A Fusion and Localization Strategy for the Visual Tracking of a Moving Object – Automatic tracking of robotic subjects on large-scale scenes has always been a challenging problem. We propose an approach to this problem that exploits the ability of a spatial system to learn a spatial distribution for autonomous tracking. Our approach provides a system-level model that can be used to learn a spatial distribution for autonomous tracking. We show that, in general, the network can be used for learning to track subjects by learning the model-level representation. Moreover, we show that the spatial representation is not only useful for learning to track subjects, but also can be employed to learn a spatial localization strategy. Experiments on a real-world dataset show that using a spatial representation can improve the tracking accuracy, as the spatial location of robotic objects is significantly more relevant to the spatial localization prediction at hand.

Learning to predict future events is challenging because of the large, complex, and unpredictable nature of the data. Despite the enormous volume of available data, supervised learning has made great progress in recent years in learning to predict the future rather than in predicting the past. In this paper, we present a framework for modeling and predicting the future of data by non-Gaussian prior approximating latent Gaussian processes. The underlying assumptions are to be established in the context of non-Gaussian prior approximating learning, and we further elaborate on these assumptions in a neural-network architecture. We evaluate this network on two datasets: the Long Short-Term Memory and the Stanford Attention Framework dataset, where we show that the model achieves state-of-the-art performance with good accuracy.

Sparse and Optimal Sparsity

A Discriminative Model for Relation Discovery

A Fusion and Localization Strategy for the Visual Tracking of a Moving Object

Learning and Analyzing Phrase Based Phrase Based Speech Recognition

Hierarchical Gaussian Process ModelsLearning to predict future events is challenging because of the large, complex, and unpredictable nature of the data. Despite the enormous volume of available data, supervised learning has made great progress in recent years in learning to predict the future rather than in predicting the past. In this paper, we present a framework for modeling and predicting the future of data by non-Gaussian prior approximating latent Gaussian processes. The underlying assumptions are to be established in the context of non-Gaussian prior approximating learning, and we further elaborate on these assumptions in a neural-network architecture. We evaluate this network on two datasets: the Long Short-Term Memory and the Stanford Attention Framework dataset, where we show that the model achieves state-of-the-art performance with good accuracy.