International Business Machines Corporation

Amazon Technologies, Inc.

Cicero Nogueira Dos Santos

Inkit Padhi

Youssef Mroueh

Tom Sercu

Bing Xiang

Bowen Zhou

Ramesh M Nallapati

Zhiguo Wang

Patrick Ng

Mattia Rigotti

Dheeru Dua

Peng Xu

John Richard Smith

Sudipta Sengupta

Alfio Massimiliano Gliozzo

Paulo Rodrigo Cavalin

Payel Das

Matthias Kormaksson

Bianca Zadrozny

Michele Merler

Maira Gatti De Bayser

Pierre L Dognin

Mauro Martino

Andrew Oliver Arnold

Henghui Zhu

Kahini Wadhawan

Igor Melnyk

Sebastian Gehrmann

Dejiao Zhang

Xiaofei Ma

Yifan Gao

Alexandre Rademaker

Luciano De Andrade Barbosa

Daria Bogdanova

Tom D J Sercu

Jarret Ross

Feng Nan

Techniques for predicting an answer to a question using a machine learning model are described. In some examples, the model predicts one or more answers to the question by: predicting at least two answers to the question using a first component of the question-answer model from a set of passages, generating, using a second component of the question-answer model, at least one question for each of the predicted at least two answers, and performing roundtrip predictions until each generated question only has one answer.

Questions disambiguation using generative evidence fusion and round-trip prediction

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

A machine learning system that implements Sobolev Independence Criterion (SIC) for feature selection is provided. The system receives a dataset including pairings of stimuli and responses. Each stimulus includes multiple features. The system generates a correctly paired sample of stimuli and responses from the dataset by pairing stimuli and responses according to the pairings of stimuli and responses in the dataset. The system generates an alternatively paired sample of stimuli and responses from the dataset by pairing stimuli and responses differently than the pairings of stimuli and responses in the dataset. The system determines a witness function and a feature importance distribution across the features that optimizes a cost function that is evaluated based on the correctly paired and alternatively paired samples of the dataset. The system selects one or more features based on the computed feature importance distribution.

Feature selection using Sobolev Independence Criterion

A computer-implemented method according to one aspect includes training a latent variable model (LVM), utilizing labeled data and unlabeled data within a data set; training a classifier, utilizing the labeled data and associated labels within the data set; and generating new data having a predetermined set of labels, utilizing the trained LVM and the trained classifier.

Generating attribute-based samples

Mechanisms are provided for implementing a Question Answering (QA) system utilizing a trained generator of a generative adversarial network (GAN) that generates a bag-of-ngrams (BoN) output representing unlabeled data for performing a natural language processing operation. The QA system obtains a plurality of candidate answers to a natural language question, where each candidate answer comprises one or more ngrams. For each candidate answer, a confidence score is generated based on a comparison of the one or more ngrams in the candidate answer to ngrams in the BoN output of the generator neural network of the GAN. A final answer to the input natural language question is selected from the plurality of candidate answers based on the confidence scores associated with the candidate answers, and is output.

Question Answering using trained generative adversarial network based modeling of text

Techniques for generation of synthetic queries from customer data for training of document querying machine learning (ML) models as a service are described. A service may receive one or more documents from a user, generate a set of question and answer pairs from the one or more documents from the user using a machine learning model trained to predict a question from an answer, and store the set of question and answer pairs generated from the one or more documents from the user. The question and answer pairs may be used to train another machine learning model, for example, a document ranking model, a passage ranking model, a question/answer model, or a frequently asked question (FAQ) model.

Systems, apparatuses, and methods to generate synthetic queries from customer data for training of document querying machine learning models

A machine learning system receives a witness function that is determined based on an initial sample of a dataset comprising multiple pairs of stimuli and responses. Each stimulus includes multiple features. The system receives a holdout sample of the dataset comprising one or more pairs of stimuli and responses that are not used to determine the witness function. The system generates a simulated sample based on the holdout sample. Values of a particular feature of the stimuli of the simulated sample are predicted based on values of features other than the particular feature of the stimuli of the simulated sample. The system applies the holdout sample to the witness function to obtain a first result. The system applies the simulated sample to the witness function to obtain a second result. The system determines whether to select the particular feature based on a comparison between the first result and the second result.

False detection rate control with null-hypothesis

Mechanisms are provided to implement a generative adversarial network (GAN) for natural language processing. With these mechanisms, a generator neural network of the GAN is configured to generate a bag-of-ngrams (BoN) output based on a noise vector input and a discriminator neural network of the GAN is configured to receive a BoN input, where the BoN input is either the BoN output from the generator neural network or a BoN input associated with an actual portion of natural language text. The mechanisms further configure the discriminator neural network of the GAN to output an indication of a probability as to whether the input BoN is from the actual portion of natural language text or is the BoN output of the generator neural network. Moreover, the mechanisms train the generator neural network and discriminator neural network based on a feedback mechanism that compares the output indication from the discriminator neural network to an indicator of whether the input BoN is from the actual portion of natural language text of the BoN output of the generator neural network.

Generative adversarial network based modeling of text for natural language processing

Techniques for content generation are provided. A plurality of discriminative terms is determined based at least in part on a first plurality of documents that are related to a first concept, and a plurality of positive exemplars and a plurality of negative exemplars are identified using the plurality of discriminative terms. A first machine learning (ML) model is trained to classify images into concepts, based on the plurality of positive exemplars and the plurality of negative exemplars. A second concept related to the first concept is then determined, based on the first ML model. A second ML model is trained to generate images based on the second concept, and a first image is generated using the second ML model. The first image is then refined using a style transfer ML model that was trained using a plurality of style images.

Automatic generation of content using multimedia

A computer-implemented method includes extracting a plurality of topics from a plurality of unlabeled social media posts, mapping the plurality of topics to a plurality of frequencies, each frequency in the plurality of frequencies indicating how often a corresponding topic in the plurality of topics occurs in the plurality of unlabeled social media posts, and predicting, based in part on the plurality of frequencies, a future social media posting behavior of a specific social media user, wherein the future social media posting behavior includes a specific topic about which the specific social media user is likely to post at a time in the future and a frequency with which the specific topic is likely to occur in posts of the specific social media user that are created at the time in the future.

Predicting user posting behavior in social media applications

Determining semantically equivalent text or questions using hybrid representations based on neural network learning. Weighted bag-of-words and convolutional neural networks (CNN) based distributed vector representations of questions or text may be generated to compute the semantic similarity between questions or text. Weighted bag-of-words and CNN based distributed vector representations may be jointly used to compute the semantic similarity. A pair-wise ranking loss function trains neural network. In one embodiment, the parameters of the system are trained by minimizing a pair-wise ranking loss function over a training set using stochastic gradient descent (SGD).

Machine learning and training a computer-implemented neural network to retrieve semantically equivalent questions using hybrid in-memory representations

Growing community of inventors

White Plains, NY, United States of America

Cicero Nogueira Dos Santos