Ronan Guivarch

On a strategy for spectral clustering with parallel computation

Spectral Clustering is one of the most important method based on space dimension reduction used in Pattern Recognition. This method consists in selecting dominant eigenvectors of a matrix called affinity matrix in order to define a low-dimensional data space in which data points are easy to cluster. By exploiting properties of Spectral Clustering, we propose a method where we apply independently the algorithm on particular subdomains and gather the results to determine a global partition. Additionally, with a criterion for determining the number of clusters, the domain decomposition strategy for parallel spectral clustering is robust and efficient.

On the Easy Use of Scientific Computing Services for Large Scale Linear Algebra and Parallel Decision Making with the P-Grade Portal

Scientific research is becoming increasingly dependent on the large-scale analysis of data using distributed computing infrastructures (Grid, cloud, GPU, etc.). Scientific computing (Petitet et al. 1999) aims at constructing mathematical models and numerical solution techniques for solving problems arising in science and engineering. In this paper, we describe the services of an integrated portal based on the P-Grade (Parallel Grid Run-time and Application Development Environment) portal (http://www.p-grade.hu) that enables the solution of large-scale linear systems of equations using direct solvers, makes easier the use of parallel block iterative algorithm and provides an interface for parallel decision making algorithms. The ultimate goal is to develop a single sign on integrated multi-service environment providing an easy access to different kind of mathematical calculations and algorithms to be performed on hybrid distributed computing infrastructures combining the benefits of large clusters, Grid or cloud, when needed.

Parallel Spectral Clustering for the Segmentation of cDNA Microarray Images

Microarray technology generates large amounts of expression level of genes to be analyzed simultaneously. This analysis implies microarray image segmentation to extract the quantitative information from spots. Spectral clustering is one of the most relevant unsupervised method able to gather data without a priori information on shapes or locality. We propose and test on microarray images a parallel strategy for the Spectral Clustering method based on domain decomposition and with a criterion to determine the number of clusters.

An hybrid approach for the parallelization of a block iterative algorithm

The Cimmino method is a row projection method in which the original linear system is divided into subsystems. At every iteration, it computes one projection per subsystem and uses these projections to construct an approximation to the solution of the linear system. n nThe usual parallelization strategy in block algorithms is to distribute the different blocks on the available processors. In this paper, we follow another approach where we do not perform explicitly this block distribution to processors within the code, but let the multi-frontal sparse solver MUMPS handle the data distribution and parallelism. The data coming from the subsystems defined by the block partition in the Block Cimmino method are gathered in an unique block diagonal sparse matrix which is analysed, distributed and factorized in parallel by MUMPS. Our target is to define a methodology for parallelism based only on the functionalities provided by general sparse solver libraries and how efficient this way of doing can be.

Parallelisation of MACOPA, A Multi-physics Asynchronous Solver

Macopa is a partial differential equations solver based on a particular local time-stepping technique dedicated to multi-physics and multi-scale problems. Here, some parallelisation strategies – multi-threading, domain decomposition, and hybrid OpenMP/MPI – are introduced for this solver. Their efficiency is evaluated on a few examples.

Sparsification on Parallel Spectral Clustering

Spectral clustering is one of the most relevant unsupervised method able to gather data without a priori information on shapes or locality. A parallel strategy based on domain decomposition with overlapping interface is reminded. By investigating sparsification techniques and introducing sparse structures, this parallel method is adapted to treat very large data set in fields of Pattern Recognition and Image Segmentation.

Segmentation of cDNA microarray images using parallel spectral clustering

Microarray technology generates large amounts of expression level of genes to be analyzed simultaneously. This analysis implies microarray image segmentation to extract the quantitative information from spots. Spectral clustering is one of the most relevant unsupervised methods able to gather data without a priori information on shapes or locality. We propose and test on microarray images a parallel strategy for the Spectral Clustering method based on domain decomposition with a criterion to determine the number of clusters.

Resource Management Architecture for Fair Scheduling of Optional Computations

Most High-Performance Computing platforms require users to submit a pre-determined number of computation requests (also called jobs). Unfortunately, this is cumbersome when some of the computations are optional, i.e., they are not critical, but their completion would improve results. For example, given a deadline, the number of requests to submit for a Monte Carlo experiment is difficult to choose. The more requests are completed, the better the results are, however, submitting too many might overload the platform. Conversely, submitting too few requests may leave resources unused and misses an opportunity to improve the results. This paper introduces and solves the problem of scheduling optional computations. It proposes a generic client-server architecture and an implementation in a production GridRPC middleware, which auto-tunes the number of requests. Real-life experiments show that several metrics are improved, such as user satisfaction, fairness and the number of completed requests. Moreover, the solution is shown to be scalable.