International Business Machines Corporation

Nuance Communications, Inc.

Pierre L Dognin

Vaibhava Goel

Igor Melnyk

Steven John Rennie

Youssef Mroueh

Petr Fousek

Payel Das

Peder Andreas Olsen

Cicero Nogueira Dos Santos

Inkit Padhi

John Randall Hershey

Tom Sercu

Tom D J Sercu

Jarret Ross

A method, system and apparatus of ensembling, including inputting a set of models that predict different sets of attributes, determining a source set of attributes and a target set of attributes using a barycenter with an optimal transport metric, and determining a consensus among the set of models whose predictions are defined on the source set of attributes.

Wasserstein barycenter model ensembling

Systems, computer-implemented methods, and computer program products to facilitate a dual learning bridge between text and a knowledge graph are provided. According to an embodiment, a system can comprise a processor that executes computer executable components stored in memory. The computer executable components comprise a model component that employs a model to learn associations between text data and a knowledge graph. The computer executable components further comprise a translation component that uses the model to bidirectionally translate second text data and one or more knowledge graph paths based on the associations.

Dual learning bridge between text and knowledge graph

A computing device receives a data X and Y, each having N samples. A function f(x,y) is defined to be a trainable neural network based on the data X and the data Y. A permuted version of the data Y is created. A loss mean is computed based on the trainable neural network f(x,y), the permuted version of the sample data Y, and a trainable scalar variable &#x3b7;. A loss with respect to the scalar variable &#x3b7; and the trainable neural network is minimized. Upon determining that the loss is at or below the predetermined threshold, estimating a mutual information (MI) between a test data Xand Y. If the estimated MI is above a predetermined threshold, the test data Xand Yis deemed to be dependent. Otherwise, it is deemed to be independent.

Mutual information neural estimation with Eta-trick

A speech processing method and arrangement are described. A dynamic noise adaptation (DNA) model characterizes a speech input reflecting effects of background noise. A null noise DNA model characterizes the speech input based on reflecting a null noise mismatch condition. A DNA interaction model performs Bayesian model selection and re-weighting of the DNA model and the null noise DNA model to realize a modified DNA model characterizing the speech input for automatic speech recognition and compensating for noise to a varying degree depending on relative probabilities of the DNA model and the null noise DNA model.

System and method for dynamic noise adaptation for robust automatic speech recognition

A method for training a deep neural network (DNN), comprises receiving and formatting speech data for the training, performing Hessian-free sequence training (HFST) on a first subset of a plurality of subsets of the speech data, and iteratively performing the HFST on successive subsets of the plurality of subsets of the speech data, wherein iteratively performing the HFST comprises reusing information from at least one previous iteration.

Systems and methods for combining stochastic average gradient and hessian-free optimization for sequence training of deep neural networks

Access is obtained to a large reference acoustic model for automatic speech recognition. The large reference acoustic model has L states modeled by L mixture models, and the large reference acoustic model has N components. A desired number of components N, less than N, to be used in a restructured acoustic model derived from the reference acoustic model, is identified. The desired number of components Nis selected based on a computing environment in which the restructured acoustic model is to be deployed. The restructured acoustic model also has L states. For each given one of the L mixture models in the reference acoustic model, a merge sequence is built which records, for a given cost function, sequential mergers of pairs of the components associated with the given one of the mixture models. A portion of the Ncomponents is assigned to each of the L states in the restructured acoustic model. The restructured acoustic model is built by, for each given one of the L states in the restructured acoustic model, applying the merge sequence to a corresponding one of the L mixture models in the reference acoustic model until the portion of the Ncomponents assigned to the given one of the L states is achieved.

Growing community of inventors

White Plains, NY, United States of America

Pierre L Dognin