Abstract: A user activity pattern may be ascertained using signal data from a set of computing devices. The activity pattern may be used to infer user intent with regards to a user interaction with a computing device or to predict a likely future action by the user. In one implementation, a set of computing devices is monitored to detect user activities using sensors associated with the computing devices. Activity features associated with the detected user activities are determined and used to identify an activity pattern based on a plurality of user activities having similar features. Examples of user activity patterns may include patterns based on time, location, content, or other context. The inferred user intent or predicted future actions may be used to provide improved user experiences, such as personalization, modifying functionality of user devices, or providing more efficient consumption of bandwidth or power.

Abstract: Certain aspects of the present disclosure are directed to methods and apparatus for deep learning in an artificial neural network. One example method generally includes receiving input data at an input to a layer of the neural network; replicating a group of neural processing units in the layer to form a superset of neural processing units, the superset comprising n instances of the group of neural processing units; processing the input data using the superset to generate output data for the layer; and determining an uncertainty of the output data. Processing the input data includes performing a dropout function by zeroing out one or more weights of a set of weights for each of the n instances of the superset of neural processing units and convolving, for each of the n instances in parallel, the input data with one or more non-zeroed out weights of the set of weights.

Abstract: A system and method for validating unsupervised machine learning models. The method includes: analyzing, via unsupervised machine learning, a plurality of sensory inputs associated with a machine, wherein the unsupervised machine learning outputs at least one normal behavior pattern of the machine; generating, based on the at least one normal behavior pattern, at least one artificial anomaly, wherein each artificial anomaly deviates from the at least one normal behavior pattern; injecting the at least one artificial anomaly into the plurality of sensory inputs to create an artificial dataset; and analyzing the artificial dataset to determine whether a candidate model is a valid representation of operation of the machine, wherein analyzing the artificial dataset further comprises running the candidate model using the artificial dataset as an input.

Abstract: A mobile device including: a position locator; a user data engine; and a reputation engine client configured to: receive a location from the position locator; operate the user data engine to provide a user profile, intent, and context data for a user, the context data including dynamic factors about the user, and the profile including relative factors about the user that are relatively static with respect to the context data from the user data engine; and determine a reputation for the location, wherein the reputation is based at least in part on a combination of the user profile, intent, and context.

Abstract: A system such as a service of a computing resource service provider includes executable code that, if executed by one or more processors, causes the one or more processors to initiate a training of a machine-learning model with a parameter for the training having a first value, the training to determine a set of parameters for the model, calculate output of the training, and change the parameter of the training to have a second value during the training based at least in part on the output. Training parameters may, in some cases, also be referred to as hyperparameters.

Abstract: A system is provided for a machine learning engine using clustered case objects in a case management system. The system includes a multi-layer neural network. The system is configured to receive case object data comprising a case object and contextual objects in the case management system associated with the case object, the contextual objects comprising word vectors, generate a context embedding for the case object using the word vectors for the contextual objects, and cluster the case object with other case objects in the case management system based on the context embedding for the case object and other context embeddings for the other case objects.

Abstract: A system of training behavior labeling model is provided. Specifically, a processing unit inputs each data of a training data set into a plurality of learning modules to establish a plurality of labeling models. The processing unit obtains a plurality of second labeling information corresponding to each data of a verification data set and generates a behavior labeling result according to the second labeling information corresponding to each data of the verification data set. The processing unit obtains a labeling change value according to the behavior labeling result and first labeling information corresponding to each data of the verification data set. The processing unit, if determining that the labeling change value is greater than a change threshold, updates the first labeling information according to the behavior labeling results, exchanges the training data set and the verification data set and reestablishes the labeling models.

Abstract: A system and method for generating analytics for entities depicted in multimedia content, including: identifying at least one social pattern based on social linking scores of a plurality of entities indicated in a social linking graph, wherein each social pattern is identified at least by comparing one of the social linking scores to a predetermined social pattern threshold, wherein each social linking score is generated based on contexts of at least one multimedia content element (MMCE) in which at least two of the plurality of entities are depicted, wherein each context is determined based on a plurality of concepts of one of the at least one MMCE, wherein each concept matches at least one signature generated for the at least one MMCE above a predetermined threshold; and generating, based on the identified at least one social pattern, analytics for the plurality of entities depicted in the social linking graph.

Abstract: A pair of records is tokenized to form a normalized representation of an entity represented by each record. The tokens are correlated to a machine learning system by determining whether a learned resolution already exists for the two entities. If not, the normalized records are compared to generate a comparison measure to determine whether the records match. The normalized records can also be used to perform a web search and web search results can be normalized and used as additional records for matching. When a match is found, the records are updated to indicate that they match, and the match is provided to the machine learning system to update the learned resolutions.

Abstract: An interface device and method of use, comprising audio and image inputs; a processor for determining topics of interest, and receiving information of interest to the user from a remote resource; an audio-visual output for presenting an anthropomorphic object conveying the received information, having a selectively defined and adaptively alterable mood; an external communication device adapted to remotely communicate at least a voice conversation with a human user of the personal interface device. Also provided is a system and method adapted to receive logic for, synthesize, and engage in conversation dependent on received conversational logic and a personality.

