Abstract: A computer-implemented method, computer program product, and computer processing system are provided for accelerating neural network data parallel training in multiple graphics processing units (GPUs) using at least one central processing unit (CPU). The method includes forming a set of chunks. Each of the chunks includes a respective group of neural network layers other than a last layer. The method further includes performing one or more chunk-wise synchronization operations during a backward phase of the neural network data parallel training, by each of the multiple GPUs and the at least one CPU.

Abstract: Embodiments include predicting transactions by an entity and identifying promotions to offer the entity. Aspects include parsing a plurality of event records corresponding to a plurality of entities respectively. Aspects also include identifying a sequence of events corresponding to the entity and discretizing time intervals and event values of the sequence of events into discrete symbolic values. Aspects further include generating a temporal pattern of events in the sequence of events, the temporal pattern including a sequence of transaction-symbols representative of the time interval and the event value of the events in the sequence of events of the entity and predicting a next transaction based on the temporal pattern.

Abstract: The present disclosure relates to systems and methods for using transfer learning in log parsing neural networks. In one implementation, a system for training a neural network to parse unstructured data may include a processor and a non-transitory memory storing instructions that, when executed by the processor, cause the system to: receive unstructured data; apply a classifier to the unstructured data to determine that the unstructured data comprises a new category of unstructured data; in response to the determination, identify an existing category of unstructured data similar to the new category; based on the identified existing category, select a corresponding neural network; reset at least one weight and at least one activation function of the corresponding neural network while retaining structure of the corresponding neural network; train the reset neural network to parse the new category of unstructured data; and output the trained neural network.

Abstract: A method, computer program product, and apparatus for adapting a trained neural network having one or more batch normalization layers are provided. The method includes adapting only the one or more batch normalization layers using adaptation data. The method also includes adapting the whole of the neural network having the one or more adapted batch normalization layers, using the adaptation data.

Abstract: A system includes a learning object storing section that stores objects to be learned, a learning result storing section that stores learning results, and a control section connected to an input section. The control section computes a principal component coefficient vector of a first feature vector of an object to be processed that is designated by the input section, computes a principal component coefficient vector of a second feature vector using a principal component basis vector stored in the learning result storing section, and computes the second feature vector of the object to be processed using the principal component coefficient vector of the second feature vector.

Abstract: Quantum annealers as analog or quantum processors can find paths in problem graphs embedded in a hardware graph of the processor, for example finding valid paths, shortest paths or longest paths. A set of input, for example nucleic acid reads, can be used to set up a graph with edges between nodes denoting overlap (i.e., common base pairs) between the reads with constraints applied to perform sequence alignment or sequencing of a nucleic acid (e.g., DNA) strand or sequence, finding a solution that has a ground state energy. At least a portion of the described approaches can be applied to other problems, for instance resource allocations problems, e.g., job scheduling problems, traveling salesperson problems, and other NP-complete problems.

Abstract: A model integration method and device are provided. The method includes: obtaining an integrated model, the integrated model having one integrated output value and a plurality of input values, the plurality of input values corresponding to a plurality of output values of a plurality of independent models; performing one or more iterations of optimizing process until a preset iteration stop condition is satisfied: acquiring a prediction output by the integrated model based on a preset test event set; determining an index value of the integrated model based on the prediction output, the index value indicates a performance evaluation of the integrated model; if the index value fails to meet a preset performance requirement; after the preset iteration stop condition is satisfied, determining the integrated model as acceptable.

Abstract: A method for implementing spiking neural network computations, the method including defining a dynamic node response function that exhibits spikes, where spikes are temporal nonlinearities for representing state over time; defining a static representation of said node response function; and using the static representation of the node response function to train a neural network. A system for implementing the method is also disclosed.

Abstract: Techniques are described for determining the value of individual facts in a knowledge base, and various applications of such fact values. In one example, the knowledge base is part of a question answering system. A ranking of knowledge base facts based on the number of times each of the knowledge base facts is used in answering user questions (e.g., as determined from question answering logs) is used to derive a fact value function that may then be used to determine the value of other facts included in or subsequently added to the knowledge base.

Abstract: A method for ranking detected anomalies is disclosed. The method includes generating a graph based on a plurality of rules, wherein the graph comprises nodes representing metrics identified in the rules, edges connecting nodes where metrics associated with connected nodes are identified in a given rule, and edge weights of the edges each representing a severity level assigned to the given rule. The method further includes ranking nodes of the graph based on the edge weights. The method further includes ranking detected anomalies based on the ranking of the nodes corresponding to the metrics associated with the detected anomalies.

