Multi-view Graph Convolutional Neural Network – Many recent methods for deep reinforcement learning (RL) rely on the use of multi-dimensional convolutional neural networks. This paper investigates the use of multi-dimensional convolutional neural networks (MDS-NNs) for non-linear reinforcement learning (NRL) tasks. We present a novel approach that employs convolutional networks for nonlinear RL tasks, which, by a neural network’s own, leads to efficient policy learning that avoids the need for costly re-training. We show that a nonlinear RL task may be more suited to a multi-dimensional MDS-NN, as it has a fully-connected network with an input manifold and a policy space. Moreover, we show that a nonlinear RL task (e.g., a simple image navigation task) may be more attractive to a multi-dimensional MDS-NN than a simple image detection task. Moreover, we obtain efficient policies for a simple RL task as a result of our approach.
Deep-learning methods have been successfully applied to the design of medical domain applications and to medical imaging. However, deep models, such as deep neural networks (DNNs), do not exhibit robustness when applied to medical data. In this paper, we propose a hybrid, deep-learning-centric, efficient and scalable deep-learning method to enhance the performance of DNNs and other deep-learning-based approaches. The proposed method aims to improve the performance of DNNs by enhancing some discriminative representations of the data using deep learning. The proposed method is tested in three different medical domain applications, the first in an online MRI data set for the purpose of validation. The performance improvements are achieved with different DNN models, for which DNNs are not available and for which deep models are not implemented. In this paper, we perform a systematic empirical evaluation of our DNN-based deep-learning method for improving the performance of DNN-based deep vision approaches. The results indicate that the proposed method is competitive in terms of its effectiveness and efficiency.
Multi-view Graph Convolutional Neural Network
Probabilistic Models for Constraint-Free Semantic Parsing with Sparse Linear Programming
3D Scanning Network for Segmentation of Medical ImagesDeep-learning methods have been successfully applied to the design of medical domain applications and to medical imaging. However, deep models, such as deep neural networks (DNNs), do not exhibit robustness when applied to medical data. In this paper, we propose a hybrid, deep-learning-centric, efficient and scalable deep-learning method to enhance the performance of DNNs and other deep-learning-based approaches. The proposed method aims to improve the performance of DNNs by enhancing some discriminative representations of the data using deep learning. The proposed method is tested in three different medical domain applications, the first in an online MRI data set for the purpose of validation. The performance improvements are achieved with different DNN models, for which DNNs are not available and for which deep models are not implemented. In this paper, we perform a systematic empirical evaluation of our DNN-based deep-learning method for improving the performance of DNN-based deep vision approaches. The results indicate that the proposed method is competitive in terms of its effectiveness and efficiency.