Abstract: System and means generates synthetic forms of social media data such as data from microblogging services (e.g., Twitter) and social networking services (e.g., Facebook). This system and means jointly generate interaction graph structures and text features similar to input social media data. First, an interaction graph is generated by mapping social network interactions in input (real) social media data to graph structures. This interaction graph is fitted to a social network model (or a composite model) by minimizing the distance between the input and the synthetic interaction graphs (of potentially different sizes). The distance is measured statistically or based on the performance of social media analytics. Various patterns (such as anomalies), interaction types and temporal dynamics are generated synthetically. Second, text features are extracted from input social media data with topic modeling and statistical analysis of word tuple distributions.

Abstract: In one aspect, the inventions include a system for control of an entertainment state system. First, an application plane layer is adapted to receive instructions regarding operation of the entertainment state system. Preferably, the application plane layer is coupled to an application plane layer interface. Second, a control plane layer includes an adaptive control unit, such as a cognitive computing unit, an artificial intelligence unit or a machine-learning unit. Third, a data plane layer includes an input interface to receive data input from one or more data sources. An adaptive control unit is trained at least in part on analyzing the behavioral responses of users in response to content provided on one or more displays to the users as sensed by one or more sensors.

Abstract: A database query comprising predicates may be received. Each predicate may operate on database columns. The database query may be determined to comprise strict operators. An upper bound neural network may be defined for calculating an adjacent upper bound and a lower bound neural network may be defined for calculating an adjacent lower bound. The upper bound neural network and the lower bound neural network may be trained using a selected value from data of a database table associated with the database query to be executed through the upper bound neural network and the lower bound neural network. The upper bound neural network and the lower bound neural network may be adjusted by passing in an expected value using an error found in expressions. The adjacent lower bound and the adjacent upper bound may be calculated in response to completion of initial training for the database columns.

Abstract: A social networking system infers a sentiment polarity of a user toward content of a page. The sentiment polarity of the user is inferred based on received information about an interaction between the user and the page (e.g., like, report, etc.), and may be based on analysis of a topic extracted from text on the page. The system infers a positive or negative sentiment polarity of the user toward the content of the page, and that sentiment polarity then may be associated with any second or subsequent interaction from the user related to the page content. The system may identify a set of trusted users with strong sentiment polarities toward the content of a page or topic, and may use the trusted user data as training data for a machine learning model, which can be used to more accurately infer sentiment polarity of users as new data is received.

Abstract: Techniques are disclosed herein for making predictions with respect to how content consumers will interact with a digital asset. For example, in the context of website visitors browsing digital assets provided via a website, web traffic data can be collected and modeled using a belief network. The belief network may represent a probability distribution for a set of variables that define the web traffic data. Examples of such variables include browser type, browsing session duration, geographic location, visitor demographic characteristics, and a browsing outcome. Certain of the embodiments disclosed herein can be used to extract knowledge from the belief network, thereby allowing statistical inferences to be drawn with respect to how certain classes of website visitors will interact with the website. The influence of one or more first variables (for example, geographic location) can be quantified with respect to one or more second variables (for example, the successful result indicator).

Abstract: Systems and methods for predictive document coding using continuous active machine learning are described herein. A method uses both a primary queue and a plurality of secondary queues, where each secondary queue is associated with a model for category of documents. The method also repeatedly classifies new batches selected from a large set of documents that have not been reviewed. The classification uses the plurality of models and updates the secondary queues from the best documents in the most recently classified batch. While the method transparently cycles through batches, the most relevant documents are provided to one or more human reviewers from secondary queues via a primary queue. The reviewer confirms relevance or non-relevance in each of the documents for each of the categories. Periodically all the models are retrained using the set of reviewed documents after a selectable number of documents have been reviewed since the most recent retraining.

Abstract: The subset encoding method and related automata designs for improving the space efficiency for many applications on the Automata Processor (AP) are presented. The method is a general method that can take advantage of the character-or ability of STEs (State Transition Elements) on the AP, and can relieve the problems of limited hardware capacity and inefficient routing. Experimental results show that after applying the subset encoding method on Hamming distance automata, up to 3.2× more patterns can be placed on the AP if a sliding window is required. If a sliding window is not required, up to 192× more patterns can be placed on the AP. For a Levenshtein distance, the subset encoding can split the Levenshtein automata into small chunks and make them routable on the AP. The impact of the subset encoding method depends on the character size of the AP.

Abstract: An analog circuit includes a first stage comprising a first stage transistor, wherein the first stage transistor is configured to receive an input signal; and a second stage coupled to the first stage, the second stage comprising a pull-up transistor and a pull-down transistor, the pull-up transistor and the pull-down transistor having the same channel type as the first stage transistor. One of the pull-up transistor and the pull-down transistor receives the input signal and the other of the pull-up transistor and the pull-down transistor receives an output of the first stage transistor. The first stage and the second stage are configured to generate a sigmoid function.

Abstract: There is provided an information processing apparatus and an information processing method to present information for improving development efficiency of a neural network to a user. The information processing method includes: comparing, by a processor, learning results by a plurality of models of neural networks and presenting comparison information relating to the learning results using a graph. In addition, an information processing apparatus includes a control unit configured to acquire learning results by a plurality of models of neural networks and control comparison information relating to the learning results. The control unit presents the comparison information using a graph. In addition, an information processing apparatus includes a comparing unit configured to compare learning results by a plurality of models of neural networks and generate comparison information relating to the learning results. The comparison information is comparison information using a graph.

