Abstract: An apparatus for baseline estimation in input signal data is configured to retrieve input signal data (I(xi)) and to subtract baseline estimation data (ƒ(xi)) from the input signal data (I(xi)) to compute output signal data. The apparatus is further configured to compute the baseline estimation data (ƒ(xi)) from a convolution using a discrete Green's function (G(xi)).

Abstract: Mechanisms are provided to implement a machine learning training model. The machine learning training model trains an image generator of a generative adversarial network (GAN) to generate medical images approximating actual medical images. The machine learning training model augments a set of training medical images to include one or more generated medical images generated by the image generator of the GAN. The machine learning training model trains a machine learning model based on the augmented set of training medical images to identify anomalies in medical images. The trained machine learning model is applied to new medical image inputs to classify the medical images as having an anomaly or not.

Abstract: Systems and methods described herein relate to training a machine-learning-based monocular depth estimator. One embodiment selects a virtual image in a virtual dataset, the virtual dataset including a plurality of computer-generated virtual images; generates, from the virtual image in accordance with virtual-camera intrinsics, a point cloud in three-dimensional space based on ground-truth depth information associated with the virtual image; reprojects the point cloud back to two-dimensional image space in accordance with real-world camera intrinsics to generate a transformed virtual image; and trains the machine-learning-based monocular depth estimator, at least in part, using the transformed virtual image.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating a synthesized signal. In some implementations, a computer-implemented system obtains generator input data including at least an input signal having one or more first characteristics, processes the generator input data to generate output data including a synthesized signal having one or more second characteristics using a generator neural network, and outputs the synthesized signal to a device. The generator neural network is trained, based on a plurality of training examples, with a discriminator neural network.

Abstract: Systems and methods for human detection are provided. The system aligns image level features between a source domain and a target domain based on an adversarial learning process while training a domain discriminator. The target domain includes humans in one or more different scenes. The system selects, using the domain discriminator, unlabeled samples from the target domain that are far away from existing annotated samples from the target domain. The system selects, based on a prediction score of each of the unlabeled samples, samples with lower prediction scores. The system annotates the samples with the lower prediction scores.

Abstract: A method can include receiving multi-channel time series data of drilling operations; training a deep neural network (DNN) using the multi-channel time series data to generate a trained deep neural network as part of a computational simulator where the deep neural network includes at least one recurrent unit; simulating a drilling operation using the computational simulator to generate a simulation result; and rendering the simulation result to a display.

Abstract: Systems and methods for lane marking and road sign recognition are provided. The system aligns image level features between a source domain and a target domain based on an adversarial learning process while training a domain discriminator. The target domain includes one or more road scenes having lane markings and road signs. The system selects, using the domain discriminator, unlabeled samples from the target domain that are far away from existing annotated samples from the target domain. The system selects, based on a prediction score of each of the unlabeled samples, samples with lower prediction scores. The system annotates the samples with the lower prediction scores.

Abstract: Systems and methods for obstacle detection are provided. The system aligns image level features between a source domain and a target domain based on an adversarial learning process while training a domain discriminator. The target domain includes one or more road scenes having obstacles. The system selects, using the domain discriminator, unlabeled samples from the target domain that are far away from existing annotated samples from the target domain. The system selects, based on a prediction score of each of the unlabeled samples, samples with lower prediction scores. The system annotates the samples with the lower prediction scores.

Abstract: The present invention relates to a method for determining a data item's membership in a database, the method comprising: a supervised training phase to obtain three trained neural networks, a phase of preparing the database by application of the first trained network to each data item of the base, and a utilization phase comprising the step of: using the first network on the data item, obtaining a binary value representative of the identity between the data item and a data item of the base by application of the third network, and selecting of those data items of the database for which the binary value obtained corresponds to an identity between the data item and the data items.

