Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network to perform any one or more of a variety of machine learning tasks. For example, the neural network can be configured as a generative neural network, e.g., an autoregressive generative neural network.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network to perform any one or more of a variety of machine learning tasks. For example, the neural network can be configured as a generative neural network, e.g., an autoregressive generative neural network.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a machine learning model using a deterministic data pipeline. One of the methods may include receiving a first request to generate a deterministic training dataset: transforming raw training examples obtained from the raw data source into pre-processed training examples; assigning a unique index to each pre-processed training example; and caching the pre-processed training examples into the cache directory specified in the received first request; receiving a second request to use the deterministic training dataset to train a machine learning model, the second request specifying a start index; and in response to receiving the second request: reading, from the cache directory, the pre-processed training examples that have indices beginning from the start index; and providing the read training examples in an order of the assigned indices for use in training the machine learning model.

Abstract: There is provided a method for providing a target position for a robotic speaker in an environment. The method includes receiving sensor information from one or more sensors related to the environment, obtaining at least one of initiation gesture information, position information, and face direction information based on the sensor information, the face direction information related to one or more users present in the environment, obtaining the target position of the robotic speaker in the environment by performing a calibration process based on the at least one of initiation gesture information, position information, or face direction information and one or more and acoustic properties of the environment, and positioning the robotic speaker at the target position.

Abstract: A method using weighted aggregated ensemble model for energy demand management of buildings includes initializing data values for integrated model to measure energy consumption, perform statistical analysis on data values to estimate accurate prediction, optimizing the data values using marine predator optimization for integrated model, analyze the output to minimize the mean square error and results show improvement in accuracy of integrated model. The data values comprise of ?, maximum number of splits, minimum leaf size, and ?. The weighted aggregated ensemble model for energy demand management of buildings shows best performance compared with other predictive models such as linear regression (LR), support vector regression (SVR), multilayer perceptron neural network (MLPNN), decision tree (DT), and generalized additive model (GAM).

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a machine learning model using a deterministic data pipeline. One of the methods may include receiving a first request to generate a deterministic training dataset: transforming raw training examples obtained from the raw data source into pre-processed training examples; assigning a unique index to each pre-processed training example; and caching the pre-processed training examples into the cache directory specified in the received first request; receiving a second request to use the deterministic training dataset to train a machine learning model, the second request specifying a start index; and in response to receiving the second request: reading, from the cache directory, the pre-processed training examples that have indices beginning from the start index; and providing the read training examples in an order of the assigned indices for use in training the machine learning model.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a machine learning model using a deterministic data pipeline. One of the methods may include receiving a first request to generate a deterministic training dataset: transforming raw training examples obtained from the raw data source into pre-processed training examples; assigning a unique index to each pre-processed training example; and caching the pre-processed training examples into the cache directory specified in the received first request; receiving a second request to use the deterministic training dataset to train a machine learning model, the second request specifying a start index; and in response to receiving the second request: reading, from the cache directory, the pre-processed training examples that have indices beginning from the start index; and providing the read training examples in an order of the assigned indices for use in training the machine learning model.

Abstract: A system and method for editing media content, by embedding graphics, text, images, icons and the like into it. An HTML overlay graphics layer is used to view the additional content in relation to the original media content. The additional content is embedded into the original media content through a rendering process once all the additions have been added by the user. The rendering may be carried out on multiple servers.

Abstract: Using stored information about the content of an electronic document which an end user has accessed, and information about changes made by other users to the electronic document, when the end user accesses a server computer from any device, the server computer can identify changed content in the electronic document which the end user has not yet accessed. When the server computer delivers the electronic document to the device used by the end user, the electronic document can be accompanied by information indicating which content of the electronic document the end user has not yet accessed. An end user application on the device can present the electronic document with information indicating the content which that end user has not yet accessed. By storing this information on the server computer, the end user can access the electronic document from any device and quickly identify the changed content which that end user has not yet accessed.

Abstract: Embodiments of the present disclosure provide a method for identifying an indoor environment location. The method includes an operation of obtaining a visibility map of an indoor environment. The visibility map may include a plurality of static markers in the indoor environment. The method can determine a direction of a first static marker with respect to magnetic north by an electronic device if the electronic device is located to point at the first static marker and can determine a direction of a second static marker with respect to magnetic north if the electronic device is disposed to point at the second static marker. The method can calculate an intersecting point based on the determined first static marker and second static marker and can identify a location of the electronic device in an indoor environment.

