Abstract: Hardware neural network processors, are provided. A neural core includes a weight memory, an activation memory, a vector-matrix multiplier, and a vector processor. The vector-matrix multiplier is adapted to receive a weight matrix from the weight memory, receive an activation vector from the activation memory, and compute a vector-matrix multiplication of the weight matrix and the activation vector. The vector processor is adapted to receive one or more input vector from one or more vector source and perform one or more vector functions on the one or more input vector to yield an output vector. In some embodiments a programmable controller is adapted to configure and operate the neural core.

Abstract: Example systems and methods of developing a learning model are presented. In one example, a sample data set to train a first learning algorithm is accessed. A number of states for each input of the sample data set is determined. A subset of the inputs is selected, and the sample data set is partitioned into a number of partitions equal to a combined number of states of the selected inputs. A second learning algorithm is created for each of the partitions, wherein each second learning algorithm receives the unselected inputs. Each of the second learning algorithms is assigned to a processor and trained using the samples of the partition corresponding to that algorithm. Decision logic is generated to direct each of a plurality of operational data units as input to one of the second learning algorithms based on states of the selected inputs of the operational data unit.

Abstract: Examples include techniques to manage training or trained models for deep learning applications. Examples include routing commands to configure a training model to be implemented by a training module or configure a trained model to be implemented by an inference module. The commands routed via out-of-band (OOB) link while training data for the training models or input data for the trained models are routed via inband links.

Abstract: Language-based machine learning approach for automatically detecting universal charset and the language of a received document is disclosed. The language-based machine learning approach employs a plurality of text document samples in different languages, after converting them to a selected Unicode style (if their original encoding schemes are not the selected Unicode), to generate a plurality of language-based machine learning models during the training stage. During the application stage, vector representations of the received document for different combinations of charsets and their respective applicable languages are tested against the plurality of machine learning models to ascertain the charset and language combination that is most similar to its associated machine learning model, thereby identifying the charset and language of the received document.

Abstract: An approach is provided for detecting and analyzing an anomaly in application performance in a client-server connection via a network. A status code of a response sent by a server to a client, a round trip latency time (RTT) of the response, and a time out of a connection between client and server are determined. Using a k-means clustering algorithm, buckets of RTT values clustered into lower and higher values, and running counts and means for the RTT values in each bucket, an RTT value is determined to exceed a threshold value. Based on the status code, the RTT value exceeding the threshold, and the connection time out, the anomaly is detected. Based on temporal and textual analyses of log entries and an environment analysis, candidate root causes of a failure that resulted in the anomaly are determined.

Abstract: Approaches presented herein enable intelligent service request classification and assignment learning. More specifically, a request comprising a free form text or spoken description is received from a user. The request description is parsed and classified by a regression-based classifier. The regression-based classifier classifies based on, for example: the description itself; the requestor's history of requests, and/or supplemental demographics about a requestor. Optionally, a user may verify the classification or select from a plurality of returned classifications. A service provider or administrator confirms that a classification is correct. If not, the incorrectly classified request is queued. If so, the correctly classified request is added to a set of training data to be used in classifying future requests.

Abstract: Embodiments of the invention relate to a neural network system for simulating neurons of a neural model. One embodiment comprises a memory device that maintains neuronal states for multiple neurons, a lookup table that maintains state transition information for multiple neuronal states, and a controller unit that manages the memory device. The controller unit updates a neuronal state for each neuron based on incoming spike events targeting said neuron and state transition information corresponding to said neuronal state.

Abstract: Embodiments are directed to data verification of business or consumer data. Certain embodiments include a data verification system that receives or selects data to be verified, selects one or more verification methods to verify, update, and/or append/enhance the data. The data verification system may verify the data with one or more data verification methods, either alone or in combination. The methods may include a web-crawling verification method, an agent web verification method, a call verification method, a direct mail method, an email method, an in-person verification method, or other methods. The system has the ability to, automatically or manually, (1) blend automatic and manual segmentation of records or elements by criteria such as industry type, best times of day/month/year to verify, update, or append, cost, and level of importance (2) select the best verification processing method(s), and (3) manage the results and properly verify, update, append/enhance records.

Abstract: A model identification system includes a device information acquiring unit that acquires device information used to identify a model of an electric device, an operation extracting unit that extracts data of a predetermined operation section, a feature quantity extracting unit that extracts a parameter used to identify the electric device, and a model identifying unit that identifies a model of an electric device, wherein the feature quantity extracting unit performs a machine learning process by sampling the data of the predetermined operation section extracted from the operation extracting unit a plurality of times, extracts a parameter corresponding to each sampling, and extracts a parameter appropriate to identify a model among a plurality of sampled parameters.

Abstract: A method for rejecting biased data using a machine learning model includes receiving a cluster training data set including a known unbiased population of data and training a clustering model to segment the received cluster training data set into clusters based on data characteristics of the known unbiased population of data. Each cluster of the cluster training data set includes a cluster weight. The method also includes receiving a training data set for a machine learning model and generating training data set weights corresponding to the training data set for the machine learning model based on the clustering model. The method also includes adjusting each training data set weight of the training data set weights to match a respective cluster weight and providing the adjusted training data set to the machine learning model as an unbiased training data set.

