Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network. One of the methods includes generating, by a speech recognition system, a matrix from a predetermined quantity of vectors that each represent input for a layer of a neural network, generating a plurality of sub-matrices from the matrix, using, for each of the sub-matrices, the respective sub-matrix as input to a node in the layer of the neural network to determine whether an utterance encoded in an audio signal comprises a keyword for which the neural network is trained.

Abstract: Embodiments of an adaptive virtual intelligent agent (“AVIA”) service are disclosed. It may include the functions of a human administrative assistant for an enterprise including customer support, customer relationship management, and fielding incoming caller inquiries. It also has multi-modal applications for the home through interaction with AVIA implemented in the home. It may engage in free-form natural language dialogs. During a dialog, embodiments maintain the context and meaning of the ongoing dialog and provides information and services as needed by the domain of the application. Over time, the service automatically extends its knowledge of the domain (as represented in the Knowledge Tree Graphs) through interaction with external resources.

Abstract: A method of dynamically modifying target selection with a neural network includes dynamically modifying a selection function by controlling an amount of imbalance of connections in the neural network. A selected neuron represents one of multiple candidate targets.

Abstract: A technology is described for managing computing capacity within a computing service environment. An example method may include determining a state of a computing service environment having a group of physical hosts that execute computing instances and determining a predicted demand for computing instance launches on the group of physical hosts. A co-location of the computing instance launches on the group of physical hosts may then be determined by applying the predicted demand for computing instance launches to the state of the computing service environment. A predicted launch time for the computing instance launches may then be determined using the co-location of the computing instance launches as an input feature. An additional physical host may then be added to the group of physical hosts to decrease the predicted launch time.

Abstract: A personal taste assessment system recommends and predicts a person's preference for a consumable or other item. The system accesses a user profile for a person. The user profile includes a preference model representing associations between the person's ratings of items and a set of item characteristics. The system also accesses a database of characteristic values for a group of items, uses the identifying information to identify a candidate item having characteristic values whose properties match characteristics associated with the rated items that the person found to be appealing, and processes the characteristic values of the candidate item with the user profile to generate a predicted rating as a prediction of how the person would rate the identified candidate item. The system then causes an electronic device to output an identification of the candidate item and the predicted rating.

Abstract: A prediction technique to predict an anomalous state of a processing environment at a future point in time. One or more values of one or more metrics of the processing system are obtained. For the one or more metrics, one or more predicted values are determined for one or more points in time in the future. Based the predicted values, one or more change values for one or more points in time are determined, and based on the one or more change values, a determination is made as to whether an anomalous state exists within the processing system.

Abstract: An information processing device includes a keyword acquiring unit configured to acquire a plurality of primary keyword and secondary keyword groups; a classifying unit configured to classify each of the plurality of secondary keywords by a plurality of topics; an estimating unit configured to estimate whether or not each primary keyword in the plurality of groups is a related keyword related to any topic having a classified secondary keyword or a mixed keyword unrelated to any of the topics; and an assigning unit configured to preferentially assign a primary keyword estimated to be a related keyword to a topic having a classified secondary keyword in the same group, and assigning a primary keyword estimated to be a mixed keyword to any of all the topics given for classification.

Abstract: An information processing device includes a keyword acquiring unit configured to acquire a plurality of primary keyword and secondary keyword groups; a classifying unit configured to classify each of the plurality of secondary keywords by a plurality of topics; an estimating unit configured to estimate whether or not each primary keyword in the plurality of groups is a related keyword related to any topic having a classified secondary keyword or a mixed keyword unrelated to any of the topics; and an assigning unit configured to preferentially assign a primary keyword estimated to be a related keyword to a topic having a classified secondary keyword in the same group, and assigning a primary keyword estimated to be a mixed keyword to any of all the topics given for classification.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for using a quantum oracle to make inference in complex machine learning models that is capable of solving artificial intelligent problems. Input to the quantum oracle is derived from the training data and the model parameters, which maps at least part of the interactions of interconnected units of the model to the interactions of qubits in the quantum oracle. The output of the quantum oracle is used to determine values used to compute loss function values or loss function gradient values or both during a training process.

Abstract: A computing resource service provider may store user data in a distributed data storage system. The distributed data storage system may contain one or more storage partitions configured to store based at least in part on prefixes of keys contained in a key-value store, where the size of the keys may vary. The one or more storage partitions may track requests to access data where the requests include a key such that the data may be located by the service provider based at least in part on the key. When a request is received a counter associated with the prefix included in the request may be incremented, the counter may be configured to decay over time. If the counter exceeds a threshold the service provider may split the prefix associated with the counter and generate new partitions responsible for the split prefix.