Abstract: A task-based learning using task-directed prediction network can be provided. Training data can be received. Contextual information associated with a task-based criterion can be received. A machine learning model can be trained using the training data. A loss function computed during training of the machine learning model integrates the task-based criterion, and minimizing the loss function during training iterations includes minimizing the task-based criterion.

Abstract: Systems and methods for construction zone segmentation are provided. The system aligns image level features between a source domain and a target domain based on an adversarial learning process while training a domain discriminator. The target domain includes construction zones scenes having various objects. The system selects, using the domain discriminator, unlabeled samples from the target domain that are far away from existing annotated samples from the target domain. The system selects, based on a prediction score of each of the unlabeled samples, samples with lower prediction scores. The system annotates the samples with the lower prediction scores.

Abstract: A method for training a deep neural network according to an embodiment includes training a deep neural network model using a first data set including a plurality of labeled data and a second data set including a plurality of unlabeled data, assigning a ground-truth label value to some of the plurality of unlabeled data, updating the first data set and the second data set such that the data to which the ground-truth label value is assigned is included in the first data set, and further training the deep neural network model using the updated first data set and the updated second data set.

Abstract: Techniques are generally described for generation of photorealistic synthetic image data. A generator network generates first synthetic image data. A first class of image data represented by a first portion of the first synthetic image data is detected and the first portion is sent to a first discriminator network. The first discriminator network generates a prediction of whether the first portion of the first synthetic image data is synthetically generated. A second class of image data represented by a second portion of the first synthetic image data is detected and the second portion is sent to a second discriminator network. The second discriminator network generates a prediction of whether the second portion of the first synthetic image data is synthetically generated. The generator network is updated based on the predictions of the discriminators.

Abstract: The present invention provides artificial intelligence technology which has machine-learning-based information understanding capability, including metric learning providing improved classification performance, classification of an object considering a semantic relationship, understanding of the meaning of a scene based on the metric learning and the classification, and the like. An electronic device according to one embodiment of the present invention comprises a memory in which at least one instruction is stored, and a processor for executing the stored instruction. Here, the processor extracts feature data from training data of a first class, obtains a feature point by mapping the extracted feature data to an embedding space, and makes an artificial neural network learn in a direction for reducing a distance between the obtained feature point and an anchor point.

Abstract: Various techniques are provided for training a neural network to classify images. A convolutional neural network (CNN) is trained using training dataset comprising a plurality of synthetic images. The CNN training process tracks image-related metrics and other informative metrics as the training dataset is processed. The trained inference CNN may then be tested using a validation dataset of real images to generate performance results (e.g., whether a training image was properly or improperly labeled by the trained inference CNN). In one or more embodiments, a training dataset and analysis engine extracts and analyzes the informative metrics and performance results, generates parameters for a modified training dataset to improve CNN performance, and generates corresponding instructions to a synthetic image generator to generate a new training dataset. The process repeats in an iterative fashion to build a final training dataset for use in training an inference CNN.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer readable media for generating shift-resilient neural network outputs based on utilizing a dense pooling layer, a low-pass filter layer, and a downsampling layer of a neural network. For example, the disclosed systems can generate a pooled feature map utilizing a dense pooling layer to densely pool feature values extracted from an input. The disclosed systems can further apply a low-pass filter to the pooled feature map to generate a shift-adaptive feature map. In addition, the disclosed systems can downsample the shift-adaptive feature map utilizing a downsampling layer. Based on the downsampled, shift-adaptive feature map, the disclosed systems can generate shift-resilient neural network outputs such as digital image classifications.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network using consistency measures. One of the methods includes processing a particular training example from a mediator training data set using a first neural network to generate a first output for a first machine learning task; processing the particular training example in the mediator training data set using each of one or more second neural networks, wherein each second neural network is configured to generate a second output for a respective second machine learning task; determining, for each second machine learning task, a consistency target output for the first machine learning task; determining, for each second machine learning task, an error between the first output and the consistency target output corresponding to the second machine learning task; and generating a parameter update for the first neural network from the determined errors.

