Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a deep neural network. One of the methods for training a deep neural network that includes a low rank hidden input layer and an adjoining hidden layer, the low rank hidden input layer including a first matrix A and a second matrix B with dimensions i×m and m×o, respectively, to identify a keyword includes receiving a feature vector including i values that represent features of an audio signal encoding an utterance, determining, using the low rank hidden input layer, an output vector including o values using the feature vector, determining, using the adjoining hidden layer, another vector using the output vector, determining a confidence score that indicates whether the utterance includes the keyword using the other vector, and adjusting weights for the low rank hidden input layer using the confidence score.

Abstract: Apparatuses, systems, methods, and computer program products are presented for performing data analytics using machine learning. An unsupervised learning module is configured to assemble an unstructured data set into multiple versions of an organized data set. A supervised learning module is configured to generate one or more machine learning ensembles based on each version of multiple versions of an organized data set and to determine which machine learning ensemble exhibits a highest predictive performance.

Abstract: Provided are methods and systems for constructing a personal knowledge graph for a user based on data contained in existing e-mail messages of the user, and using the personal knowledge graph to provide the user with contextually-relevant content and/or contact suggestions while the user is composing an e-mail message. A personal knowledge graph is constructed based on relations/connections between users and content identified from data contained in e-mail messages sent and/or received by the user. Such relations include content-content relations, user-content relations, and user-(content)-user relations. When a user is composing an e-mail message, the system responsively processes, analyzes, and indexes composing e-mail message data. The composing e-mail message data is used to fetch relevant information from the user's personal knowledge graph and generate one or more content and/or contact suggestions for presentation to the user alongside an e-mail message composing view.

Abstract: An automaton hardware engine employs a transition table organized into 2n rows, where each row comprises a plurality of n-bit storage locations, and where each storage location can store at most one n-bit entry value. Each row corresponds to an automaton state. In one example, at least two NFAs are encoded into the table. The first NFA is indexed into the rows of the transition table in a first way, and the second NFA is indexed in to the rows of the transition table in a second way. Due to this indexing, all rows are usable to store entry values that point to other rows.

Abstract: A media control system enables a device-agnostic and source-agnostic entertainment experience through use of an internet-enabled user device. The user device includes a client application for navigating through media or entertainment content, controlling media devices according to a type of media content selected by the user, and sharing media experiences via social networks. The user device includes smartphones, tablet computers, and other internet-enabled processor-based devices. The media control system leverages the internet access of the user device to enable search and discovery of all available media content. A recommendation engine coupled to the client application learns media preferences from user behavior, generates from numerous disparate media sources recommended media choices corresponding to the media preferences, and presents the recommended media choices on the user device.

Abstract: A quantum computing device that includes a plurality of semiconductor adiabatic qubits is described herein. The qubits are programmed with local biases and coupling terms between qubits that represent a problem of interest. The qubits are initialized by way of a tuneable parameter, a local tunnel coupling within each qubit, such that the qubits remain in a ground energy state, and that initial state is represented by the qubits being in a superposition of |0> and |1> states. The parameter is altered over time adiabatically or such that relaxation mechanisms maintain a large fraction of ground state occupation through decreasing the tunnel coupling barrier within each qubit with the appropriate schedule. The final state when tunnel coupling is effectively zero represents the solution state to the problem represented in the |0> and |1> basis, which can be accurately read at each qubit location.

Abstract: Systems and methods of pre-launching applications in a computer system, said applications being likely to be activated by a user from a terminated and/or suspended process state, are disclosed. The pre-launching of an application may be based on the assessed probability of the application being activated—as well as the level of availability of system resources to affect such pre-launching. Applications may be pre-launched based on these and other conditions/considerations, designed to improve the user's experience of a quick launch of applications in the background. Several prediction models are presented to provide a good estimate of the likelihood of an application being activated by a user. Such prediction models may comprise an adaptive predictor (based on past application usage situations) and/or a switch rate predictor (based on historic data of an application being switched and, possibly, having a decay rate applied to such switch rate measure).

Abstract: Various techniques for chronology based content processing are disclosed herein. For example, in one embodiment, a method includes receiving a content containing a plurality of facts and determining a subject and a chronological value for each of the facts contained in the content. The subject includes at least one of a person, a place, an object, or an event, and wherein the chronological value includes at least a date. The method also includes assembling the facts in the content based on the determined subjects and corresponding chronological values for the individual facts.

Abstract: In one exemplary embodiment, a method of a computerized media-content recommender includes receiving a user-judgment score based on an historical user-listening data with respect to a media content. A first prediction score for a user with respect to the media content is calculated with a media-content recommender. The media-content recommender includes a first set of prediction parameters. A first prediction error including a difference between the user-judgment score and the first prediction score is determined. At least one parameter value of the first set of prediction parameters is modified with a machine-learning optimization technique to generate a second set of prediction parameters. A second prediction score for the user with respect to the media content is calculated with a media-content recommender. A second prediction error including a difference between the user-judgment score and the second prediction score is calculated.

