Adversarial Data Analysis in Multi-label Classification – We use the model for both action recognition and classification tasks. Unlike previous approaches, we do not require a large number of examples to learn the structure, and the structure is learned automatically. Therefore, it is natural to ask whether the structure of the task is more informative than the examples it is learning from. This paper proposes a new model based on the deep reinforcement learning method. The model is built with three layers: a layer in which an agent can control the environment, a layer in which an agent uses its actions and a layer called the hidden layer to represent the reward-value relationship between actions. The hidden layer is learned from the learned model through reinforcement learning, and the reward-value relationship between actions is learned by using the reinforcement learning techniques. An evaluation on the UCI dataset of 9,891 actions demonstrates the effectiveness of the model of learning from examples.

The paper presents a practical and robust method for learning and computing face models in the presence of natural occlusion. Our algorithm is based on a discriminative representation over faces, which is an essential step to learning the structure of a face database. We prove that both the face recognition and face estimation are NP-hard, without taking into account the presence of occlusion. We apply our method to several complex face datasets and show results on simulated and real-world datasets.

The Computational Phenomenon of Quantum-like Particles

Learning to rank for automatic speech synthesis

Adversarial Data Analysis in Multi-label Classification

Deep Multi-Objective Goal Modeling

Lip Localization via Semi-Local KernelsThe paper presents a practical and robust method for learning and computing face models in the presence of natural occlusion. Our algorithm is based on a discriminative representation over faces, which is an essential step to learning the structure of a face database. We prove that both the face recognition and face estimation are NP-hard, without taking into account the presence of occlusion. We apply our method to several complex face datasets and show results on simulated and real-world datasets.