Abstract: An apparatus and method of training a convolutional neural network (CNN) are provided. A method of training a CNN including a plurality of convolution layers stored in a memory involves approximating, using a processor, a convolution layer among the plurality of convolution layers using a low-rank approximation; reducing the number of output reconstruction filters of the approximated convolution layer; and modifying a structure of the CNN based on an approximation result and the reduced number of output reconstruction filters.

Abstract: A method of detecting anomalies in a time series is disclosed. A training time series corresponding to a process is extracted from an initial time series corresponding to the process, the training time series including a subset of the initial time series. Outlier data points in the training time series are modified based on predetermined acceptability criteria. A plurality of prediction methods are trained using the training time series. An actual data point corresponding to the initial time series is received. The plurality of prediction methods are used to determine a set of predicted data points corresponding to the actual data point. It is determined whether the actual data point is anomalous based on a calculation of whether each of the set of predicted data points is statistically different from the actual data point.

Abstract: A learning apparatus in a digital environment is advantageous to interaction and communication among users who use a knowledge point structure for learning. The learning apparatus in the digital environment constructs a structurized knowledge library by editing knowledge points, tags of the knowledge points and a relationship among the knowledge points; then records a user's mastery degree for the knowledge points on the basis of the tags; and records knowledge learning information and social attribute information thereof for each user, so as to construct a general-class learning record, a single-class learning record and a general knowledge learning system of the user, which can also be used for the user to view the popularity of each tag in the knowledge library.

Abstract: Methods to, responsive to receiving a case by a deep question answering (deep QA) system, identify, in a corpus of information, a first variable for which a value was not specified in the case, compute an importance score for the first variable based on a concept in the corpus, wherein the concept is associated with the first variable, and upon determining that the importance score exceeds an importance threshold, determine that specifying a value for the first variable increases a confidence score of a response returned by the deep QA system beyond a confidence threshold.

Abstract: Approaches for optimizing network demand forecasting models and network topology using hyperparameter selection are provided. An approach includes defining a pool of features that are usable in models that predict demand of network resources, wherein the pool of features includes at least one historical forecasting feature and at least one event forecasting feature. The approach also includes generating, using a computer device, an optimal model using a subset of features selected from the pool of features. The approach further includes predicting future demand on a network using the optimal model. The approach additionally includes allocating resources in the network based on the predicted future demand.

Abstract: Systems and computer program products to, responsive to receiving a case by a deep question answering (deep QA) system, identify, in a corpus of information, a first variable for which a value was not specified in the case, compute an importance score for the first variable based on a concept in the corpus, wherein the concept is associated with the first variable, and upon determining that the importance score exceeds an importance threshold, determine that specifying a value for the first variable increases a confidence score of a response returned by the deep QA system beyond a confidence threshold.

Abstract: A system for monitoring an environment may include an input device for monitoring and capturing pattern-based states of a model of the environment. The system may also include a 5 thalamobot embodied in at least a first processor, in which the first processor is in communication with the input device. The thalamobot may include at least one filter for monitoring captured data from the input device and for identifying at least one state change within the captured data. The system may also include at least one critic and/or at least one recognition system.

Abstract: In one embodiment, a method includes accessing a first set of entities, with which a user has interacted, and a second set of entities in a social-networking system. A first set of vector representations of the first set of entities are determined using a deep-learning model. A target entity is selected from the first set of entities, and the vector representation of the target entity is removed from the first set. The remaining vector representations in the first set are combined to determine a vector representation of the user. A second set of vector representations of the second set of entities are determined using the deep-learning model. Similarity scores are computed between the user and each of the target entity and the entities in the second set of entities. Vector representations of entities in the second set of entities are updated based on the similarity scores using the deep-learning model.

Abstract: A method of training a neural network includes encouraging one or more filters in the neural network to have a low rank.

Abstract: Systems, methods, and apparatus, including computer programs encoded on a computer storage medium for processing a network input through a neural network having one or more initial neural network layers followed by a softmax output layer. In one aspect, the methods include obtaining a layer output generated by the one or more initial neural network layers and processing the layer output through the softmax output layer to generate a neural network output. Processing the layer output through the softmax output layer includes determining, for each possible output value, a number of occurrences in the layer output values; for each possible output value occurring in the layer output values, determining a respective exponentiation measure; determining a normalization factor for the layer output by combining the exponentiation measures in accordance with the number of occurrences of the possible output values; and determining, for each of layer output values, a softmax probability value.

Abstract: An analytics processing system generates analytics from a collection of unstructured data by identifying trends in the data and deriving associations or correlations between series of values. Each series is generated from a set of field labeled values in the set, and compared to other series in the collection. Identified relationships in the series are scored based on depiction of an illustrative, predictive, or non-random association, and ranked by a scoring metric for analytical value. A visualization of the relationships are ranked and rendered such that the visualization highlights the association in a manner not achievable by simple inspection of the field values. Relationships are graphed by lines, circles, bars (histogram) on labeled axes based on the series. In this manner, a user may generate analytic results from a large data set, and pinpoint significant associations by paging through renderings scored as the most illustrative of notable trends.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for receiving, by a computational graph system, a request to process a computational graph; obtaining data representing a subgraph of the computational graph, the computational graph comprising a plurality of nodes and directed edges, wherein each node represents a respective operation, wherein each directed edge connects a respective first node to a respective second node, the subgraph assigned to a first device by a placer in the computational graph system; determining that the first device comprises a hardware accelerator having a plurality of streams; in response to determining, generating instructions that when executed by the first device cause the first device to: assign the operation represented by each node in the subgraph to a respective stream; and perform the operations represented by the nodes in the subgraph in accordance with the assignment.