Abstract: Apparatus and methods for developing robotic controllers comprising parallel networks. In some implementations, a parallel network may comprise at least first and second neuron layers. The second layer may be configured to determine a measure of discrepancy (error) between a target network output and actual network output. The network output may comprise control signal configured to cause a task execution by the robot. The error may be communicated back to the first neuron layer in order to adjust efficacy of input connections into the first layer. The error may be encoded into spike latency using linear or nonlinear encoding. Error communication and control signal provision may be time multiplexed so as to enable target action execution. Efficacy associated with forward and backward/reverse connections may be stored in individual arrays. A synchronization mechanism may be employed to match forward/reverse efficacy in order to implement plasticity.

Abstract: Transient survivability metrics are used to select improvements to distribution automation network designs. The approach combines survivability analysis and power flow analysis to assess the survivability of the distribution power grid network. Available investment decisions are then automatically optimized with respect to survivability and investment costs.

Abstract: A system and method is provided that processes a training database of human-generated requests in each of a plurality of task categories with a machine learning algorithm to develop a task classifier model that may be applied to subsequent user requests to determine the most likely one of the task categories for the subsequent user request.

Abstract: A remote processor interacts with a transactional memory that has a memory, local BWC (Byte-Wise Compare) resources, and local NFA (Non-deterministic Finite Automaton) engine resources. The processor causes a byte stream to be transferred into the transactional memory and into the memory. The processor then uses the BWC circuit to find a character signature in the byte stream. The processor obtains information about the character signature from the BWC circuit, and based on the information uses the NFA engine to process the byte stream starting at a byte position determined based at least in part on the results of the BWC circuit. From the time the byte stream is initially written into the transactional memory until the time the NFA engine completes, the byte stream is not read out of the transactional memory.

Abstract: Troubleshooting a technical problem on a user device using a network-based remediation platform. Receiving problem statements relating to technical problems associated with a user device, activating a domain of cases, assigning a score for the cases based on a scoring algorithm, and determining one or more remediation actions to suggest based on the score.

Abstract: A method, computer-readable storage device and apparatus for determining a regression of a medical condition are disclosed. For example, a method receives characteristics of motion information, wherein the characteristics of motion information is based upon gait information, compares the characteristics of motion information over a time period to a profile of the medical condition, wherein the profile of the medical condition comprises a plurality of signatures associated with different stages of the medical condition, determines a potential presence of the regression of the medical condition when the characteristics of motion information matches one of the plurality of signatures, and transmits a notification of the potential presence of the regression of the medical condition.

Abstract: A method, computer-readable storage device and apparatus for calculating a health quality measure are disclosed. For example, a method receives characteristics of motion information, wherein the characteristics of motion information is based upon gait information, monitors the characteristics of motion information over a time period to determine a plurality of different modes of motion within the time period, and calculates the health quality measure based upon the plurality of different modes of motion.

Abstract: In one embodiment, a request is received from a requesting node in a network to assist in distributing a task of the requesting node. Upon receiving the message, a capability to perform the task of one or more helping nodes in the network is evaluated, and a helping node of the one or more helping nodes is selected to perform the task based on the evaluated capability of the selected helping node. The distribution of the task is then authorized from the requesting node to the selected helping node.

Abstract: A method for determining a correction characteristic curve for adapting a characteristic curve of an injection system, in which the correction characteristic curve includes at least one deviation of a measured characteristic curve from a setpoint characteristic curve, the at least one deviation including a sum tolerance of at least two components of the injection system, which have an effect on the characteristic curve.

Abstract: Systems and methods are disclosed for customizing, distributing and processing audio fingerprint data. An example method includes receiving, at a first device, an activation signal and a first audio fingerprint via first wireless communications between the first device and a communications network, the receiving occurring while the first device is not recording audio via a microphone of the first device; based on the activation signal, recording audio using the microphone during a first time period; generating a second audio fingerprint representative of the recorded audio; determining whether the second audio fingerprint matches the first audio fingerprint; and sending an indication of whether the second audio fingerprint matches the first audio fingerprint to an audience measurement entity via second wireless communications between the first device and the communications network.

Abstract: An apparatus and a method for determining careless driving are provided and determine more reliable careless driving by generating normal driving patterns using driving performance data for a reference time at the beginning of driving. In addition, careless driving patterns greater than a predetermined number are detected using the normal driving pattern and a boundary between the normal driving and the careless driving is determined using a supervised learning method. The careless driving of the driver is then determined based on the determined boundary.