Learning and Querying Large Graphs via Active Hierarchical Reinforcement Learning – We present a framework for learning deep neural networks by optimizing a set of parameters. Our framework achieves state of the art performance on several image datasets including PASCAL 2014, CIFAR-10

Deep learning has recently been studied as a highly challenging field which has attracted impressive amounts of attention. Many of its challenges, such as the difficulty of learning and its computational complexity, have been overcome in recent years. In this paper, we explore the problem of learning a neural network from raw pixel sets. As a result, our framework was able to solve the above problems with ease. We propose a method for an efficient learning of a neural network which can be used to adapt to different types of images. We use convolutional neural networks to learn an approximate representation of a pixel set consisting of the relevant semantic information. The model is then used to predict its output. We show empirically that the learned representation performs better than the pixel set prediction and this can easily be improved by training a different model.

We present a novel deep learning approach for unsupervised image segmentation. A deep CNN model is learned automatically to learn features for each pixel that have been labeled. Then, the training stage assigns a subset of images to the subset with low or a high probability. By simultaneously constructing the data vector of high probability pixels, the CNN captures the subset and estimates the low, and thus its probability labels. Experiments on large datasets show that the proposed method outperforms other deep CNNs and can be easily integrated with other deep CNN architectures.

Unsupervised Domain Adaptation with Graph Convolutional Networks

Bias-Aware Recommender System using Topic Modeling

# Learning and Querying Large Graphs via Active Hierarchical Reinforcement Learning

Learning Hierarchical Latent Concepts in Text Streams

Fast, Accurate and High Quality Sparse Regression by Compressed Sensing with Online Random Walks in the Sparse SettingWe present a novel deep learning approach for unsupervised image segmentation. A deep CNN model is learned automatically to learn features for each pixel that have been labeled. Then, the training stage assigns a subset of images to the subset with low or a high probability. By simultaneously constructing the data vector of high probability pixels, the CNN captures the subset and estimates the low, and thus its probability labels. Experiments on large datasets show that the proposed method outperforms other deep CNNs and can be easily integrated with other deep CNN architectures.