Abstract: During a training phase, a first machine learning system is trained using actual data, such as multimodal images of a hand, to generate synthetic image data. During training, the first system determines latent vector spaces associated with identity, appearance, and so forth. During a generation phase, latent vectors from the latent vector spaces are generated and used as input to the first machine learning system to generate candidate synthetic image data. The candidate image data is assessed to determine suitability for inclusion into a set of synthetic image data that may be used for subsequent use in training a second machine learning system to recognize an identity of a hand presented by a user. For example, the candidate synthetic image data is compared to previously generated synthetic image data to avoid duplicative synthetic identities. The second machine learning system is then trained using the approved candidate synthetic image data.

Abstract: Various methods for the detection and enhanced visualization of a particular structure or pathology of interest in a human eye are discussed in the present disclosure. An example method to visualize a given pathology (e.g., CNV) in an eye includes collecting optical coherence tomography (OCT) image data of the eye from an OCT system. The OCT image data is segmented to identify two or more retinal layer boundaries located in the eye. The two or more retinal layer boundaries are located at different depth locations in the eye. One of the identified layer boundaries is moved and reshaped to optimize visualization of the pathology located between the identified layer boundaries. The optimized visualization is displayed or stored or for a further analysis thereof.

Abstract: A method for selecting a deep learning network which is optimal for solving an image processing task obtaining a type of the image processing task, selecting a data set according to the type of problem, and dividing selected data set into training data and test data. Similarities between different training data are calculated, and a batch size of the training data is adjusted according to the similarities of the training data. A plurality of deep learning networks is selected according to the type of problem, and the plurality of deep learning networks is trained through the training data to obtain network models. Each of the network models is tested through the test data, and the optimal deep learning network with the best test result is selected from the plurality of deep learning networks appropriate for image processing.

Abstract: Various implementations disclosed herein include devices, systems, and methods that involve federated learning techniques that utilize locally-determined ground truth data that may be used in addition to, or in the alternative to, user-provided ground truth data. Some implementations provide an improved federated learning technique that creates ground truth data on the user device using a second prediction technique that differs from a first prediction technique/model that is being trained. The second prediction technique may be better but may be less suited for real time, general use than the first prediction technique.

Abstract: Approaches for multitask learning as question answering include a method for training that includes receiving a plurality of training samples including training samples from a plurality of task types, presenting the training samples to a neural model to generate an answer, determining an error between the generated answer and the natural language ground truth answer for each training sample presented, and adjusting parameters of the neural model based on the error. Each of the training samples includes a natural language context, question, and ground truth answer. An order in which the training samples are presented to the neural model includes initially selecting the training samples according to a first training strategy and switching to selecting the training samples according to a second training strategy. In some embodiments the first training strategy is a sequential training strategy and the second training strategy is a joint training strategy.

Abstract: The present invention relates to a method for learning class descriptors for the detection and the automatic location of objects in a video, each object belonging to a class of objects from among a set of classes, the method using: a learning base, composed from reference videos and containing annotated frames each comprising one or more labels identifying each object detected in the frames, descriptors associated with these labels and learned previously by a preprocessing neural network from the annotated frames of the learning base, an architecture of neural networks defined by parameters centralized on a plurality of parameter servers, and a plurality of computation entities working in parallel, a method in which, for each class of objects, one of the neural networks of the architecture is trained by using as input data the descriptors and the labels to define class descriptors, each computation entity using, for the computation of the class descriptors, a version of the parameters of the parameter ser

Abstract: Maximum expressivity can be received representing a ratio between maximum and minimum input weights to a neuron of a neural network implementing a weighted real-valued logic gate. Operator arity can be received associated with the neuron. Logical constraints associated with the weighted real-valued logic gate can be determined in terms of weights associated with inputs to the neuron, a threshold-of-truth, and a neuron threshold for activation. The threshold-of-truth can be determined as a parameter used in an activation function of the neuron, based on solving an activation optimization formulated based on the logical constraints, the activation optimization maximizing a product of expressivity representing a distribution width of input weights to the neuron and gradient quality for the neuron given the operator arity and the maximum expressivity. The neural network of logical neurons can be trained using the activation function at the neuron, the activation function using the determined threshold-of-truth.

