Abstract: Systems and methods include initializing a trainees population (TP), calculating an objective function (OF) of the TP to identify a trainer. A teaching pool is created using variables of each trainee and the identified trainer, and unique variables are added to obtain an updated teaching pool (UTP). Search is performed on the UTP to obtain ‘m’ subset of variables and OFs. The OFs of ‘m’ subset are compared with OFs of the trainer's and each trainee's variable and one of the trainer or each trainee are updated accordingly. An updated learning pool (ULP) is created for selected trainee and the trainees, by adding unique variables to obtain ‘n’ subset. The OF of ‘n’ subset are compared with objective functions of selected trainee and the trainees and variables are updated accordingly. These steps are iteratively performed to obtain an optimal subset of variables that is selected for teaching and learning phase.

Abstract: Machine learning is utilized to analyze respective execution times of a plurality of tasks in a job performed in a distributed computing system to determine that a subset of the plurality of tasks are straggler tasks in the job, where the distributed computing system includes a plurality of computing devices. A supervised machine-learning algorithm is performed using a set of inputs including performance attributes of the plurality of tasks, where the supervised machine learning algorithm uses labels generated from determination of the set of straggler tasks, the performance attributes include respective attributes of the plurality of tasks observed during performance of the job, and applying the supervised learning algorithm results in identification of a set of rules defining conditions, based on the performance attributes of the plurality of tasks, indicative of which tasks will be straggler tasks in a job. Rule data is generated to describe the set of rules.

Abstract: According to some embodiments, a server may access a data store containing electronic records, each electronic record representing a risk association for an entity in connection with a plurality of relationships, and each electronic record may contain a set of record characteristic values. The server may automatically designate a first sub-set of the set of record characteristic values as fixed effect variables and a second sub-set as random effect variables. A data analytics mixed effect predictive model may then generate, based on the fixed and random effect variables, a future performance estimation value for the risk association of each entity in connection with its plurality of relationships. An indication associated with the future performance estimation value for the risk association of at least one entity in connection with its plurality of relationships may then be transmitted to generate an interactive user interface display.

Abstract: A mechanism is provided in a data processing system for question answering using time weighted evidence. The mechanism receives an input question. The mechanism determines a time focus for the input question and defines a weighting function. The weighting function is a bell curve having a peak at the time focus on a time axis. The mechanism decomposes the input question into one or more queries and applies the one or more queries to a corpus of information to obtain a set of hypothesis evidence. Each item of information within the hypothesis evidence has an associated time value. The mechanism weights the set of hypothesis evidence based on the associated time values according to the weighting function to form time weighted evidence and generates hypotheses for answering the input question based on the time weighted evidence.

Abstract: Techniques are described for formally expressing whether sequences of operations performed on block storage devices are sequential or random. In embodiments, determinations of whether these sequences of operations are sequential or random may be used to predict latencies involved with running particular workloads, and to predict representative workloads for particular latencies.

Abstract: A method and system to identify a time lagged indicator of an event to be predicted are described. The method includes receiving information including an indication of a factor, the factor being a different event than the event to be predicted, and identifying a window period within which the event is statistically correlated with the factor. The method also includes collecting data for a duration of the window period, the data indicating occurrences of the factor and the event, and identifying a time lagged dependency of the event on the factor based on analyzing the data.

Abstract: Each user is represented by a mixture of topics, e.g., one or more topics, and a probability of interest in each topic in the mixture, and given the target user, one or more other users can be recommended, each user that is recommended to the target user is determined to have a topical interest similarity with the target user, e.g., the target user's interest in one or more topics of the mixtures of topics is determined to be similar to a recommended interest in the one or more topics of the mixture of topics. The target user and the one or more recommended users can be said to have similar topical interests. The target user can use the user recommendation to establish an interactive dialogue, for example, with one or more users identified in the user recommendation.

Abstract: A neural network unit has at least one RAM, an output buffer and an array of neural processing units that: read first time step context layer node values from the output buffer; read second time step input layer node values from the RAM; generate second time step hidden layer node values based on the read input and context layer node values; output the hidden layer node values to the output buffer rather than to the RAM; read the hidden layer node values from the output buffer; generate second time step context layer node values based on the read hidden layer node values; output the context layer node values to the output buffer rather than to the RAM; generate output layer node values using the hidden layer node values; write the output layer node values to the RAM; and repeat for a sequence of time steps.

Abstract: A mechanism is provided for computing a solution to a plan recognition problem. The plan recognition problem includes the model and a partially ordered sequence of observations or traces. The plan recognition is transformed into an AI planning problem such that a planner can be used to compute a solution to it. The approach is general. It addresses unreliable observations: missing observations, noisy observations (or observations that need to be discarded), and ambiguous observations). The approach does not require plan libraries or a possible set of goals. A planner can find either one solution to the resulting planning problem or multiple ranked solutions, which maps to the most plausible solution to the original problem.

Abstract: A computer-implemented method includes receiving multimodal data. The computer-implemented method further includes generating one or more kernel matrices from the multimodal data. The computer-implemented method further includes generating an equivalent kernel matrix using one or more coefficient matrices, wherein the one or more coefficient matrices are constrained by a nuclear norm. The computer-implemented method further includes initiating one or more iterative processes. Each of the one or more iterative processes includes: calculating an error for the one or more coefficient matrices of the equivalent kernel matrix based on a training set, and initiating a line search for the one or more coefficient matrices of the equivalent kernel matrix. The computer-implemented method further includes, responsive to generating an optimal coefficient matrix, terminating the one or more iterative processes. The method may be embodied in a corresponding computer system or computer program product.

