Hyperspectral Data Clustering Using Hellinger Divergence

Clustering is an important task in hyperspectral image processing. Despite the existence of a large number of clustering algorithms, little attention has been paid to the use of non-Euclidean dissimilarity measures in the clustering of hyperspectral data. This paper proposes a clustering technique based on the Hellinger divergence as a dissimilarity measure. The proposed technique uses Lloyd’s ideas of the k-means algorithm and gradient descent-based procedure to update clusters centroids. The proposed technique is compared with an alternative fast k-medoid algorithm implemented using the same metric from the viewpoint of clustering error and runtime. Experiments carried out using an open hyperspectral scene have shown the advantages of the proposed technique.

Utilizing Amari-Alpha Divergence to Stabilize the Training of Generative Adversarial Networks

Generative Adversarial Nets (GANs) are one of the most popular architectures for image generation, which has achieved significant progress in generating high-resolution, diverse image samples. The normal GANs are supposed to minimize the Kullback–Leibler divergence between distributions of natural and generated images. In this paper, we propose the Alpha-divergence Generative Adversarial Net (Alpha-GAN) which adopts the alpha divergence as the minimization objective function of generators. The alpha divergence can be regarded as a generalization of the Kullback–Leibler divergence, Pearson χ 2 divergence, Hellinger divergence, etc. Our Alpha-GAN employs the power function as the form of adversarial loss for the discriminator with two-order indexes. These hyper-parameters make our model more flexible to trade off between the generated and target distributions. We further give a theoretical analysis of how to select these hyper-parameters to balance the training stability and the quality of generated images. Extensive experiments of Alpha-GAN are performed on SVHN and CelebA datasets, and evaluation results show the stability of Alpha-GAN. The generated samples are also competitive compared with the state-of-the-art approaches.

Enhancing resolution of natural methylome reprogramming behavior in plants

BackgroundNatural methylome reprogramming within chromatin involves changes in local energy landscapes that are subject to thermodynamic principles. Signal detection permits the discrimination of methylation signal from dynamic background noise that is induced by thermal fluctuation. Current genome-wide methylation analysis methods do not incorporate biophysical properties of DNA, and focus largely on DNA methylation density changes, which limits resolution of natural, more subtle methylome behavior in relation to gene activity.ResultsWe present here a novel methylome analysis procedure, Methyl-IT, based on information thermodynamics and signal detection. Methylation analysis involves a signal detection step, and the method was designed to discriminate methylation regulatory signal from background variation. Comparisons with commonly used programs and two publicly available methylome datasets, involving stages of seed development and drought stress effects, were implemented. Information divergence between methylation levels from different groups, measured in terms of Hellinger divergence, provides discrimination power between control and treatment samples. Differentially informative methylation positions (DIMPs) achieved higher sensitivity and accuracy than standard differentially methylated positions (DMPs) identified by other methods. Differentially methylated genes (DMG) that are based on DIMPs were significantly enriched in biologically meaningful networks.ConclusionsMethyl-IT analysis enhanced resolution of natural methylome reprogramming behavior to reveal network-associated responses, offering resolution of gene pathway influences not attainable with previous methods.