Abstract: Helper neural network can play a role in augmenting authentication services that are based on neural network architectures. For example, helper networks are configured to operate as a gateway on identification information used to identify users, enroll users, and/or construct authentication models (e.g., embedding and/or prediction networks). Assuming, that both good and bad identification information samples are taken as part of identification information capture, the helper networks operate to filter out bad identification information prior to training, which prevents, for example, identification information that is valid but poorly captured from impacting identification, training, and/or prediction using various neural networks. Additionally, helper networks can also identify and prevent presentation attacks or submission of spoofed identification information as part of processing and/or validation.

Abstract: The present disclosure relates to method and system for performing classification of real-time input sample using compressed classification model. Classification system receives classification model configured to classify training input sample. Relevant neurons are identified from neurons of the classification model. Classification error is identified for each class. Reward value is determined for the relevant neurons based on relevance score of each neuron and the classification error. Optimal image is generated for each class based on the reward value of the relevant neurons. The optimal image is provided to the classification model for generating classification error vector for each class. The classification error vector is used for identifying pure neurons from the relevant neurons. A compressed classification model comprising the pure neurons is generated. The generated compressed classification model is used for performing the classification of real-time input sample.

Abstract: An electronic apparatus is disclosed.

Abstract: Embodiments of the present disclosure disclose a target detecting method and apparatus, a training method, an electronic device, and a medium. The target detecting method includes: separately extracting, by means of a neural network, characteristics of a template frame and a detection frame, where the template frame is a detection box image of a target object, and the template frame is smaller than the detection frame in image size; obtaining a classification weight and a regression weight of a local region detector based on the characteristic of the template frame; inputting the characteristic of the detection frame into the local region detector to obtain classification results and regression results of multiple alternative boxes output by the local region detector; and obtaining a detection box for the target object in the detection frame according to the classification results and regression results of the multiple alternative boxes output by the local region detector.

Abstract: A method of performing a convolutional operation in a convolutional neural network includes: obtaining input activation data quantized with a first bit from an input image; obtaining weight data quantized with a second bit representing a value of a parameter learned through the convolutional neural network; binarizing each of the input activation data and the weight data to obtain a binarization input activation vector and a binarization weight vector; performing an inner operation of the input activation data and weight data based on a binary operation with respect to the binarization input activation vector and the binarization weight vector and distance vectors having the same length as each of the first bit and the second bit, respectively; and storing a result obtained by the inner operation as output activation data.

Abstract: The present disclosure provides a computation method and product thereof. The computation method adopts a fusion method to perform machine learning computations. Technical effects of the present disclosure include fewer computations and less power consumption.

Abstract: The present disclosure provides a computation method and product thereof. The computation method adopts a fusion method to perform machine learning computations. Technical effects of the present disclosure include fewer computations and less power consumption.

Abstract: Embodiments of the innovation relate to an emotional quality estimation device comprising a controller having a memory and a processor, the controller configured to execute a training engine with labelled training data to train a neural network and generate a classroom analysis machine, the labelled training data including historical video data and an associated classroom quality score table; receive a classroom observation video from a classroom environment; execute the classroom analysis machine relative to the classroom observation video from the classroom environment to generate an emotional quality score relating to the emotional quality of the classroom environment; and output the emotional quality score for the classroom environment.

