Abstract: Hybrid training of deep networks includes a multi-layer neural network. The training includes setting a current learning algorithm for the multi-layer neural network to a first learning algorithm. The training further includes iteratively applying training data to the neural network, determining a gradient for parameters of the neural network based on the applying of the training data, updating the parameters based on the current learning algorithm, and determining whether the current learning algorithm should be switched to a second learning algorithm based on the updating. The training further includes, in response to the determining that the current learning algorithm should be switched to a second learning algorithm, changing the current learning algorithm to the second learning algorithm and initializing a learning rate of the second learning algorithm based on the gradient and a step used by the first learning algorithm to update the parameters of the neural network.

Abstract: Artificial neural networks (ANNs) are a distributed computing model in which computation is accomplished with many simple processing units, called neurons, with data embodied by the connections between neurons, called synapses and by the strength of these connections, the synaptic weights. An attractive implementation of ANNs uses the conductance of non-volatile memory (NVM) elements to record the synaptic weight, with the important multiply—accumulate step performed in place, at the data. In this application, the non-idealities in the response of the NVM such as nonlinearity, saturation, stochasticity and asymmetry in response to programming pulses lead to reduced network performance compared to an ideal network implementation. A method is shown that improves performance by distributing the synaptic weight across multiple conductances of varying significance, implementing carry operations between less-significant signed analog conductance-pairs to more-significant analog conductance-pairs.

Abstract: In order to facilitate machine learning for prediction using distinct dataset types, systems and methods include collecting content information from archived websites databases. Collecting historical event information from online sources, where the historical event information is associated with a plurality of historical events. Generating event-dependent products training datasets based on the content information and the historical event information, where the event-dependent content training datasets defines for content historical events that are associated with attributes of the content, attribute change of the content, or both. Training an attribute prediction machine learning model based on the event-dependent content training datasets. Applying the trained attribute prediction machine learning model to additional event information to predict, for content, a future attribute estimate, a future attribute change estimate, or both.

Abstract: The technology disclosed proposes using a combination of computationally cheap, less-accurate bag of words (BoW) model and computationally expensive, more-accurate long short-term memory (LSTM) model to perform natural processing tasks such as sentiment analysis. The use of cheap, less-accurate BoW model is referred to herein as “skimming”. The use of expensive, more-accurate LSTM model is referred to herein as “reading”. The technology disclosed presents a probability-based guider (PBG). PBG combines the use of BoW model and the LSTM model. PBG uses a probability thresholding strategy to determine, based on the results of the BoW model, whether to invoke the LSTM model for reliably classifying a sentence as positive or negative. The technology disclosed also presents a deep neural network-based decision network (DDN) that is trained to learn the relationship between the BoW model and the LSTM model and to invoke only one of the two models.

Abstract: Ensemble classification algorithms having subclass resolution are disclosed. An example disclosed apparatus includes a fingerprint generator to generate a fingerprint of class probabilities of each of a plurality of samples, a distribution creator to create a distribution of the samples based on the generated fingerprints, and a distribution applicator to apply the distribution to a population to predict sub-class probabilities of each of the population.

Abstract: Disclosed is a data processing system which includes compile time logic configured to section a graph into a sequence of subgraphs, the sequence of subgraphs including at least a first subgraph. The compile time logic configures the first subgraph to generate a plurality of output tiles of an output tensor. A runtime logic configured with the compile time logic is to execute the sequence of subgraphs to generate, at the output of the first subgraph, the plurality of output tiles of the output tensor, and write the plurality of output tiles in a memory in an overlapping configuration. In an example, an overlapping region between any two neighboring output tiles of the plurality of output tiles comprises a summation of a corresponding region of a first neighboring output tile and a corresponding region of a second neighboring output tile.

Abstract: Methods and systems of classification model evaluation are described. A processor may generate a classifier track comprising visual indicators representing the set of classifier labels. The classifier labels may be based on output data of a classification model. The processor may generate a label track comprising visual indicators representing a set of observed labels received from a device. The processor may output the classifier track and the label track on a user interface. The processor may receive a request to evaluate the classification model, where the request may indicate a performance metric. The processor may identify a set of operators associated with the performance metric. The processor may execute the identified set of operators on the classifier track and the label track. The processor may generate a performance track indicating the performance metric of the classification model. The processor may output the performance track on the user interface.

Abstract: A system and method for providing customized content recommendations to a user based on extrapolated data is described. The system may receive an answer to a question from a user and determine a second unanswered question having a threshold relationship level with the question. The system may compute a predicted probability that the user would answer the second question correctly, calculate a first comparative skill level of the user among a cohort of similar users, and rank the first comparative skill level of the user against a second comparative skill level of the cohort to determine a skill gap of the user. Further, in some implementations, the system may generate a search query based on the skill gap of the user, and determine recommended content customized to the skill gap of the user.

Abstract: Techniques of forecasting web metrics involve generating, prior to the end of a period of time, a probability of a metric taking on an anomalous value, e.g., a value indicative of an anomaly with respect to web traffic, at the end of the period based on previous values of the metric. Such a probability is based on a distribution of predicted values of the metric at some previous period of time. For example, a web server may use actual values of the number of bounces collected at hourly intervals in the middle of a day to predict a number of bounces at the end of the current day. Further, the web server may also compute a confidence interval to determine whether a predicted end-of-day number of bounces may be considered anomalous. The width of the confidence interval indicates the probability that a predicted end-of-day number of bounces has an anomalous value.

