Learning Sparse Feature Representations Using Probabilistic Quadtrees and Deep Belief Nets

Abstract

Learning sparse feature representations is a useful instrument for solving an unsupervised learning problem. In this paper, we present three labeled handwritten digit datasets, collectively called n-MNIST. Then, we propose a novel framework for the classification of handwritten digits that learns sparse representations using probabilistic quadtrees and Deep Belief Nets. On the MNIST and n-MNIST datasets, our framework shows promising results and significantly outperforms traditional Deep Belief Networks.

Progressively Growing Generative Adversarial Networks for High Resolution Semantic Segmentation of Satellite Images

“…Discriminative features used to evaluate candidate rooftops include building shadows, geometry and spectral characteristics [10], [11]. Several approaches have used LIDAR alone or in addition to multispectral images [12]- [14] Newer-generation machine learning techniques [15] have also been applied in satellite image classification [16] and in rooftop segmentation specifically [17]- [19]. Convolutional neural networks (CNNs) have greatly improved the state of the art in semantic segmentation tasks wherein each pixel in an image is associated with a class label [20], [21].…”

Pixel-Level Reconstruction and Classification for Noisy Handwritten Bangla Characters

“…Quadtrees have been used previously to compress images [15] and represent spatial data [16]. In [4], the authors use probabilistic quadtrees to represent character images and classify them using a deep belief network (DBN).…”

“…We inject these three types of noises into the Bangla Basic Character dataset consisting of 50 classes resulting in the Noisy Bangla Basic Character Dataset. In [4], once a block has been chosen for decomposition in any one image based on the homogeneity criterion, that block is identically decomposed for every other image. We use a saliency map instead which improves the representation and we also use another DBN for character reconstruction removing noise.…”

“…We build upon recent developments in Deep Learning to develop techniques for efficient representation of sparse features [3], [4], [5]. Sparse representations are usually compact representations of signals or features that have unnecessary default values [6] [7].…”

PCGAN-CHAR: Progressively Trained Classifier Generative Adversarial Networks for Classification of Noisy Handwritten Bangla Characters

“…In this work, we build on recent work in adversarial training [14] to improve on the state-of-the-art in representing sparse features [3,14,23]. We define sparse representations as noisy, generally compact, representations of signals [12] [7].…”