Abstract: A method and an apparatus for outputting data are provided. The method includes: obtaining a set of human-face key point data, where the human-face key point data characterizes a position of a key point of a human face in a target human-face image; determining human-eye feature data for characterizing a shape feature of a human eye, based on the set of the human-face key point data; and inputting the human-eye feature data into a human-eye size recognition model obtained by pre-training to obtain a degree value for characterizing a size of the human eye, and outputting the degree value. The human-eye size recognition model characterizes a correspondence between human-eye feature data and a degree value. With the above method, the human-face key point data is effectively utilized to determine the size of the human eye, improving the accuracy of recognizing the size of the human eye.

Abstract: Provided is a data augmentation system including at least one processor, the at least one processor being configured to: input, to a machine learning model configured processor to perform recognition, input data; identify a feature portion of the input data to serve as a basis for recognition by the machine learning model in which the input data is used as input; acquire processed data by processing at least a part of the feature portion; and perform data augmentation based on the processed data.

Abstract: A method ranks image brands. An image brand model is trained to generate an image brand rank from image features. An augmented image brand model is trained to generate an augmented image brand rank from the image brand rank. Predicted financial features are generated from the augmented image brand rank using a feature generation model. A neural network model is trained to generate a predicted augmented image brand rank from the predicted financial features.

Abstract: Systems, devices, methods and instructions are described for detecting GAN generated images. On embodiment involves receiving an images, generating co-occurrence matrices on color channels of the image, generating analysis of the image by using a convolutional neural network trained to analyze image features of the images based on the generated co-occurrence matrices and determining whether the image is a GAN generated image based on the analysis.

Abstract: An image-recognition method is provided. The method includes the following steps: receiving a plurality of check-point images, and classifying the check-point images into a plurality of groups; classifying the check-point images in each group into a plurality of types to generate first structured data, wherein the first structured data includes a first layer and a second layer, and the first layer indicates the groups in different statuses, the second layer is located with directories of the first layer, and the types in each group of the second layer indicate different components in a status corresponding to each group; and balancing a number of the check-point images in each type of each group in the first structured data to generate second structured data, wherein the second structured data is used to train an AI model for image recognition.

Abstract: Described is a system and method for accurate image and/or video scene classification. More specifically, described is a system that makes use of a specialized convolutional-neural network (hereafter CNN) based technique for the fusion of bottom-up whole-image features and top-down entity classification. When the two parallel and independent processing paths are fused, the system provides an accurate classification of the scene as depicted in the image or video.

Abstract: Embodiments of the present disclosure provide a video processing method, a video processing device and a related non-transitory computer readable storage medium. The method includes the following. Frame sequence data of a low-resolution video to be converted is obtained. Pixel tensors of each frame in the frame sequence data are inputted into a pre-trained neural network model to obtain high-resolution video frame sequence data corresponding to the video to be converted output by the neural network model. The neural network model obtains the high-resolution video frame sequence data based on high-order pixel information of each frame in the frame sequence data.

Abstract: A computer-implemented method is provided. The computer-implemented method includes classifying an image using a classification model having a residual network. Classifying the image using the classification model includes inputting an input image into the residual network having N number of residual blocks sequentially connected, N?2, (N?1) number of pooling layers respectively between two adjacent residual blocks of the N number of residual blocks, and (N?1) number of convolutional layers respectively connected to first to (N?1)-th residual blocks of the N number of residual blocks; processing outputs from the first to the (N?1)-th residual blocks of the N number of residual blocks respectively through the (N?1) number of convolutional layers; vectorizing outputs respectively from the (N?1) number of convolutional layers to generate (N?1) number of vectorized outputs; vectorizing an output from a last residual block of the N number of residual blocks to generate a last vectorized output.

Abstract: Disclosed are computer-implemented methods, non-transitory computer-readable media, and systems for identity document face image quality recognition. One computer-implemented method includes pairing, for each user of a plurality of users and to form a pair of face images, an identity document (ID) face image and a live face image. For each pair of face images and based on a face similarity between the ID face image and the live face image, a similarity score for the ID face image is generated. Based on ID face images and similarity scores corresponding to the ID face images, a model for ID face image quality recognition is trained.

