A computer, including a processor and a memory, the memory including instructions to be executed by the processor to train a quantile neural network to input an image and output a lower quantile (LQ) prediction, a median quantile (MQ) prediction and an upper quantile (UQ) prediction corresponding to an object in the image, wherein an LQ loss, an MQ loss and a UQ loss are determined for the LQ prediction, the MQ prediction and the UQ prediction respectively and wherein the LQ loss, the MQ loss and the UQ loss are combined to form a base layer loss and output the quantile neural network.

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training an image embedding model. In one aspect, a method comprises: obtaining training data comprising a plurality of training examples, wherein each training example comprises: an image pair comprising a first image and a second image; and selection data indicating one or more of: (i) a co-click rate of the image pair, and (ii) a similar-image click rate of the image pair; and using the training data to train an image embedding model having a plurality of image embedding model parameters.

Methods, systems, and computer program products for learning models for entity resolution using active learning are provided herein. A computer-implemented method includes determining a set of data items related to a task associated with structured knowledge base creation, and outputting the set of data items to a user for labeling. Such a method also includes generating, based on a user-labeled version of the set of data items, a candidate model for executing the task, and one or more generalized versions of the candidate model. Additionally, such a method can also include generating a final model based on one or more iterations of analysis of the candidate model and analysis of the one or more generalized versions of the candidate model, and performing the task by executing the final model on one or more datasets.

A method and apparatus comprising generating a dynamic personalized webpage is disclosed. At least two webpages are loaded in a fashion that is hidden from the user. Content from the at least two webpages is extracted based on classification “of interest” by an artificial intelligence algorithm. A dynamic personalized webpage comprising extracted content is then generated and displayed to the user. In the preferred embodiment, the user's dynamic personalized webpage will be filled with advertisements tailored to the user and the user would receive at least some revenue from advertisements.

A server acquires a feature label vector of each seed user and forms a first number of clusters corresponding different information categories according to the feature label vectors of the seed users. The server calculates a central vector of each cluster according to the feature label vectors of the seed users in the cluster. The server acquires a feature weight vector corresponding to the information categories. The server acquires a feature label vector of each potential user. The server calculates first distances from the potential users to the central vector of the information categories according to the feature label vectors of the potential users, feature weight vectors and central vectors corresponding to the information categories. The server selects a second number of potential users corresponding to the shortest first distances from the first distances and sends them information that is matched with corresponding information categories of the target users.

A computer-implemented tool for automated classification and interpretation of documents, such as life science documents supporting clinical trials, is configured to perform a combination of raw text, document construct, and image analyses to enhance classification accuracy by enabling a more comprehensive machine-based understanding of document content. The combination of analyses provides context for classification by leveraging relative spatial relationships among text and image elements, identifying characteristics and formatting of elements, and extracting additional metadata from the documents as compared to conventional automated classification tools, wherein natural language processing (NLP) is applied to associate text with tokens, and relevant differences and similarities between protocols are identified.

Embodiments of the present disclosure provide systems and methods for training a machine-learning model for predicting emotions from received media data. Methods according to the present disclosure include displaying a user interface. The user interface includes a predefined media content, a plurality of predefined emotion tags, and a user interface control for controlling a recording of the user imitating the predefined media content. Methods can further include receiving, from a user, a selection of one or more emotion tags from the plurality of predefined emotion tags, receiving the recording of the user imitating the predefined media content, storing the recording in association with the selected one or more emotion tags, and training, based on the recording, the machine-learning model configured to receive input media data and predict an emotion based on the input media data.

Example freight management systems and methods are described. In one implementation, techniques receive at least one wide angle camera image from a sensor tower, where the sensor tower is located proximate a loading dock and the wide angle camera image is associated with at least a portion of the loading dock. The techniques also receive multiple high precision camera images from the sensor tower, where the plurality of high precision camera images are associated with at least a portion of the loading dock. The techniques process the wide angle image using a first convolutional neural network (CNN) and process the multiple high precision images using a second CNN. The techniques identify a freight item proximate the loading dock based on the processed high precision images.

In an embodiment, a method includes receiving a trigger of machine learning model generation. In addition, the method includes algorithmically eliminating at least some of rows and at least some of columns of a training dataset, the algorithmically eliminating yielding a size-reduced training dataset. The method additionally includes generating, for a prediction target, a plurality of machine learning models via a plurality of machine learning algorithms. The method also includes measuring prediction accuracies of the plurality of machine learning models relative to the prediction target. Furthermore, the method includes selecting a particular machine learning model. Moreover, the method includes applying the particular machine learning model to a data source.

A vehicle collects video data of an environment surrounding the vehicle including traffic entities, e.g., pedestrians, bicyclists, or other vehicles. The captured video data is sampled and presented to users to provide input on a traffic entity's state of mind. The user responses on the captured video data is used to generate a training dataset. A machine learning based model configured to predict a traffic entity's state of mind is trained with the training dataset. The system determines input video frames and associated dimension attributes for which the model performs poorly. The dimension attributes characterize stimuli and/or an environment shown in the input video frames. The system generates a second training dataset based on video frames that have the dimension attributes for which the model performed poorly. The model is retrained using the second training dataset and provided to an autonomous vehicle to assist with navigation in traffic.