Abstract: Representations of sets of descriptors of reference objects are stored in a repository, with individual descriptors including information about entities identified in the reference objects. In response to a request to extract content from a particular data object, a reference object which satisfies a similarity criterion with respect to the particular data object is identified from the repository using the descriptors. A structural comparison of the particular data object and the reference object is performed to determine an entity related to another entity identified in the particular data object.

Abstract: Root cause estimation for a data set corresponding to customer returns of a product may use a probabilistic model to associate customer-entered product return data with probability distributions relating to possible root causes for the returns. A particular application relates to applying a Bayesian network to customer-selected return reason codes and customer-entered return reason comments to estimate a probability distribution for root causes of a plurality of returns and uncertainties relating to the probability distribution estimation. A bag-of-n-grams can be used to enable the Bayesian network to process natural language portions of the customer-entered product return data. The output of the model and other data relating to the root cause estimation can be conveyed to a seller of the returned products via a user interface.

Abstract: Techniques for performing learnt image compression and object detection using compressed image data are described. A system may perform image compression using an image compression model that includes an encoder, an entropy model, and a decoder. The encoder, the entropy model, and the decoder may be jointly trained using machine learning based on training data. After training, the encoder and the decoder may be separated to encode image data to generate compressed image data or to decode compressed image data to generate reconstructed image data. In addition, the system may perform object detection using a compressed object detection model that processes compressed image data generated by the image compression model. For example, the compressed object detection model may perform partial decoding using a single layer of the decoder and perform compressed object detection on the partially decoded image data.

Abstract: The present application relates to a multi-stage encoder/decoder system that provides image compression using hierarchical auto-regressive models and saliency-based masks. The multi-stage encoder/decoder system includes a first stage and a second stage of a trained image compression network, such that the second stage, based on the image compression performed by the first stage, identify certain redundancies that can be removed from the bit string to reduce the storage and bandwidth requirements. Additionally, by using saliency-based masks, distortions in different sections of the image can be weighted differently to further improve the image compression performance.

Abstract: Techniques for learned lossy image compression are described. A system may perform image compression using an image compression model that includes an encoder to compress an image and a decoder to reconstruct the image. The encoder and the decoder are trained using machine learning techniques. After training, the encoder can encode image data to generate compressed image data and the decoder can decode compressed image data to generate reconstructed image data.

Abstract: At an artificial intelligence system, a neural network model is trained iteratively to generate similarity scores for image pairs. The model includes a first subnetwork with a first number of convolution layers, and a second subnetwork with a different number of convolution layers. A given training iteration includes determining, using a version of the model generated in an earlier iteration, similarity scores for a set of image pairs, and then selecting a subset of the pairs based on the similarity scores. The selected subset is used to train a subsequent version of the model. After the model is trained, it may be used to generate similarity scores for other image pairs, and responsive operations may be initiated if the scores meet a criterion.

Abstract: Techniques are disclosed herein for reactively identifying software products, available from an electronic marketplace, that are exhibiting anomalous behavior. Data associated with software products is accessed and analyzed to determine anomalous behavior. The data analyzed may include, but is not limited to, crash data, ratings data, marketplace data, usage data, and the like. A machine learning mechanism may be used to classify the application into a category relating to whether a potential anomaly is identified for the software product. A score may also be calculated for the software applications that indicates a severity of the anomalous behavior. The classification and/or the score may be used to determine whether to perform further analysis or testing with regard to a software product. For instance, the score may be used to determine that the software product is to be tested by a testing service.

Abstract: Techniques are disclosed herein for identifying software products, available from an electronic marketplace, to be tested. Data associated with software products is accessed and analyzed to determine what software products to test. The data analyzed may include, but is not limited to, download data, crash data, ratings data, marketplace data, usage data, and the like. A machine learning mechanism may be used to predict a popularity of a software product, classify the application into a category relating to whether a potential anomaly is identified for the software product, and determine whether to test the software product. A score may also be calculated for the software products that indicates whether or not to test the software product. The predicted popularity, the classification and/or the score may be used to determine whether to perform further analysis or testing with regard to a software product.

Abstract: A workflow processing system may receive a representation of one or more objects corresponding to the current state of a workflow. A rule may be mapped to one or more activities of the workflow. The rule may comprise an any-of clause, an all-of clause, and a none-of clause. Each clause may contain zero or more conditions that are evaluated with respect to the current state of the workflow. When each of the any-of, all-of, and none-of clauses evaluates to true, the rule may be satisfied and the activities to which the rule is mapped are performed.