Abstract: A first input is processed via a first configuration of a neural network to produce a first output. The first configuration defines attributes of the neural network, such as connections between neural elements of the neural network. If the neural network requires a context switch to process a second input, a second configuration is applied to the neural network to change the attributes, and the second input is processed via the second configuration of the neural network to produce a second output.

Abstract: A computer-implemented method includes receiving a batch of neural network inputs to be processed using a neural network on a hardware circuit. The neural network has multiple layers arranged in a directed graph and each layer has a respective set of parameters. The method includes determining a partitioning of the neural network layers into a sequence of superlayers. Each superlayer is a partition of the directed graph that includes one or more layers. The method includes processing the batch of inputs using the hardware circuit, which includes, for each superlayer in the sequence: i) loading the respective set of parameters for the layers in the superlayer into memory of the hardware circuit, and ii) for each input in the batch, processing the input through each of the layers in the superlayer using the parameters in the memory of the hardware circuit to generate a superlayer output for the input.

Abstract: Optimization of machine intelligence utilizes a systemic process through a plurality of computer architecture manipulation techniques that take unique advantage of efficiencies therein to minimize clock cycles and memory usage. The present invention is an application of machine intelligence which overcomes speed and memory issues in learning ensembles of decision trees in a single-machine environment. Such an application of machine intelligence includes inlining relevant statements by integrating function code into a caller's code, ensuring a contiguous buffering arrangement for necessary information to be compiled, and defining and enforcing type constraints on programming interfaces that access and manipulate machine learning data sets.

Abstract: A computer-based crime risk forecasting system and corresponding method are provided for generating crime risk forecasts and conveying the forecasts to a user. With the conveyed forecasts, the user can more effectively gauge both the level of increased crime threat and its potential duration. The user can then leverage the information conveyed by the forecasts to take a more proactive approach to law enforcement in the affected areas during the period of increased crime threat.

Abstract: Briefly, embodiments disclosed herein may relate to formulating synthetic questions, such as in response to a search query, for example. Candidate synthetic questions may be presented to a user who may initiate a search at least in part by selecting one or more candidate synthetic questions, for example, in an embodiment.

Abstract: A method, system, and computer program product for social collaboration in probabilistic prediction are provided in the illustrative embodiments. A set of predictions is sent to a user device. A prediction in the set of predictions is a probability of an outcome of an event. The probability is computed using a prediction model trained with training data corresponding to the event. An input is received from the user device. The input comprises a new prediction made at the user device using a new prediction model executing on the user device. A difference is determined between the prediction and the new prediction. The prediction model is revised to produce a revised prediction. A revised difference between the revised prediction and the new prediction is smaller than the difference.

Abstract: A system and method for matching an advertisement item to a multimedia content element based on sentiments. The method comprises: extracting at least one multimedia content element from a web-page requested for display on a user node; generating a signature for each of the at least one multimedia content element, wherein each signature represents a concept, wherein each concept is an abstract description of one of the at least one multimedia content element; correlating the concepts of the generated signatures to determine a context of the at least one multimedia content element, wherein the context indicates at least a brand sentiment; searching for at least one advertisement item based on the signatures and the context; and causing a display of the at least one advertisement item within a display area of the web-page.

Abstract: A method detects anomalies in time series data, by first learning a final set of exemplars by summarizing training time series data using a divide-and-conquer procedure. Then, for each window of testing time series data, a distance to a nearest exemplar in the final set of exemplars is determined, wherein the distance is an anomaly score. Finally, an anomaly is signaled when the anomaly score for a window is greater than a threshold.

Abstract: Methods, systems, and apparatus for receiving a particular set of user data; obtaining a predictive model that estimates a likelihood of a user to order a food item, wherein the predictive model is generated using observation data that includes historic user data and user data from other user devices; providing the particular set of user data to the predictive model; obtaining an indication of a likelihood of the user to order a food item; based on the indication of a likelihood of the user to order a food item, determining whether to output a notification on the user device inviting the user to order a food item; and in response to determining to output a notification on the user device inviting the user to order a food item, selectively outputting a notification on the user device inviting the user to order a food item.

Abstract: A system for sensing data includes one or more sensors formed on a substrate, including flexible substrates. A plurality of transistors are coupled to the one or more sensors and formed on the substrate. Each transistor of the plurality of transistors is constructed with a channel formed of a nanoscale material. The plurality of transistors are configured to perform computing tasks such that data processing and classification are performed directly on the sensor substrate. The nanoscale material can include carbon nanotubes.