A Comparative Analysis of Support Vector Machines – We present a principled alternative to the conventional hardware approach to non-convex and semi-supervised non-parametric classification using deep neural networks (DNNs). In contrast to prior approaches, the DNN formulation can be directly modeled by a matrix and an unaligned matrix. Hence, we provide a principled framework for embedding DNN models in the model space through convolutional neural networks (CNNs). Such an approach is also applicable to general-purpose classification tasks in which CNNs are used as a proxy for the data of the target classification task. We show that this framework is applicable to unsupervised and supervised learning tasks, and demonstrate its superior performance in various instances. We further provide an empirical evaluation demonstrating the effectiveness of our approach for supervised and unsupervised classification tasks.

We propose a deep generative model that can learn to produce a variety of emotions. Our model includes an external representation of the emotions of a given scene in which the emotions of human beings are encoded. To learn a complex emotion representation for the scenes, we combine human-level language and external representations of the emotions from the world as a generative model of the scene. By leveraging our generative model, we generate visual summaries of the emotion of human beings that we can then use to make predictions about the emotions of the human and the environment. The models use these summaries for the task of estimating human emotion recognition.

Learning to Predict Saccadic Charts from Data Captions

Learning from non-deterministic examples

A Comparative Analysis of Support Vector Machines

Object Classification through Deep Learning of Embodied Natural Features and Subspace

Learning a Human-Level Auditory Processing UnitWe propose a deep generative model that can learn to produce a variety of emotions. Our model includes an external representation of the emotions of a given scene in which the emotions of human beings are encoded. To learn a complex emotion representation for the scenes, we combine human-level language and external representations of the emotions from the world as a generative model of the scene. By leveraging our generative model, we generate visual summaries of the emotion of human beings that we can then use to make predictions about the emotions of the human and the environment. The models use these summaries for the task of estimating human emotion recognition.