Abstract: Devices, systems, and methods are provided for real-time object placement guidance in augmented reality experience. An example method may include receiving, by a device having a sensor, an indication of an object to be viewed in an physical environment of the device. The example method may also include determining a 3D model of the physical environment using data of the physical environment captured by the sensor. The example method may also include determining that a first surface in the 3D model of the environment is a first floor space, and a second surface in the 3D model of the environment is a first wall space. The example method may also include determining that a portion of the first surface is unoccupied and sized to fit the object.

Abstract: Embodiments of the present invention provide improved techniques for determining long term relevance and user behavior using query chains. The query chains may first be detected and then annotated into different types of chains based at least in part on various decision rules, machine-learned classifiers, and inter-query relationships. The query chains may then be subsequently used to train models for predicting user behavior and providing more relevant results to a user's queries. A content provider system according to various embodiments may aggregate historical data associated with previous search and/or transaction data, which may be analyzed to detect query chains, for example, whether queries are chained to each other. Determining whether queries are chained to each other may involve incorporating decision rules and reformulation models, analyzing temporal windows between queries, and/or analyzing inter-query relationships.

Abstract: A contextual label compression framework is presented that uses trained sequence to sequence models. A set of training data including received queries and related content can be processed to generate sequences of semantic encodings. These sequences can be used to train the sequence to sequence models, in order to be able to predict queries for instances of content when the relevant information for those instances is processed by the model. When such information is received for an instance, that information can be processed to generate a semantic encoding sequence which can then be processed by the model. A resulting semantic sequence output by the model can be segmented and decoded to produce a set of relevant queries for the instance of content. This information can then be provided to an entity associated with the instance of content for purposes in managing aspects relating to that content.

Abstract: Various embodiments attempt to predict items in a new product category that may be of interest to a particular user when the user has not interacted with or made purchases in that product category. In accordance with various embodiments of the present disclosure, this is accomplished by training machine learning models, such as deep neural networks, using both the historical purchase data from the particular user as well as the historical purchase data from many other users to infer what new category of items the particular user might be interested in. In some embodiments, the previous purchase data of the other users are used to construct a graph, and the graph is used to train the machine learning model rather than the raw data. The graph can represent the entirety of the previous purchase data with less actual data, thereby requiring less memory and other computational resources.

Abstract: Techniques for improved text normalization are provided. Signatures are generated for a first word and a second word using a locality-sensitive hashing technique. A graph is constructed based on the first and second signatures, by creating a first node in the graph for the first word, creating a second node in the graph for the second word, and creating an edge in the graph connecting the first and second nodes upon determining that the first and second signatures match. A mapping from the first word to the second word is then generated based on the graph.

Abstract: The present embodiments provide visual search techniques which produces results that include both accurate similar items as well diversified items through attribute manipulation. In some embodiments, a feature vector describing the item of interest is obtained. A target feature vector is then generated at least partially from the original feature vector, in which the target feature vector shares only a subset of attribute values with the original feature vector and includes at least some values that are different from the original feature vector. An electronic catalog of items is then queried using the target feature vector, and a set of candidate items are determined from the electronic catalog based at least in part on similarity to the target feature vector. The original feature vector may be used to query for a set of similar items that are as similar as possible to the item of interest.

Abstract: Systems and methods are described for determining clusters for large collections of content items. A fast cluster-identifying algorithm can be used to find high density areas where certain less interesting content items might be clustered in a feature space. An example algorithm is a mean shift algorithm. Once these high-density clusters are located, a system can remove them and proceed to analyze the remaining data. Removing these clusters of featureless content items can greatly reduce the collection size and also enhance the overall quality of the collection. Labels can then be applied to clusters and, when a content item is received, classification algorithms can be used to assign an appropriate label to the content item.

Abstract: Systems, methods, and computer-readable media are disclosed for universal anchor points for mobile robots. In one embodiment, an example mobile robot may include a chassis, a removable assembly comprising a support member, and a first anchor point coupled to the chassis. The first anchor point may be configured to anchor the support member to the chassis. The first anchor point may be configured to be coupled to support members of different diameters or widths.

Abstract: Various embodiments provide techniques for generating a virtual wardrobe in which wearing of items can be tracked through computer vision. In various embodiments, computer vision techniques can be used to identify individual clothing items from an image or video of a closet containing a collection of clothes, in which some of the items may be partially obscured, such as positioned behind other items, folded, at an odd viewing angle, shadowed, or the like. Through the present techniques, the clothing items may be automatically cataloged, and utilization of individual items can be tracked automatically based on additional image data of instances in which an item is worn. Various recommendations can then be made based on the attributes of the items in the virtual wardrobe or utilization of the items.

Abstract: Various embodiments enable a computing device to perform tasks such as processing an image to recognize text or an object in an image to identify a particular product or related products associated with the text or object. In response to recognizing the text or the object as being associated with a product available for purchase from an electronic marketplace, one or more advertisements or product listings associated with the product can be displayed to the user. Accordingly, additional information for the associated product can be displayed, enabling the user to learn more about and purchase the product from the electronic marketplace through the portable computing device.

