Accenture Global Solutions Limited

Timothy M Shea

Andrew E Fano

Lavinia Andreea Danielescu

David William Vinson

Kenneth Stewart

Maziyar Baran Pouyan

Yao A Yang

Saeideh Shahrokh Esfahani

Eric Michael Gallo

Noah Gideon Pacik-Nelson

Jesus Sanchez-Macias

Kenneth Michael Stewart

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for updating a trained gesture recognition model deployed on a neuromorphic processor that has been trained to process data that characterizes the new gesture and to determine a gesture classification for the gesture are described. A method includes receiving data that characterizes a new gesture and processing the data to generate a new embedding in a latent space. For each of multiple clusters of reference embeddings in the latent space, a respective distance in the latent space between the cluster of reference embedding and the new embedding is determined. A determination is made, based on applying one or more learning rules to the distances, one or more procedures to update the gesture recognition model. A determination is made, in accordance with the determined procedure(s), an update to values of one or more parameters of the gesture recognition model.

Self-learning neuromorphic gesture recognition models

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for recognizing speech using a spiking neural network acoustic model implemented on a neuromorphic processor are described. In one aspect, a method includes receiving, a trained acoustic model implemented as a spiking neural network (SNN) on a neuromorphic processor of a client device, a set of feature coefficients that represent acoustic energy of input audio received from a microphone communicably coupled to the client device. The acoustic model is trained to predict speech sounds based on input feature coefficients. The acoustic model generates output data indicating predicted speech sounds corresponding to the set of feature coefficients that represent the input audio received from the microphone. The neuromorphic processor updates one or more parameters of the acoustic model using one or more learning rules and the predicted speech sounds of the output data.

Self-learning neuromorphic acoustic model for speech recognition

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for clustering data are disclosed. In one aspect, a method includes the actions of receiving feature vectors. The actions further include accessing rules that each relate one or more values of the feature vectors to a respective label of a plurality of labels. The actions further include, based on the rules, generating heuristics that each identify related values of the feature vectors. The actions further include, for each of the heuristics, generating a matrix that reflects a similarity of the feature vectors. The actions further include, based on the matrices that each reflects a respective similarity of the feature vectors, generating clusters that each include a subset of the feature vectors. The actions further include, for each cluster, determining a label of the plurality of labels.

Data clustering

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for clustering data are disclosed. In one aspect, a method includes the actions of receiving feature vectors. The actions further include, for a subset of the feature vectors, accessing a first label. The actions further include generating a classifier that is configured to associate a given feature vector with a feature vector of the subset of the feature vectors. The actions further include applying the feature vectors that are not included in the subset of the feature vectors to the classifier. The actions further include generating a dissimilarity matrix. The actions further include, based on the dissimilarity matrix, generating a graph. The actions further include, for each node of the graph, determining a second label. The actions further include, based on the second labels and the first labels, determining a training label for each feature vector.

Growing community of inventors

Merced, CA, United States of America

Timothy M Shea