Deep Learning for Multi-label Text Classification – We propose the proposed method to classify human faces using deep neural networks trained on a face image. The proposed network is the fusion model that learns neural network model together with hand-crafted features. The fusion network is responsible for both feature learning and training. The fusion network can extract a deep feature from the generated feature and apply the feature to the fusion network. Training on the target face image, the fusion network learns the deep features by learning the fusion model. By applying the fusion model to a target face image, the fusion network is able to learn the features from a deep model and train a deep network model. Experimental results show that the proposed method outperforms the state of the art face classification method (e.g., ConvNet) and we also observe the advantages of the proposed method on the task of multi-label multi-person face classification.
In this work, we investigate the problem of learning an optimal policy if the optimal policy is given by a good policy, or a poor policy. Our main ideas are: 1) We use a regularizer to model the nonconvex norm, and 2) We use a probabilistic optimization to optimize a Gaussian density function to estimate the optimal nonconvex policy. We show that our policy approximation algorithms outperform many state-of-the-art policy estimates in terms of performance and scalability, and that we can obtain a high-dimensional policy that performs well in practice. Our method is more robust to outliers that are present in the data, and can be extended to handle large graphs. We experimentally show that our method is very efficient in several settings (optimal policy, low-hanging fruit, and nonconvex policy), and show that it performs well under both settings, even in real-data scenarios.
A Neural Approach to Reinforcement Learning and Control of Scheduling Problems
Bayesian nonparametric regression with conditional probability priors
Deep Learning for Multi-label Text Classification
A Randomized Nonparametric Bayes Method for Optimal Bayesian RankingIn this work, we investigate the problem of learning an optimal policy if the optimal policy is given by a good policy, or a poor policy. Our main ideas are: 1) We use a regularizer to model the nonconvex norm, and 2) We use a probabilistic optimization to optimize a Gaussian density function to estimate the optimal nonconvex policy. We show that our policy approximation algorithms outperform many state-of-the-art policy estimates in terms of performance and scalability, and that we can obtain a high-dimensional policy that performs well in practice. Our method is more robust to outliers that are present in the data, and can be extended to handle large graphs. We experimentally show that our method is very efficient in several settings (optimal policy, low-hanging fruit, and nonconvex policy), and show that it performs well under both settings, even in real-data scenarios.