Abstract: Contextual graph matching based anomaly detection may include evaluating computer-generated log file data to create a master directed graph that specifies known events and transitions between the known events. The master directed graph may be processed to determine a plurality of decomposed master graph walks. Incoming computer-generated log file data may be evaluated to create an incoming directed graph that specifies unknown events and transitions between the unknown events. The incoming directed graph may be processed to determine a decomposed incoming walk. Overlap, distance difference, and correlation scores may be determined for each walk pair of a plurality of walk pairs including each of the plurality of decomposed master graph walks and the decomposed incoming walk. One of the decomposed master graph walks may be selected based on the overlap score, the difference score, and the correlation score, to detect an anomaly.

Abstract: Disclosed is a system, method, and computer program product that employs high dynamic range (HDR) image processing and manipulation algorithms for capturing and measuring real-time sky conditions for processing into control input signals to a building's automated fenestration (AF) system, daylight harvesting (DH) system and HVAC system. The photometer comprises a color camera and a fitted fish-eye lens to capture 360-degree, hemispherical, low dynamic range (LDR) color images of the sky. Both camera and lens are housed in a sealed enclosure protecting them from environmental elements and conditions. In some embodiments the camera and processes are controlled and implemented by a back-end computer.

Abstract: Described is a multiple-camera system and process for disambiguating between multiple users and identifying which of the multiple users performed an event. For example, when an event is detected, user patterns near the location of the event are determined, along with touch points at the location of the event. User pattern orientation and/or arm trajectories between the event location and the user patterns may be determined and processed to disambiguate between multiple users and determine which user pattern is involved in the event.

Abstract: The disclosure relates to technology for identifying a root cause of anomalous behavior in a communications network. A key quality indicator (KQI) indicative of a performance level associated with a source is received in the communication network. The KQI includes a performance measurement value to identify a performance level of the source having the anomalous behavior. An improved KQI indicative of a level of degradation is calculated based on the anomalous behavior at the source by recovering the KQI to a historical value. Sources of the anomalous behavior are clustered into subsets according to the level of degradation based on the calculated KQI improvement. A global or partial degradation type of the root cause source in the subset having a severe level of degradation is determined using hypothesis testing, and a confidence value is provided for the accepted degradation type.

Abstract: Systems and methods for classifying a test object are provided. For each respective target object in a plurality of target objects, a first procedure is performed comprising (a) posing the test object against the respective target thereby obtaining an interaction between the test and target, and (b) scoring the interaction with a first classifier. Each such score across the plurality of targets forms a test vector that is inputted into a second classifier thereby obtaining an indication of a target object. The second classifier is trained on training vectors, each being the output from instances of the first classifier after inputting a corresponding training object in a plurality of training objects in accordance with the first procedure. Each object in one subset of the training objects is uniquely associated with one of the targets. Another subset of the training objects is not associated with the targets.

Abstract: System, method, and computer program product to determine that a question received by a deep question answering system is speculative, generate, by one or more predictive algorithms, a set of candidate answers, compute a score for each candidate answer in the set of candidate answers, and return a first candidate answer, of the set of candidate answers, as responsive to the speculative question received by the deep question answering system.

Abstract: A first set of execution platforms is deployed for a set of operations of a training phase of a machine learning model. Prior to the completion of the training phase, a triggering condition for deployment of a different set of execution platforms is detected. The different set of execution platforms is deployed for a subsequent set of training phase operations.

Abstract: To precisely predict future data even when the number of pieces of time-series data is small, in predicting the future data, using the time-series data. When the future data is predicted using the time-series data, whether present time data is used is determined based on prediction variation or a data transition, and then the prediction of the future data is performed.

Abstract: Methods and systems for learning, recognition, classification and analysis of real-world cyclic patterns using a model having n oscillators, with primary frequency ?1, ?2, . . . , ?n. The state of the oscillators is evolved over time using sensor observations, which are also used to determine the sensor characteristics, or the sensor observation functions. Once trained, a set of activity detection filters may be used to classify a sensor data stream as being associated with an activity.

Abstract: An apparatus for classifying data using a neural network includes an input layer configured to receive input data; an output value to output a plurality of first output values of the input data with respect to each of at least one of all classes, and output only one first output value of the input data with respect to each of a rest of all of the classes; and a boost pooling layer to receive one or more output values output for each class, and output one second output value for each class.

Abstract: Systems and methods are provided for a computer-implemented method for identifying pairs of cohesive words within a text. A supervised model is trained to detect cohesive words within a text to be scored. Training the supervised model includes identifying a plurality of pairs of candidate cohesive words in a training essay and an order associated with the pairs of candidate cohesive words based on an order of words in the training essay. The pairs of candidate cohesive words are filtered to form a set of evaluation pairs. The evaluation pairs are provided via a graphical user interface based on the order associated with the pairs of candidate cohesive words. An indication of cohesion or no cohesion is received for the evaluation pairs via the graphical user interface. The supervised model is trained based on the evaluation pairs and the received indications.