Abstract: Devices, systems, and methods are provided for real-time object placement guidance in augmented reality experience. An example method may include receiving, by a device having a sensor, an indication of an object to be viewed in an physical environment of the device. The example method may also include determining a 3D model of the physical environment using data of the physical environment captured by the sensor. The example method may also include determining that a first surface in the 3D model of the environment is a first floor space, and a second surface in the 3D model of the environment is a first wall space. The example method may also include determining that a portion of the first surface is unoccupied and sized to fit the object.

Abstract: Systems and methods are disclosed for optimizing responses to queries. Analyses of user interactions and other behaviors can lead to association of queries with signals, including ASINs and other product descriptions. The associations can be algorithmically graphed and analyzed on a disaggregated basis and individually weighted to improve search recall and reduce the risk of returning defective search results. Machine learning techniques can further optimize the associations and/or the search results.

Abstract: Systems, methods, and computer-readable media are disclosed for systems and methods to implement preferred pathways in mobile robots. Example methods may include obtaining, via at least one of a user interface and a corresponding Application Programming Interface API call, at least one preferred pathway for an autonomous mobile robot, transmitting the at least one preferred pathway to the autonomous mobile robot, generating a planned path for the autonomous mobile robot based at least in part on an influence function, the influence function being representative of an amount of bias towards the at least one preferred pathway on a motion planning decision of the autonomous mobile robot, the amount of bias being based at least in part on a metric associated with the at least one preferred pathway, and causing the autonomous mobile robot to move from a start point to an end point along the planned path.

Abstract: Technologies are provided for reformulation of a tail query to a head query with the same purchase intent by mapping the tail query to the head query. In some of the technologies, a reasonable embedding can be learned on historical head queries. The embedding can then be refined by leveraging rewards generated from a persistently noisy oracle that compensates for the lack of historical behavioral signal for tail queries. Further, a contextual sampling technique that uses text-based rewards or oracle-based rewards, or both, can be implemented in order to avoid biases introduced by persistent noise in the oracle. Numerical experiments on large scale e-commerce datasets demonstrate that the provided technologies can outperform several conventional approaches to query reformulation.

Abstract: Various embodiments enable a user editing a document to tag product references in the document. These product references can then be recognizable when the document is published online, thereby enabling ad units (or other personalized units associated with the content) to be inserted either at a location associated with a particular product reference or within the document in a designated location. For example, when editing or drafting a document, a client-side tool could enable a user to tag particular words within the text to create a hyperlink to product pages of an electronic marketplace. In another example, a client-side tool could auto-insert a list of words associated with subject matter of the same. Accordingly, mentions of these words in content could be auto-converted into text-links or hyperlinks to a respective items page of the electronic marketplace.

Abstract: Approaches described and suggested herein relate to generating an enhanced point cloud representation of an objection and generating a surface mesh from the enhanced point cloud. The surface mesh can be used to render three-dimensional representations of objects on personal devices such as smartphones and personal computers, for example. Generating an enhanced point cloud of an object includes capturing a plurality of images of the object from a plurality of viewpoints about the object, generating an initial point cloud representation of the object from the plurality of images, generating a preliminary surface mesh from the point cloud using a Delauney-based meshing algorithm, and sampling points from the preliminary surface mesh. The sampled points are then added to the point cloud to form the enhanced point cloud. A final surface mesh can then be generated from the enhanced point cloud using a Poisson-based meshing algorithm.

Abstract: A first router establishes an internet connection between a remote router management platform (RRMP) and the first router via a second router, the second router being part of a local area network. A router identifier of the first router is registered with a user account at the RRMP, the user account being associated with router configuration data for the local area network. The first router sends a request with the router identifier to the RRMP using the internet connection, a request for the router configuration data and receives the router configuration data from the RRMP via the internet connection responsive to the remote router management platform verifying that the first router is registered to the user account. At least a portion of the router configuration data, when executed by the first router, causes the first router to implement the local area network in tandem with the second router.

Abstract: Large scale time series forecasting models are described that leverage deep learning. This can include the utilization of temporal convolution networks and long short-term memory (LTSM) units of recurrent neural networks (RNNS). The model architectures can handle very large data sets with a large number of time series. Diverse scaling is provided through use of a scale-free leveling network architecture, and sparse time-series data is managed using a gating approach. A deep temporally regularized matrix factorization approach to time-series forecasting is utilized that can leverage correlations between the time series during both training and prediction.

Abstract: Various embodiments provide for recommending products from an electronic catalog that are aesthetically compatible with an existing item in a physical space based on a live camera view of the space, and rendering augmented reality views of the recommended products into the live camera view. Specifically, an image of the existing item is used to search a database of images of designed spaces that includes a corresponding object visually similar to the existing item. The image of the designed space may also include other items that are visually compatible with the corresponding object and therefore visually compatible with the existing item. Thus, the images of the other items are used to search an electronic catalog for available items that are visually similar to the other items and therefore would be visually compatible with the existing item.

Abstract: A computing device is used to capture image data of a physical environment. The image data is of a live camera view from the camera. The image data includes a representation of a physical environment. A selection of an aspect of the representation is determined. The image data is analyzed using a trained neural network and using the selection to determine one or more types of items for the representation. At least two items associated with the one or more types of items are generated. The at least two items and at least one visible marker associated with the aspect are overlaid in the live camera view to provide an augmented reality view of the physical environment.