Abstract: In one embodiment, techniques are shown and described relating to traffic-based inference of influence domains in a network by using learning machines. In particular, in one embodiment, a management device computes a time-based traffic matrix indicating traffic between pairs of transmitter and receiver nodes in a computer network, and also determines a time-based quality parameter for a particular node in the computer network. By correlating the time-based traffic matrix and time-based quality parameter for the particular node, the device may then determine an influence of particular traffic of the traffic matrix on the particular node.

Abstract: A network of apparatuses that characterizes items is presented. A self-updating apparatus includes a processing unit that has a memory storing parameters that are useful for characterizing different items, and a processing module configured to automatically select sources from which to receive data, modify the parameters based on the data that is received, and to select recipients of modified parameters. Selection of sources and recipients is based on comparison of parameters between the processing module and the sources, and between the processing module and the recipients, respectively. The processing unit may include an artificial intelligence program (e.g., a neural network such as a machine learning program). When used in a network, the processing units may “train” other processing units in the network such that the characterization accuracy and range of each processing unit improves over time.

Abstract: A method for establishing semantic equivalence between a plurality of concepts including: providing an Orthogonal Semantic Equivalence Map in which first, second, and third extensional concept models are related; selecting or de-selecting a concept in the first concept model; selecting or deselecting a (relation, concept) pair representing an intensional relation from a concept in the first concept model to a concept in the second concept model over a concept in the third concept model; determining a subset of intensional relations from the selected concepts in the first concept model to concepts in the second concept model; determining a set of concepts from the first concept model that are related to concepts in the second concept model over the selected (relation, concept) pairs; and determining the narrowest common extension of the set of concepts from the first, second, or third concept models that are related over the selected intensional relations.

Abstract: Described is a system for online characterization of biomechanical and cognitive factors relevant to physical rehabilitation and training efforts. A biosensing subsystem senses biomechanical states of a user based on the output of sensors and generates a set of biomechanical data. The set of biomechanical data is transmitted in real-time to an analytics subsystem. The set of biomechanical data is analyzed by the analytics subsystem, and control guidance is sent through a real-time control interface to adjust the user's motions. In one aspect control guidance is sent to a robotic exoskeleton worn by the user to adjust the user's motions.

Abstract: A processing device and method of classifying data are provided. The method comprises the computer-implemented steps of selecting a M number of model sets, a R number of data representation sets, and a T number of sampling sets, generating a M*R*T number of classifiers comprising a three-dimensional (3D) array of classifiers, testing each individual classifier in the 3D array of classifiers on a testing set to obtain accuracy scores for the each individual classifier, and assigning a weight value to the each individual classifier corresponding to each accuracy score, wherein the 3D array of classifiers comprises a 3D array of weighted classifiers.

Abstract: A method, system and computer-usable medium for performing cognitive computing operations comprising receiving streams of data from a plurality of data sources; processing the streams of data from the plurality of data sources, the processing the streams of data from the plurality of data sources performing data enriching for incorporation into a cognitive graph; defining a travel-related cognitive persona within the cognitive graph, the travel-related cognitive persona corresponding to an archetype user model, the travel-related cognitive persona comprising a set of nodes in the cognitive graph; associating a user with the travel-related cognitive persona; defining a travel-related cognitive profile within the cognitive graph, the travel-related cognitive profile comprising an instance of the travel-related cognitive persona that references personal data associated with the user; associating the user with the travel-related cognitive profile; and, performing a cognitive computing operation based upon the tra

Abstract: A method for predicting subject trustworthiness includes using at least one classifier to predict truthfulness of subject responses to prompts during a local or remote interview, based on subject responses and response times, as well as interviewer impressions and response times, and, in embodiments, also biometric measurements of the interviewer. Data from the subject interview is normalized and analyzed relative to an experience database previously created using data obtained from test subjects. Classifier prediction algorithms incorporate assumptions that subject response times are indicators of truthfulness, that subjects will tend to be consistently truthful or deceitful, and that conscious and subconscious impressions of the interviewer are predictive of subject trustworthiness. Data regarding interviewer impressions can be derived from interviewer response times, interviewer questionnaire answers, and/or interviewer biometric data.

Abstract: Exemplary systems, apparatus, and methods for evaluating and predicting athletic performance are described. Systems may include a receiver that gathers non-deterministic data on one or more aspects of athletic performance, a deterministic model of the athletic performance, a hybrid processor that creates a conditional probabilistic model from these elements, and a display presenting the evaluated or predicted performance. The system may include sensors affixed to an athlete or their equipment to convey position, acceleration, heart rate, respiration, biomechanical attributes, and detached sensors to record video, audio, and other ambient conditions. Apparatus may include a hybridization processor that communicates the output of conditional probabilistic models directly to athletes, coaches, and trainers using sound, light, or haptic signals, or to spectators using audiovisual enhancements to broadcasts.

Abstract: The present disclosure is directed to the generation of a compact artificial neural network by removing individual nodes from the artificial neural network. Individual nodes of the artificial neural network may be deactivated randomly and/or selectively during training of the artificial neural network. In some embodiments, a particular node may be randomly deactivated approximately half of the time during processing of a set of training data inputs. Based on the accuracy of the results obtained when the node is deactivated compared to the accuracy of the results obtained when the node is activated, an activation probability may be generated. Nodes can then be selectively removed from the artificial neural network based on the activation probability.

Abstract: In implementations of the subject matter described herein, each token for containing an element in the training data is sampled according to a factorization strategy in training. Instead of using a single proposal, the property value of the target element located at the token being scanned is iteratively updated one or more times based on a combination of an element proposal and a context proposal. The element proposal tends to accept a value that is popular for the target element independently of the current piece of data, while the context proposal tends to accept whenever the property value that is popular in the context of the target data or popular for the element itself. The proposed modeling training approach can converge in a quite efficient way.