Abstract: Systems and methods are disclosed to enable delivering a contextually relevant action for an underlying focal point of a communication (an “entity”) between users over computing devices. Delivery of a contextually relevant action entails identifying the entity and associated descriptors or amplifying words in the communication surrounding the entity, reviewing databases of actions taken with respect to the identified entity and associated descriptors, reviewing the functions and features of platforms and applications supported on users' computing devices, computing correlations between the actions taken and entity involved and computing devices' available functions and features, and selecting a contextually relevant action from the computed correlation. The selected contextually relevant action is displayed simply as an executable action for a user to take or as a description of the entity or as a series of possible executable actions to take.

Abstract: A method of generating data representing a structure of a room on a device, the method including: outputting an audio signal at a first position in the room; acquiring a first reflection signal of the audio signal at the first position; outputting the audio signal at a second position in the room; acquiring a second reflection signal of the audio signal at the second position; estimating distance information about the room by using the acquired first and second reflection signals; acquiring at least one angle information about the room; and generating a floor plan illustrating the structure of the room using the estimated distance information and the acquired angle information.

Abstract: A plurality of data keys are associated with a plurality of type values; query frequencies of the data keys are known. A computer memory is divided into a plurality of tranches, each tranche including a probabilistic or non-probabilistic data structure. The data keys are stored in the tranches in accordance with their query frequencies such that, e.g., frequently queried data keys are stored in data structures having higher accuracy and infrequently queried keys are stored in data structure having less accuracy (and consequently require less memory space).

Abstract: Embodiments of the present disclosure provide a method for identifying an indoor environment location. The method includes an operation of obtaining a visibility map of an indoor environment. The visibility map may include a plurality of static markers in the indoor environment. The method can determine a direction of a first static marker with respect to magnetic north by an electronic device if the electronic device is located to point at the first static marker and can determine a direction of a second static marker with respect to magnetic north if the electronic device is disposed to point at the second static marker. The method can calculate an intersecting point based on the determined first static marker and second static marker and can identify a location of the electronic device in an indoor environment.

Abstract: A method of generating data representing a structure of a room on a device, the method including: outputting an audio signal at a first position in the room; acquiring a first reflection signal of the audio signal at the first position; outputting the audio signal at a second position in the room; acquiring a second reflection signal of the audio signal at the second position; estimating distance information about the room by using the acquired first and second reflection signals; acquiring at least one angle information about the room; and generating a floor plan illustrating the structure of the room using the estimated distance information and the acquired angle information.

Abstract: A method for generating feature graphs employed for creation of a head note in a legal document is provided. The method enables identifying one or more predetermined features in a plurality of legal documents. The one or more predetermined features are based on grammatical constituents of text in the legal document. The plurality of legal documents is manually identified as headnote and non headnote. The method further enables obtaining data related to the availability of the one or more identified predetermined features in the sentences manually identified as headnote and non headnote in the plurality of legal documents. Furthermore, the method enables computing likelihood of a sentence being a headnote based on the obtained data. The method further enables generating feature graphs corresponding to each predetermined feature based on the computed likelihood and obtained data and storing the generated feature graphs in a repository.

Abstract: A plurality of data keys are associated with a plurality of type values; query frequencies of the data keys are known. A computer memory is divided into a plurality of tranches, each tranche including a probabilistic or non-probabilistic data structure. The data keys are stored in the tranches in accordance with their query frequencies such that, e.g., frequently queried data keys are stored in data structures having higher accuracy and infrequently queried keys are stored in data structure having less accuracy (and consequently require less memory space).

Abstract: Disclosed is a system and method for email management that leverages information associated with an incoming email message in order to organize the recipient's inbox and display relevant message content from the inbox. The present disclosure identifies messages by their relevance to a recipient, and effectively displays the relevant portions of the message to the recipient in an efficient manner which allows users to interact with the messages from the inbox. According to some embodiments, the present disclosure categorizes emails and provides a user interface for presenting the emails prioritized by importance or relevance to the recipient. The present disclosure logically categorizes incoming emails and displays enriched snippets extracted from the messages from the inbox without the user having to open the message, thereby allowing the user to view and/or interact with the message from the inbox.

Abstract: A cardinality of an incoming data stream is maintained in real time; the cardinality is maintained in a data structure that is represented by an unsorted list at low cardinalities, a linear counter at medium cardinalities, and a PCSA at high cardinalities. The conversion to the linear counter makes use of the data in the unsorted list, after which that data is discarded. The conversion to the PCSA uses only the data in the linear counter.