Abstract: A method for constructing and training a Decentralized Migration Diagram Neutral Network (DMDNN) model for a production process, including: determining a production task input management node, distributed management nodes and a granularity of each of the distributed management nodes, and constructing a production system network; constructing network calculation nodes on each of the distributed management nodes; constructing and training a DMDNN model; and applying the trained DMDNN model in management and control of the production process.

Abstract: A request for an inference from a customer is received at a machine learning (ML) decentralized application (DAPP) platform, where the request includes a data record associated with a user that is associated with the customer. The data record is distributed by the ML DAPP platform among a number of service providers. An inference is received at the ML DAPP platform from each service provider. The received inferences are returned to the customer by the ML DAPP platform.

Abstract: Historical device positioning data captured from one or more devices over a period of time is received. The historical device positioning data includes historical latitude, longitude, and elevation data of the one or more devices. Building boundaries for a give building are identified based upon the historical latitude and longitude data. The historical device positioning data corresponding to locations within the building boundaries of the building is clustered using a machine learning-based clustering algorithm, resulting in clusters with corresponding cluster centroids. The cluster centroids are associated with respective floors within the building. A current floor of the building on which a specific device is located is determined by mapping current device positioning data of the specific device to the closest cluster centroid.

Abstract: Some embodiments provide a neural network inference circuit for implementing a neural network that includes multiple computation nodes at multiple layers. Each of a set of the computation nodes includes (i) a linear function that includes a dot product of input values and weight values and (ii) a non-linear activation function. The neural network inference circuit includes (i) a set of dot product circuits to compute dot products for the plurality of computation nodes and (ii) at least one computation node post-processing circuit to (i) receive a dot product for a computation node computed by the set of dot product circuits, (ii) compute a result of the linear function for the computation node based on the dot product, and (iii) use a lookup table to compute the non-linear activation function of the computation node from the result of the linear function to determine an output of the computation node.

Abstract: Apparatuses and methods of manufacturing same, systems, and methods are described. In one aspect, a method includes generating a convolutional neural network (CNN) by training a CNN having three or more convolutional layers, and performing cascade training on the trained CNN. The cascade training includes an iterative process of one or more stages, in which each stage includes inserting a residual block (ResBlock) including at least two additional convolutional layers and training the CNN with the inserted ResBlock.

Abstract: A quantum processor is operable as a universal adiabatic quantum computing system. The quantum processor includes physical qubits, with at least a first and second communicative coupling available between pairs of qubits via an in-situ tunable superconducting capacitive coupler and an in-situ tunable superconducting inductive coupler, respectively. Tunable couplers provide diagonal and off-diagonal coupling. Compound Josephson junctions (CJJs) of the tunable couplers are responsive to a flux bias to tune a sign and magnitude of a sum of a capacitance of a fixed capacitor and a tunable capacitance which is mediated across a pair of coupling capacitors. The qubits may be hybrid qubits, operable in a flux regime or a charge regime. Qubits may include a pair of CJJs that interrupt a loop of material and which are separated by an island of superconducting material which is voltage biased with respect to a qubit body.

Abstract: Systems and methods for optimizing and/or solving objective functions are provided. Angle-based stochastic gradient descent (AG-SGD) can be used to alleviate pattern deviation(s) not resolved by related art systems and methods. AG-SGD can use the angle between the current gradient (CG) and the previous gradient (PG) to determine the new gradient (NG).

Abstract: An interface device and method of use, comprising audio and image inputs; a processor for determining topics of interest, and receiving information of interest to the user from a remote resource; an audio-visual output for presenting an anthropomorphic object conveying the received information, having a selectively defined and adaptively alterable mood; an external communication device adapted to remotely communicate at least a voice conversation with a human user of the personal interface device. Also provided is a system and method adapted to receive logic for, synthesize, and engage in conversation dependent on received conversational logic and a personality.

Abstract: Systems and methods are described for developing and using neural network models. An example method of training a neural network includes: oscillating a learning rate while performing a preliminary training of a neural network; determining, based on the preliminary training, a number of training epochs to perform for a subsequent training session; and training the neural network using the determined number of training epochs. The systems and methods can be used to build neural network models that efficiently and accurately handle heterogeneous data.

Abstract: Techniques in advanced deep learning provide improvements in one or more of accuracy, performance, energy efficiency, and cost. In a first embodiment, a scaled array of processing elements is implementable with varying dimensions of the processing elements to enable varying price/performance systems. In a second embodiment, an array of clusters communicates via high-speed serial channels. The array and the channels are implemented on a Printed Circuit Board (PCB). Each cluster comprises respective processing and memory elements. Each cluster is implemented via a plurality of 3D-stacked dice, 2.5D-stacked dice, or both in a Ball Grid Array (BGA). A processing portion of the cluster is implemented via one or more Processing Element (PE) dice of the stacked dice. A memory portion of the cluster is implemented via one or more High Bandwidth Memory (HBM) dice of the stacked dice.