On the Reliable Detection of Non-Linear Noise in Continuous Background Subtasks – This paper presents a novel method for the detection of non-linear noise in a continuous background task. We construct a graph-space to model the background, and apply the method to solve a real-world problem in recommender system for automatic recommendation. The graph structures are derived using an alternating direction method of multiplicative and univariate analysis, and its similarity of the model structure to the input graph is estimated using a graph classifier. The graph classifier achieves performance at both classification and benchmark with the highest classification result. The graph classifier achieves a good performance for multi-output classification.

The object oriented visualization task is of great importance for many visual science applications such as robotics, computer vision and object retrieval. A major challenge for visualization tasks is that, due to the large training set, tasks are hard to train efficiently. In this paper, we propose a novel method to learn a deep convolutional neural network which is capable of modeling the object oriented visual representation as a sequence of multi-level object pose and orientation hierarchies. In particular, we propose a three-stage learning approach to encode pose and orientation hierarchies in the deep network, and to learn the hierarchies in the deep network without any prior knowledge of its parameters. This method is a first step towards the use of object oriented visual representation through the use of neural networks to create a multi-level object-oriented object model. Experiments on the Cityscapes dataset, Cityscapes-01 and Cityscapes-02 datasets illustrate the effectiveness of our approach compared to other CNNs.

Diet in the Wild: Large-Scale Detection of Exercise-Related Events from Body States using Mobile Phones

Multiagent Learning with Spatiotemporal Context: Application to Predicting Consumer’s Behaviors

On the Reliable Detection of Non-Linear Noise in Continuous Background Subtasks

Evaluating the Accuracy of Text Trackers using the Inductive Logic Problem

Learning Visual Representations of Objects and their Attributes in HEMThe object oriented visualization task is of great importance for many visual science applications such as robotics, computer vision and object retrieval. A major challenge for visualization tasks is that, due to the large training set, tasks are hard to train efficiently. In this paper, we propose a novel method to learn a deep convolutional neural network which is capable of modeling the object oriented visual representation as a sequence of multi-level object pose and orientation hierarchies. In particular, we propose a three-stage learning approach to encode pose and orientation hierarchies in the deep network, and to learn the hierarchies in the deep network without any prior knowledge of its parameters. This method is a first step towards the use of object oriented visual representation through the use of neural networks to create a multi-level object-oriented object model. Experiments on the Cityscapes dataset, Cityscapes-01 and Cityscapes-02 datasets illustrate the effectiveness of our approach compared to other CNNs.