Abstract: Logic may identify feature contributions to erroneous predictions by predictive models. Logic may provide a set of two or more models. Each model may train based on a training dataset and test based on a testing dataset and two or more models may be unique. Logic may test the set during a monitoring period. Logic may perform residual modeling on each model in the set during the monitoring period and may determine a list of input features that contribute to a residual of each model of the set. A residual comprises a difference between a predicted result and an expected result. Logic may generate a combined list of the input features from the set and may rank the input features. Logic may perform a voting process to generate the ranks for the input features. And logic may classify features as exogenous or endogenous based on a threshold and the ranks.

Abstract: According to embodiments, methods, systems, and computer program products are provided for receiving one or more input compositions comprising one or more materials, assigning a material vector to each material, learning, for each of the input compositions, a composition vector based on the material vectors of the materials that form each composition, assigning predicted rating values having a confidence level to each of the composition vectors, selecting a composition to be rated based on the confidence levels, presenting the selected composition to be rated to a user, receiving a user rating for the composition to be rated; adjusting the predicted rating values and confidence levels of the composition vectors that have not been rated by the user, and generating a predictive model to predict a user's ratings for compositions when confidence levels of each composition vector is above a predetermined threshold value.

Abstract: A mechanism is provided for determining a predicted performance of a database com. A first model for a database is determined using machine learning and training data based on monitoring the database operating in a production environment. A second model for the database is determined by combining the first model and a knee of curve formula for the database. The second model is stored for use in determining the predicted performance of the database in response to a database query.

Abstract: To realize a reservoir computing system easily implemented as hardware, provided is a reservoir computing system including a reservoir operable to output an inherent output signal in response to an input signal. An input node is operable to supply the reservoir with an input signal corresponding to input data, and an output node is operable to output an output value corresponding to an output signal that is output by the reservoir in response to the input data. An adaptive filter is operable to output output data based on a result obtained by weighting a plurality of the output values output from the output node at a plurality of timings with a plurality of weights. Also provided are a learning method and a computer program product.

Abstract: Embodiments of the present disclosure allow accuracy of prediction of pollution to be improved. In operation, a prediction of pollution in a future time period is determined. The prediction of pollution indicates predicted data related to a pollution index. Then, matching historical data for the predicted data is determined from historical data related to the pollution index. The matching historical data is obtained in a historical time period corresponding to the future time period. Based on the matching historical data, the prediction of pollution is refined.

Abstract: According to embodiments, methods, systems, and computer program products are provided for receiving one or more input compositions comprising one or more materials, assigning a material vector to each material, learning, for each of the input compositions, a composition vector based on the material vectors of the materials that form each composition, assigning predicted rating values having a confidence level to each of the composition vectors, selecting a composition to be rated based on the confidence levels, presenting the selected composition to be rated to a user, receiving a user rating for the composition to be rated; adjusting the predicted rating values and confidence levels of the composition vectors that have not been rated by the user, and generating a predictive model to predict a user's ratings for compositions when confidence levels of each composition vector is above a predetermined threshold value.

Abstract: To realize a reservoir computing system easily implemented as hardware, provided is a reservoir computing system including a reservoir operable to output an inherent output signal in response to an input signal. An input node is operable to supply the reservoir with an input signal corresponding to input data, and an output node is operable to output an output value corresponding to an output signal that is output by the reservoir in response to the input data. An adaptive filter is operable to output output data based on a result obtained by weighting a plurality of the output values output from the output node at a plurality of timings with a plurality of weights. Also provided are a learning method and a computer program product.

Abstract: Example implementations described herein involve an interface for calculating and displaying missing links for data represented as a bipartite network, along with novel methods for improving link prediction algorithms in the related art. Through example implementations described herein, the accuracy of link prediction algorithms can be improved upon, thereby providing the user with a more accurate understanding of the data in the bipartite network.

Abstract: Herein, horizontally scalable techniques efficiently configure machine learning algorithms for optimal accuracy and without informed inputs. In an embodiment, for each particular hyperparameter, and for each epoch, a computer processes the particular hyperparameter. An epoch explores one hyperparameter based on hyperparameter tuples. A respective score is calculated from each tuple. The tuple contains a distinct combination of values, each of which is contained in a value range of a distinct hyperparameter. All values of a tuple that belong to the particular hyperparameter are distinct. All values of a tuple that belong to other hyperparameters are held constant. The value range of the particular hyperparameter is narrowed based on an intersection point of a first line based on the scores and a second line based on the scores. A machine learning algorithm is optimally configured from repeatedly narrowed value ranges of hyperparameters. The configured algorithm is invoked to obtain a result.

Abstract: The present disclosure provides systems and methods for synthetic data generation. A recurrent neural network can be trained for synthetic data generation by obtaining a sequence of elements and determining, using a classifier, that the sequence corresponds to a token. In response to the determination, a recurrent neural network configured to use a first vocabulary including the elements can be modified to use a second vocabulary, the second vocabulary including the token and the first vocabulary. The modified recurrent neural network can be trained using the token and the sequence of elements. The trained recurrent neural network can be used to generate synthetic data. A classifier can detect sequences of elements in the synthetic data corresponding to tokens. The tokens can replace the sequences of elements in the generated synthetic data and can be provided to the trained recurrent neural network to continue synthetic data generation.

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 durations 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 duration 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.