Abstract: A question answering system that determines whether a question is off-topic by performing the following steps: (i) receiving, by a question answering system, a set of documents; (ii) identifying topical subset(s) for each document of the set of documents using named entity recognition, where each topical subset relates to a corresponding topic; (iii) assigning a set of topic score(s) for each topical subset using natural language processing, where each topic score relates to a corresponding probability associated with the respective topical subset under a probabilistic language model; and (iv) determining, based, at least in part, on the topic score(s) corresponding to the topical subset(s), whether a question input into the question answering system is off-topic.

Abstract: Systems and methods may provide for partitioning a plurality of training samples into a first sequential list of centroids, removing one or more repeating centroids in the first sequential list of centroids to obtain a first reduced list of centroids and generating a set of Hidden Markov Model (HMM) parameters based on the first reduced list of centroids. Additionally, a plurality of detection samples may be partitioned into a second sequential list of centroids, wherein one or more repeating centroids in the second sequential list of centroids may be removed to obtain a second reduced list of centroids. The second reduced list of centroids may be used to determine a match probability for the plurality of detection samples against the set of HMM parameters. In one example, the reduced lists of centroids lack temporal variability.

Abstract: A method for determining the predictive value of a feature may include: (a) performing predictive modeling procedures associated with respective predictive models, wherein performing each modeling procedure includes fitting the associated model to an initial dataset representing an initial prediction problem; (b) determining a first accuracy score of each of the fitted models, representing an accuracy with which the fitted model predicts an outcome of the initial prediction problem; (c) shuffling values of a feature across observations included in the initial dataset, thereby generating a modified dataset representing a modified prediction problem; (d) determining a second accuracy score of each of the fitted models, representing an accuracy with which the fitted model predicts an outcome of the modified prediction problem; and (e) determining a model-specific predictive value of the feature for each of the fitted models based on the first and second accuracy scores of the fitted model.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for modifying a computational graph to include send and receive nodes. Communication between unique devices performing operations of different subgraphs of the computational graph can be handled efficiently by inserting send and receive nodes into each subgraph. When executed, the operations that these send and receive nodes represent may enable pairs of unique devices to conduct communication with each other in a self-sufficient manner. This shifts the burden of coordinating communication away from the backend, which affords the system that processes this computational graph representation the opportunity to perform one or more other processes while devices are executing subgraphs.

Abstract: An algorithm according to an embodiment of the present invention provides for latent signal detection of adverse events. Embodiments infer the presence of adverse drug events from large observational databases housed by the FDA, WHO, and other governmental organizations. The disclosed algorithms do not require the adverse event to be reported explicitly. Instead, the algorithms infer the presence of adverse events through more common secondary effects. In an embodiment, machine learning techniques are used for this purpose.

Abstract: According to an aspect, a heterogeneous graph in a data store is accessed. The heterogeneous graph includes a plurality of nodes having a plurality of node types. The nodes are connected by edges having a plurality of relation types. One or more intermediary graphs are created based on the heterogeneous graph. The intermediary graphs include intermediary nodes that are the relation types of the edges of the heterogeneous graph and include intermediary links between the intermediary nodes based on shared instances of the nodes between relation types in the heterogeneous graph. The intermediary graphs are traversed to find sets of relations based on intermediary links according to a template. An inference rule is extracted from the heterogeneous graph based on finding sets of relations in the intermediary graphs. The inference rule defines an inferred relation type between at least two of the nodes of the heterogeneous graph.

Abstract: Described is a system for learning, prediction, and recall of spatiotemporal patterns. An input spatiotemporal sequence is learned using a recurrent spiking neural network by first processing the input spatiotemporal sequence using the recurrent spiking neural network. The recurrent spiking neural network comprises neurons having excitatory synaptic connections and inhibitory synaptic connections. Balanced inhibitory connectivity exists between neurons having excitatory synaptic connections. The recurrent spiking neural network uses distinct forms of synaptic plasticity for excitatory synaptic connections and inhibitory synaptic connections, such that excitatory synaptic connections strengthen and inhibitory synaptic connections weaken. In another aspect, the system is able to recall the learned spatiotemporal sequence and predict a future spatiotemporal sequence through activation of the recurrent spiking neural network.

Abstract: A machine tool is provided with an operation evaluation section that outputs evaluation data on an operation of the machine tool and a machine learning device that performs machine learning of the movement amount of an axis. The machine learning device calculates a reward from physical-amount data on the machine tool and the evaluation data, performs adjustment of the movement amount of the axis based on a machine learning result of the adjustment of the movement amount of the axis and the physical-amount data, and performs the machine learning of the adjustment of the movement amount of the axis based on the adjusted movement amount of the axis, the physical-amount data after the operation of the machine tool based on the movement amount of the axis, and the reward.