Abstract: According to an embodiment, a learning device includes one or more processors. The processors calculate a latent vector of each of a plurality of first target data, by using a parameter of a learning model configured to output a latent vector indicating a feature of a target data. The processors calculate, for each first target data, first probabilities that the first target data belongs to virtual classes on an assumption that the plurality of first target data belong to the virtual classes different from each other. The processors update the parameter such that a first loss of the first probabilities, and a second loss that is lower as, for each of element classes to which a plurality of elements included in each of the plurality of first target data belong, a relation with another element class is lower, become lower.

Abstract: In one embodiment, a first device may receive, from a second device, a reference landmark map identifying locations of facial features of a user of the second device depicted in a reference image and a feature map, generated based on the reference image, representing an identity of the user. The first device may receive, from the second device, a current compressed landmark map based on a current image of the user and decompress the current compressed landmark map to generate a current landmark map. The first device may update the feature map based on a motion field generated using the reference landmark map and the current landmark map. The first device may generate scaling factors based on a normalization facial mask of pre-determined facial features of the user. The first device may generate an output image of the user by decoding the updated feature map using the scaling factors.

Abstract: A plurality of on-road vehicles moving along streets of a city, while utilizing onboard cameras for autonomous driving functions, are bound to capture, over time, random images of various city objects such as structures and street signs. Images of a certain city object may be captured by many of the vehicles at various times while passing by that object, thereby resulting in a corpus of imagery data containing various random images and other representations of that certain object. Per each of at least some of the city objects appearing more than once in the corpus of imagery data, appearances are detected and linked, and then analyzed to uncover behaviors associated with that city object, thereby eventually, over time, revealing various city dynamics associated with many city objects.

Abstract: In the method, a predetermined first number of sensed positions (Pi) of the object (20) are provided, the positions being provided with respect to a vehicle coordinate system (K) of the vehicle (10). Depending on a predetermined second number of the provided positions, a second coordinate system (K?) is determined. The provided positions are transformed into the second coordinate system (K?) and a function is determined which approximates a course of the provided positions in the second coordinate system (K?). For a predetermined third number of sampling points of the function, at least one analytical property of the function is determined at the respective sampling point. The coordinates and the at least one analytical property of the respective sampling point are transformed into the vehicle coordinate system (K).

Abstract: An information processing method implemented by a computer includes: obtaining a piece of first data, and a piece of second data not included in a training dataset for training an inferencer; calculating, using a piece of first relevant data obtained by inputting the first data to the inferencer trained by machine learning using the training dataset, a first contribution representing contributions of portions constituting the first data to a piece of first output data output by inputting the first data to the inferencer; calculating, using a piece of second relevant data obtained by inputting the second data to the inferencer, a second contribution representing contributions of portions constituting the second data to a piece of second output data output by inputting the second data to the inferencer; and determining whether to add the second data to the training dataset, according to the similarity between the first and second contributions.

Abstract: Methods and systems for capturing motion and/or determining the shapes and positions of one or more objects in 3D space utilize cross-sections thereof. In various embodiments, images of the cross-sections are captured using a camera based on reflections therefrom or shadows cast thereby.

Abstract: A method for detecting real lateral locations of target plants includes: recording an image of a ground area at a camera; detecting a target plant in the image; accessing a lateral pixel location of the target plant in the image; for each tool module in a set of tool modules arranged behind the camera and in contact with a plant bed: recording an extension distance of the tool module; and recording a lateral position of the tool module relative to the camera; estimating a depth profile of the plant bed proximal the target plant based on the extension distance and the lateral position of each tool module; estimating a lateral location of the target plant based on the lateral pixel location of the target plant and the depth profile of the plant bed surface proximal the target plant; and driving a tool module to a lateral position aligned with the lateral location of the target plant.

Abstract: A method for training a machine learning model to recognize an object topology of an object from an image of the object. The method includes: obtaining a 3D model of the object; determining a descriptor component value for each vertex of the grid; generating training data image pairs each having a training input image and a target image. The target image is generated by determining the vertex positions in the training input image; assigning the descriptor component value determined for the vertex at the vertex position to the position in the target image; and adapting at least some of the descriptor component values assigned to the positions in the target image or adding descriptor component values to the positions of the target image.

Abstract: Systems and methods for object detection generate a feature pyramid corresponding to image data, and rescaling the feature pyramid to a scale corresponding to a median level of the feature pyramid, wherein the rescaled feature pyramid is a four-dimensional (4D) tensor. The 4D tensor is reshaped into a three-dimensional (3D) tensor having individual perspectives including scale features, spatial features, and task features corresponding to different dimensions of the 3D tensor. The 3D tensor is used with a plurality of attention layers to update a plurality of feature maps associated with the image data. Object detection is performed on the image data using the updated plurality of feature maps.

Abstract: Techniques for anomaly detection are described. An exemplary method includes receiving one or more requests to train an anomaly detection machine learning model using feedback-based training, the request to indicate one or more of a type of analysis to perform, a model selection indication, and a configuration for a training dataset; training the anomaly detection machine learning model according to the one or more requests using the training data; performing feedback-based training on the trained anomaly detection machine learning model; and using the retrained anomaly detection machine learning model.