Abstract: The subject-matter described herein relates to a computer-implemented method of enabling interoperability between a first knowledge base and a second knowledge base. Each knowledge base is graphically represented and includes a plurality of nodes each defining a concept and a plurality of relations linking the plurality of nodes. The first knowledge base and the second knowledge base are encoded using different coding standards. The method comprises: identifying an entity from the plurality of entities in the second knowledge base; obtaining a mapping between the identified entity from the second knowledge base and a matching entity from the first knowledge base; and creating and storing a link between the identified entity from the second knowledge base and the matching entity from the first knowledge base.

Abstract: A system, method and program product for a computer-based project collaboration system using a data log for cross-domain collaboration. A cognitive log stores log entries based on domain-specific project data sources. An ontology translator includes domain-specific ontologies and a mapping ontology that defines relationships among the domain-specific ontologies. A cross-domain query includes domain parameters from one domain-specific ontology and returns and displays results based on log entries with domain parameters from another domain-specific ontology using the ontology translator.

Abstract: A hierarchy of agents is constructed from a set of agents. Each agent in the hierarchy is trained to answer a question according to a corresponding corpus associated with the agent, which contains a portion of knowledge about a subject-matter. The question is submitted to a first subset of agents, the agents in the first subset occupying a first level in the hierarchy. From a first agent in the first subset, a first answer is propagated to a second agent in a second subset of agents, the first agent computing the first answer using a first portion of knowledge about the subject-matter. to form a first morphed answer, a second answer is added to the first answer, the second answer being computed by the second agent using a second portion of knowledge about the subject-matter. The morphed answer is produced in response to the question.

Abstract: Systems and methods for building a distributed learning framework, including generating a sparse communication network graph with a high overall spectral gap. The generating includes computing model parameters in distributed shared memory of a cluster of a plurality of worker nodes; determining a spectral gap of an adjacency matrix for the cluster using a stochastic reduce convergence analysis, wherein a spectral reduce is performed using a sparse reduce graph with a highest possible spectral gap value for a given network bandwidth capability; and optimizing the communication graph by iteratively performing the computing and determining until a threshold condition is reached. Each of the plurality of worker nodes is controlled using tunable approximation based on available bandwidth in a network in accordance with the generated sparse communication network graph.

Abstract: The present teaching relates to distributed deep machine learning on a cluster. In one example, a request is received for estimating one or more parameters associated with a machine learning model on a cluster including a plurality of nodes. A set of data is obtained to be used for estimating the one or more parameters. The set of data is divided into a plurality of sub-sets of data, each of which corresponds to one of the plurality of nodes. Each sub-set of data is allocated to a corresponding node for estimating values of the one or more parameters based on the sub-set of data. Estimated values of the one or more parameters obtained based on a corresponding sub-set of data allocated to the node, are received from each of the plurality of nodes. The one or more parameters of the machine learning model are estimated based on the estimated values of the one or more parameters generated by at least some of the plurality of nodes.

Abstract: A feature extraction is performed on transaction data to obtain a user classification feature and a transaction classification feature. A first dimension feature is constructed based on the user classification feature and the transaction classification feature. A dimension reduction processing is performed on the first dimension feature to obtain a second dimension feature. A probability that the transaction data relates to a risky transaction is determined based on a decision classification of the second dimension feature, where the decision classification is based on a pre-trained deep forest network including a plurality of levels of decision tree forest sets.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a target sequence including a respective output at each of multiple output time steps from respective encoded representations of inputs in an input sequence. The method includes, for each output time step, starting from the position, in the input order, of the encoded representation that was selected as a preceding context vector at a preceding output time step, traversing the encoded representations until an encoded representation is selected as a current context vector at the output time step. A decoder neural network processes the current context vector and a preceding output at the preceding output time step to generate a respective output score for each possible output and to update the hidden state of the decoder recurrent neural network. An output is selected for the output time step using the output scores.

Abstract: The technology disclosed provides a quasi-recurrent neural network (QRNN) encoder-decoder model that alternates convolutional layers, which apply in parallel across timesteps, and minimalist recurrent pooling layers that apply in parallel across feature dimensions.

Abstract: Described are methods and systems to identify analyzing a social network to predict member actions, queries, or ranks within a social networking system. According to various embodiments, the system detects changes within a first data set of a first member. The system identifies an entity associated with the change in the first data set, determines an action probability of the entity in response to the change, and identifies a second data set associated with a second member having at least one common element with the first data set. The system identifies a set of elements in the first data set and an entity data set corresponding to the change and generates a customized user interface screen comprising a representation of the entity and a portion of the set of elements.

Abstract: Methods and systems for pruning process execution logs include learning a predictive model from a set of execution traces that characterize a process, where the predictive model determines a likelihood of a given instance reaching a specified outcome; identifying attributes in the predictive model that fall below a threshold measure of relevance to the specified outcome using a processor; and removing the identified attributes from the set of execution traces.