Abstract: A computer-implemented method, system and computer program product for training a machine learning model using noisily labeled data. A classification model is built in which a classified dataset is inputted, where the classified dataset includes label noise. Based on the input, the classification model generates a prediction of class probabilities. Furthermore, a second model is built with the same architecture as the classification model, where the second model is a moving average of the classification model, and where the second model generates a prediction of class probabilities. Weight factors used to weight such predictions of these models are generated by the artificial neural network (ANN), in which the weighted predictions are used by the ANN to obtain a prediction of class probabilities. The predictions of class probabilities of the ANN and the classification model are then combined to train the machine learning model.

Abstract: Various embodiments are provided for providing machine learning with noisy label data in a computing environment using one or more processors in a computing system. A label corruption probability of noisy labels may be estimated for selected data from a dataset using temporal inconsistency in a machine model prediction during a training operation in a neural network.

Abstract: A method for generating and displaying an embedding of multivariate time series data in an embedded space is provided. The method may include optimizing a generative deep learning neural network by adding a regularization term to train the generative deep learning neural network, wherein the regularization term maintains a temporal relationship between data points associated with the multivariate time series data when representing the data points of the multivariate time series data in an embedded space. The method may further include, in response to receiving the multivariate time series data, applying the optimized generative deep learning neural network to the input, and representing and displaying the multivariate time series data in the embedded space such that the temporal relationship between the data points from the input is captured by and presented in the representation, wherein the representation comprises an embedding of the multivariate time series data in the embedded space.

Abstract: A method, a computer program product, and a computer system fuse features for multi-modal classifications for a plurality of modality inputs. The method includes receiving a request indicative of the modality inputs to be selected. The method includes performing an embeddings level fusion operation to concatenate features from the modality inputs. The method includes performing a multi-modal discriminative feature level fusion operation that integrates feature representations learned by applying different network structures on the modality inputs. The method includes determining weights of the concatenated features and the feature representations based on a measure of the concatenated features and the feature representations indicative of affecting a final prediction performance. The method includes generating fused features for the modality inputs based on the concatenated features, the feature representations, and the weights.

Abstract: Various embodiments are provided for providing machine learning with noisy label data in a computing environment using one or more processors in a computing system. A label corruption probability of noisy labels may be estimated for selected data from a dataset using temporal inconsistency in a machine model prediction during a training operation in a neural network.

Abstract: One embodiment provides a method for face liveness detection. The method comprises receiving a first image comprising a face of a user, determining one or more two-dimensional (2D) facial landmark points based on the first image, and determining a three-dimensional (3D) pose of the face in the first image based on the one or more determined 2D facial landmark points and one or more corresponding 3D facial landmark points in a 3D face model for the user. The method further comprises determining a homography mapping between the one or more determined 2D facial landmark points and one or more corresponding 3D facial landmark points that are perspectively projected based on the 3D pose, and determining liveness of the face in the first image based on the homography mapping.

Abstract: One embodiment provides a method for face liveness detection. The method comprises receiving a first image comprising a face of a user, determining one or more two-dimensional (2D) facial landmark points based on the first image, and determining a three-dimensional (3D) pose of the face in the first image based on the one or more determined 2D facial landmark points and one or more corresponding 3D facial landmark points in a 3D face model for the user. The method further comprises determining a homography mapping between the one or more determined 2D facial landmark points and one or more corresponding 3D facial landmark points that are perspectively projected based on the 3D pose, and determining liveness of the face in the first image based on the homography mapping.