Abstract: In a case where the operation program is started, a CPU of the mini-batch learning apparatus functions as a calculation unit, a specifying unit, and an evaluation unit. The calculation unit calculates an area ratio of each of a plurality of classes in mini-batch data. The specifying unit specifies, as a correction target class, a rare class of which the area ratio is lower than a setting value. The evaluation unit evaluates the class determination accuracy of the machine learning model by using a loss function. As correction processing, the evaluation unit sets a weight for a loss value of the rare class to be larger than a weight for a loss value of a class other than the rare class.

Abstract: Techniques for generating machine learning training data which corresponds to one or more downstream tasks are disclosed. In one example, a computer implemented method comprises generating one or more synthetic data instances for training a machine learning model, and determining a value of respective ones of the one or more synthetic data instances with respect to at least one task. One or more additional synthetic data instances for training the machine learning model are generated based at least in part on the values of the respective ones of the one or more synthetic data instances.

Abstract: A long-term object tracker employs a continuous learning framework to overcome drift in the tracking position of a tracked object. The continuous learning framework consists of a continuous learning module that accumulates samples of the tracked object to improve the accuracy of object tracking over extended periods of time. The continuous learning module can include a sample pre-processor to refine a location of a candidate object found during object tracking, and a cropper to crop a portion of a frame containing a tracked object as a sample and to insert the sample into a continuous learning database to support future tracking.

Abstract: Speed of first work is compared with speed of second work based on a first working period when a worker is caused to perform the first work of setting annotation data to first image data and a second working period when the worker is caused to perform the second work of correcting advance annotation data set based on a recognition result obtained by causing a predetermined recognizer to recognize the first image data, and, in a case where the first work is faster than the second work, the worker is requested to correct second image data in which advance annotation data is not set, while, in a case where the second work is faster than the first work, the worker is requested to correct advance annotation data set based on a recognition result obtained by causing the recognizer to recognize the second image data.

Abstract: A system is provided that considers allele fraction shifts as a function of copy number and clonal heterogeneity. The system leverages differences between allele frequencies to differentiate between somatic and normal variants in impure tumor samples. In solid tumors, stromal cells and infiltrating lymphocytes are typically interspersed among the tumor cells. The normal cell contamination in tumors can be leveraged to differentiate somatic from germline variants. We explicitly model allelic copy number and clonal sample fractions so that we can examine how these factors impact the power to detect somatic variants. The system models the copy number alterations, which can also affect the allele frequencies of both somatic and germline variants. The expected allele frequencies can be calculated. The expected allele frequencies for somatic and germline differ with tumor content for different copy number alterations.

Abstract: Improved (e.g., high-throughput, low-noise, and/or low-artifact) X-ray Microscopy images are achieved using a deep neural network trained via an accessible workflow. The workflow involves selection of a desired improvement factor (x), which is used to automatically partition supplied data into two or more subsets for neural network training. The neural network is trained by generating reconstructed volumes for each of the subsets. The neural network can be trained to take projection images or reconstructed volumes as input and output improved projection images or improved reconstructed volumes as output, respectively. Once trained, the neural network can be applied to the training data and/or subsequent data—optionally collected at a higher throughput—to ultimately achieve improved de-noising and/or other artifact reduction in the reconstructed volume.

Abstract: An automated method of inspecting a pipe includes: positioning the pipe with respect to a laser scanner using a positioning apparatus; scanning a size of the positioned pipe by the laser scanner; identifying a specification and historical data of the pipe's type by inputting the scanned size to an artificially intelligent module trained through machine learning to match input size data to standardized pipe types and output corresponding specifications and historical data of the pipe types; scanning dimensions of the positioned pipe by the laser scanner using a dimension portion of the identified historical data; comparing the scanned dimensions with standard dimensions from the identified specification; detecting a dimension nonconformity when the scanned dimensions are not within acceptable tolerances of the standard dimensions; and in response to detecting the dimension nonconformity, generating an alert and updating the dimension portion of the identified historical data to reflect the detected dimension n

Abstract: Systems and methods are provided for identifying a product in an image and outputting stock keeping units of the product. The system comprises three main components: a database server, a data analytics system and a standard dashboard. The database server contains real-time inventory images as well as historical images of each product type. The data analytics system is executed by a computer processor configured to apply a multi-head self-supervised learning-based classifier to detect product information captured by the image. The data analytics system is also configured to determine hierarchical classification categories for the product. The standard dashboard is configured to output a report regarding the product information.

Abstract: Systems and methods are disclosed for parsing resume documents using computer vision and optical character recognition technology in combination with a user feedback interface system to facilitate user feedback to improve the overall processing quality of the resumes that are imported into computer resume processing systems. In at least one embodiment, the system and method prompt a user to upload an input resume document, which is processed with a first parsing pass to generate initial resume data by extracting a plurality of resume text blocks. Further processing identifies an initial set of bounding blocks and to visually displays the initial resume data for user review and feedback to regroup one or more of the initial set of bounding blocks into a regrouped bounding block. Additional processing consolidates into a group text block each of the resume text blocks corresponding to the regrouped one or more of the initial set of bounding blocks.

Abstract: An artificial intelligence (AI) system utilizing a machine learning algorithm such as deep learning for controlling an electronic device when a video is reproduced and a user's voice instruction is received, to acquire a frame corresponding to the time point when the input of the user's voice instruction is received, and obtain a search result for information on objects in the frame using an AI model trained according to at least one of machine learning, a neural network or a deep learning algorithm.