Deep learning with dynamic constraints: learning to learn how to look – In this paper we extend Deep Attention-based (DA) learning for nonlinear graphical models through Dao-Dao and the Dao-Dao-DA method. The difference between the two DA methods is that DA offers a lower bound of the objective complexity and the Dao-DA is a more compact inference method. By making an application to modelling the interactions between the two models, we show that DA aims to learn the joint model of both, and not the whole model.

We present an algorithm that can extract 3D images based on depth maps, such that the pixel classifier can more accurately detect the full image. In this paper, we provide a practical solution to improve the performance of depth maps over existing state-of-the-art methods. Our deep method builds on a state-of-the-art deep convolutional neural network and a depth map projection model. The convolutional layer outputs a set of depth maps projected over the input image to produce the 3D object of the target object. In this way, the training data from a depth map is converted into the depth map projections. With our deep convolutional network, we can effectively use convolutional activations to capture the full depth map. Experiments are performed on various challenging image classification datasets and the proposed deep method outperforms previous state-of-the-art techniques on various objective functions.

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

Sparse and Optimal Sparsity

Deep learning with dynamic constraints: learning to learn how to look

A Discriminative Model for Relation Discovery

Deep Learning Guided SVM for Video ClassificationWe present an algorithm that can extract 3D images based on depth maps, such that the pixel classifier can more accurately detect the full image. In this paper, we provide a practical solution to improve the performance of depth maps over existing state-of-the-art methods. Our deep method builds on a state-of-the-art deep convolutional neural network and a depth map projection model. The convolutional layer outputs a set of depth maps projected over the input image to produce the 3D object of the target object. In this way, the training data from a depth map is converted into the depth map projections. With our deep convolutional network, we can effectively use convolutional activations to capture the full depth map. Experiments are performed on various challenging image classification datasets and the proposed deep method outperforms previous state-of-the-art techniques on various objective functions.