Balanced Clustering: A Uniform Model and Fast Algorithm

Clustering is a fundamental research topic in data mining and machine learning. In addition, many specific applications demand that the clusters obtained be balanced. In this paper, we present a balanced clustering model that is to minimize the sum of squared distances to cluster centers, with uniform regularization functions to control the balance degree of the clustering results. To solve the model, we adopt the idea of the k-means method. We show that the k-means assignment step has an equivalent minimum cost flow formulation when the regularization functions are all convex. By using a novel and simple acceleration technique for the k-means and network simplex methods our model can be solved quite efficiently. Experimental results over benchmarks validate the advantage of our algorithm compared to the state-of-the-art balanced clustering algorithms. On most datasets, our algorithm runs more than 100 times faster than previous algorithms with a better solution.

Development and Validation of an X-Ray Tomograph for Two-Phase Flows

This paper describes the development and validation of a high spatial resolution X-ray tomograph devoted to investigate air-water two-phase flows. The device hardware is mainly composed of a 60 keV X-ray source, a detector and an accurate mechanical bench. This paper concentrates on accuracy quantification and emphasis is given on the reconstruction procedure. It is well known that absorption gradients induce reconstruction artifacts when using standard algorithms based on uniform regularization. In the particular case of two-phase flows in pipes, this leads to a poor measurement accuracy in the vicinity of the walls. To overcome such effects, improved algorithms have been developed in this study, involving different spatially adaptative regularization methods. A first calibration performed on static phantoms clearly exhibits the benefit brought by such advanced reconstruction algorithms. A validation procedure has been carried out on an air-water bubble column, equipped with an optical probe which can be translated in order to explore the 80 mm × 80 mm square cross section. Comparisons of local void fraction measurements have been performed pixel by pixel, and demonstrate the accuracy improvement induced by the advanced reconstruction algorithms.