Abstract: A method of preprocessing, prior to encoding with an external encoder, image data using a preprocessing network comprising a set of inter-connected learnable weights is provided. At the preprocessing network, image data from one or more images is received. The image data is processed using the preprocessing network to generate an output pixel representation for encoding with the external encoder. The preprocessing network is configured to take as an input display configuration data representing one or more display settings of a display device operable to receive encoded pixel representations from the external encoder. The weights of the preprocessing network are dependent upon the one or more display settings of the display device.

Abstract: A process for reducing time of transmission for single-band, multiple-band or hyperspectral imagery using Machine Learning based compression is disclosed. The process uses Machine Learning to compress single-band, multiple-band and hyperspectral imagery, thereby decreasing the needed bandwidth and storage-capacity requirements for efficient transmission and data storage. The reduced file size for transmission accelerate the communications and reduces the transmission time. This enhances communications systems where there is a greater need for on or near real-time transmission, such as mission critical applications in national security, aerospace and natural resources.

Abstract: A machine vision system comprising receiving means configured to receive image data indicative of an object to be classified where there is provided processing means with an initial neural network, the processing means configured to determine a differential equation describing the initial neural network algorithm based on the neural network parameters, and to determine a solution to the differential equation in the form of a series expansion; and to convert the series expansion to a finite series expansion by limiting the number of terms in the series expansion to a finite number; and to determine the output classification in dependence on the finite series expansion.

Abstract: A training method for video stabilization and an image processing device using the same are proposed. The method includes the following steps. An input video including low dynamic range (LDR) images is received. The LDR images are converted to high dynamic range (HDR) images by using a first neural network. A second neural network for video stabilization is trained to generate stabilized HDR images in a time-dependent manner.

Abstract: A computer-implemented method of determining an explainability mask for classification of an input image by a trained neural network. The trained neural network is configured to determine the classification and classification score of the input image by determining a latent representation of the input image at an internal layer of the trained neural network. The method includes accessing the trained neural network, obtaining the input image and the latent representation thereof and initializing a mask for indicating modifications to the latent representation. The mask is updated by iteratively adjusting values of the mask to optimize an objective function, comprising i) a modification component indicating a degree of modifications indicated by the mask, and ii) a classification score component, determined by applying the indicated modifications to the latent representation and determining the classification score thereof. The mask is scaled to a spatial resolution of the input image and output.

Abstract: This disclosure relates generally to a system and method to identify various products on a plurality of images of various shelves of a retail store to facilitate compliance with respect to planograms. Planogram is a visual plan, which designates the placement of products on shelves and merchandising display fixtures of a retail store. Planograms are used to create consistency between store locations, to provide proper shelf space allocation, to improve visual merchandising appeal, and to create product-pairing suggestions. There are a few assumptions considering one instance per product class is available beforehand and the physical dimension of each product template is available in some suitable unit of length. In case of absence of physical dimension of the products, a context information of the retail store will be used. The context information is that the products of similar shapes or classes are arranged together in the shelves for consumers' convenience.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing images using recurrent attention. One of the methods includes determining a location in the first image; extracting a glimpse from the first image using the location; generating a glimpse representation of the extracted glimpse; processing the glimpse representation using a recurrent neural network to update a current internal state of the recurrent neural network to generate a new internal state; processing the new internal state to select a location in a next image in the image sequence after the first image; and processing the new internal state to select an action from a predetermined set of possible actions.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training an action selection neural network that is used to control a robotic agent interacting with a real-world environment.

Abstract: Systems and methods for generating text corpora comprising realistic optical character recognition (OCR) errors and training language models using the text corpora are provided. An example method comprises: generating, by a computer system, an initial set of images based on an input text corpus comprising text; overlaying, by the computer system, one or more simulated defects over the initial set of images to generate an augmented set of images; generating an output text corpus based on the augmented set of image; and training, using the output text corpus, a language model for optical character recognition.