply the weight of each scenic spot with the potential vector of the scenic spot and accumulate to obtain the long-term preference of the current tourist, and then splicing the long-term preference of the current tourist with the current preference of the tourist and multiplying by the weight to obtain the final vector; Perform the dot product operation between the final vector and the current preferences of tourists to obtain the estimated rating of the scenic spots, and normalize the estimated ratings of the scenic spots to get the predicted probability of each scenic spot, and take the scenic spots corresponding to the top K scores to get the top_k scenic spot recommendations list.