Abstract: The invention relates to a system and to a method for determining a state of a device by means of a trained support-vector machine. According to the invention, an operating parameter space is divided into classification volumes, at least one of which indicates a normal state and at least one other of which indicates a fault state of the device. A current state of the device can therefore be determined by determining where a current operating parameter point is to be arranged in the operating parameter space. The invention further relates to methods and to variants of the system in order to facilitate a cause evaluation and to determine particularly relevant operating parameters for the fault determination.

Abstract: A computing system trains a classification model using distributed training data. In response to receipt of a first request, a training data subset is accessed and sent to each higher index worker computing device, the training data subset sent by each lower index worker computing device is received, and a first kernel matrix block and a second kernel matrix block are computed using a kernel function and the accessed or received training data subsets. (A) In response to receipt of a second request from the controller device, a first vector is computed using the first and second kernel matrix blocks, a latent function vector and an objective function value are computed, and the objective function value is sent to the controller device. (A) is repeated until the controller device determines training of a classification model is complete. Model parameters for the trained classification model are output.

Abstract: Embodiments for planning vehicular driving actions in the presence of non-recurrent events by a processor. One or more dynamics of non-recurrent events in a transport network may be learned according to one or more contextual factors. One or more vehicle-specific factors may be learned in relation to a historical journey and current journey of a vehicle. One or more action responses associated with the one or more non-recurrent events and the one or more vehicle-specific factors may be generated.

Abstract: Recommendations for new experiments are generated via a pipeline that includes a predictive model and a preference procedure. In one example, a definition of a development task includes experiment parameters that may be varied, the outcomes of interest and the desired goals or specifications. Existing experimental data is used by machine learning algorithms to train a predictive model. The software system generates candidate experiments and uses the trained predictive model to predict the outcomes of the candidate experiments based on their parameters. A merit function (referred to as a preference function) is calculated for the candidate experiments. The preference function is a function of the experiment parameters and/or the predicted outcomes. It may also be a function of features that are derived from these quantities. The candidate experiments are ranked based on the preference function.

Abstract: The Quanton virtual machine approximates solutions to NP-Hard problems in factorial spaces in polynomial time. The data representation and methods emulate quantum computing on classical hardware but also implement quantum computing if run on quantum hardware. The Quanton uses permutations indexed by Lehmer codes and permutation-operators to represent quantum gates and operations. A generating function embeds the indexes into a geometric object for efficient compressed representation. A nonlinear directional probability distribution is embedded to the manifold and at the tangent space to each index point is also a linear probability distribution. Simple vector operations on the distributions correspond to quantum gate operations. The Quanton provides features of quantum computing: superpositioning, quantization and entanglement surrogates. Populations of Quantons are evolved as local evolving gate operations solving problems or as solution candidates in an Estimation of Distribution algorithm.

Abstract: While artificial neural networks can be used to predict particular values in certain contexts, convolutional neural networks are not typically used in these contexts—instead they may be employed for image recognition. However, raw transactional data may be structured to take advantage of convolutional neural network (CNN) techniques by arranging the data such that correlations are increased between nearby other data. In arranging data in this manner, the structured CNN (SCNN) can operate efficiently without having to make use of engineered data features, the generation and maintenance of which can be a time-consuming process.

Abstract: Examples include techniques to manage training or trained models for deep learning applications. Examples include routing commands to configure a training model to be implemented by a training module or configure a trained model to be implemented by an inference module. The commands routed via out-of-band (OOB) link while training data for the training models or input data for the trained models are routed via inband links.

Abstract: A traceability system includes: a Equipment table that stores production data of a product manufactured in a first process, in which an individual ID is appended to a product; a Equipment table that stores production data of a product manufactured in a second process, in which an individual ID is not appended to a product; a training data setting unit that creates a training data table that stores the Equipment table and the Equipment table, which are correlated with each other; a feature amount extracting unit that calculates a cycle time of a predetermined number of products manufactured in the past in the first process; a model creation section that creates a production time estimation model for estimating a production time at which a product has been manufactured in the second process on the basis of the cycle time of the products; and a production time estimating unit.

Abstract: A computational system can include digital circuitry and analog circuitry, for instance a digital processor and a quantum processor. The quantum processor can operate as a sample generator providing samples. Samples can be employed by the digital processing in implementing various machine learning techniques. For example, the computational system can perform unsupervised learning over an input space, for example via a discrete variational auto-encoder, and attempting to maximize the log-likelihood of an observed dataset. Maximizing the log-likelihood of the observed dataset can include generating a hierarchical approximating posterior.

Abstract: Methods, systems, and apparatus for mining feedback are described. A set of one or more lexical patterns associated with one or more of a suggestion and a defect report are determined and the set of one or more lexical patterns are matched against a plurality of feedback items to generate a distance learning training set. A distance learning technique is applied to the distance learning training set to generate a distance learning model and the distance learning model is used to identify one or more candidate feedback items of the plurality of feedback items, each of which is one or more of a candidate suggestion and a candidate defect report.