The Open Science Grid (OSG) User School is soliciting applications for students to attend the 2016 session, July 25-29, 2016. This school is intended for researchers who want to learn how they could use high-performance computing in their work. The school takes place on the campus of the University...
posted on January 30, 2014 The Blue Waters program is sponsoring 20 undergraduate internships for 2014-15. Interns will participate in petascale computing research and development projects. The research should begin in Summer 2014 and continue through the school year. Interns will work with a...
The MSI Undergraduate Internship Program wrapped up its 20th summer last month. Eleven undergraduates from colleges and universities around the country worked on projects in chemistry, physics, geophysics, medicinal chemistry, biomedical engineering, astronomy, biochemistry, and chemical...
The MSI Undergraduate Internship Program wrapped up its 19th summer last month. Twelve undergraduates from colleges and universities around the country worked on projects in chemistry, physics, geophysics, astronomy, biochemistry, medicinal chemistry, and drug design. In this annual program,...
One of the most exciting areas that researchers use MSI for is computer-generated visualizations. Avery Musbach, a graduate student in the Department of Computer Science and Engineering , is the lead author on a paper that demonstrates the power of scientific computing to create visualizations. The...
<h3 class="red">Genomewide Predictions of Maize Performance</h3><p><span style="background-color: rgb(255, 255, 255);">Genomewide predictions allow the evaluation of maize lines or hybrids without field testing (phenotyping) of the lines or hybrids themselves. In particular, genomewide marker effects are estimated from a training population that has been previously genotyped and phenotyped. These marker effects are then used to assess the performance of new lines or hybrids that have been genotyped, but not yet phenotyped. Genomewide predictions therefore leverage the lower costs of genotyping (about $15 per line or hybrid) than of phenotyping (about $120 per line or hybrid). Since 2011, the Bernardo research group has been given access by Monsanto to about $25 million worth of phenotypic and marker datasets from its own maize breeding program. These datasets have allowed the group to investigate ways to optimize genomewide predictions in maize breeding. Research efforts are now focusing not only on predicting performance averaged across environments, but also in specific environments. The scale of the datasets (969 populations, >4 million phenotypic data points, > 11 million marker data points, and environmental covariables on 432 year-location combinations) has necessitated the use of high-performance computing resources</span>.</p><p>Return to this PI's <a href="https://www.msi.umn.edu/pi/98e6ab632f9f76148151360678ed6cb6/10219">main page</a>.</p>
<h3 class="red">Scalable Ring-Shaped Hotspot Detection</h3><p>Given a collection of geo-located activities (e.g., crime reports), ring-shaped hotspot detection (RHD) finds rings where the concentration of activities inside the ring is much higher than outside. RHD is important for the applications such as crime analysis, where it may focus the search for crime source’s location, e.g. the home of a serial criminal. RHD is challenging because of the large number of candidate rings and the high computational cost of the statistical significance test.</p><p>This group's previous work considereda Dual Grid Based Pruning (DGP) algorithm to detect ring-shaped hotspots. Based on theoretical proofs and experimental evaluation, the DGP algorithm improves the computational cost of the naïve approach substantially. However, the DGP algorithm is still computationally intensive due to the extensive ring enumeration using a four dimensional parameter space (i.e. four parameters of a ring) and the requirement for a significance test using Monte Carlo simulations which increase its computational cost. Moreover, sequential procedures that are used for ring enumeration in parameter space do not scale up to large number of grid cells due to the four dimensional nature of the parameter grid. Therefore the researchers plan to improve the computational efficiency of the DGP algorithm using an enhanced refine phase (by applying the Best Enclosing Ring algorithm) and an enhanced pruning phase (using multi-cell-size pruning and Local Maxima Enumeration).</p><p>The researchers are using MSI for two tasks:</p><ul><li>To perform an extensive experimental evaluation for the DGP algorithm, which can help in executing the algorithm for a large number of grid cells and a large number of Monte Carlo simulations in a reasonable amount of time. The researchers will extensively evaluate the performance of the DGP algorithm as compared to the naive approach. Experiments will evaluate the effect of varying multiple parameters (i.e. cell size, number of activities, likelihood ratio threshold, number of Monte Carlo simulations) on the algorithm performance.</li><li>To develop a parallel formulation of the DGP algorithm using a parallel computing platform (e.g. Hadoop, Spark, GPU, etc). The researchers will parallelize the DGP algorithm over a set of machines using a parallel computing platform (e.g. Hadoop, Spark, GPU, etc) to meet the high performance requirements imposed. They will both use task partitioning and data partitioning among the machines and the goal of the parallelization will be to improve the scalability as well as reduce the execution time of DGP. In data partitioning schemes, the four dimensional parametric grid will be partitioned among different machines and each parametric grid cell will be independently processed. In the task partitioning schemes, the significance test with Monte Carlo simulations will be given to different machines so that each machine completes a number of simulations.</li></ul><p>Return to this PI's <a href="https://www.msi.umn.edu/pi/ddda1f906e2114d5048233b4594cee5c/54249">main page</a>.</p>
<h3 class="red">Metal Cluster Structures and Reactivities: Computational Elucidation of Anion Photoelectron Spectra</h3><p>Research in this group focuses on the chemical bonding between transition metal atoms in ligand-free diatomics and clusters, and their reactivities with small gas phase molecules. Experiments in the laboratory employ anion photoelectron spectroscopy, flow tube ion-molecule chemistry, and mass spectrometry to study these anions and the corresponding neutral species. To help assign the photoelectron spectra, they compare the experimental results with those calculated using density functional methods, which are used to predict the equilibrium geometries, vibrational frequencies, electronic state energies, and spin multiplicities of the anionic and neutral systems. By simulating the Franck-Condon photodetachment spectra based upon the DFT results and comparing those predictions for possible electronic states and (for polyatomics) different isomers directly to the experimental spectra, the researchers are often able to deduce convincing assignments for the observed species, even if they have never been studied before either spectroscopically or computationally. The group's spectroscopic studies can also provide useful benchmarks to aid in the further development, by other researchers, of improved theoretical methods with which to treat these small but computationally challenging transition metal clusters and partially-ligated organometallics.</p><p>Current work includes studies of bare diatomics incorporating transition metals from Groups 5 and/or 6, which can exhibit very high-order multiple bonding. For example, the researchers are investigating the Group 6, third transition series homonuclear diatomic W<sub>2</sub> (tungsten dimer), which has a formal bond order of 6, as well as the anions of bimetallic dimers incorporating metals from both Groups 5 and 6, such as NbCr<sup>-</sup>, NbMo<sup>-</sup> and NbW<sup>-</sup>, which can also exhibit sextuple bonds. They are also studying (or plan to study next year) organometallic complexes produced upon reaction of transition metal atoms with simple hydrocarbons (such as methane, ethylene or butadiene) or with CO<sub>2</sub>. These results can contribute to the understanding of the relationships between the chemical reactivities of various transition metals and the configurations and spin multiplicities of their ground and low-lying electronic states. In a broader context, these results can contribute to the development of an improved understanding of catalytic processes mediated by transition metal systems.</p><p>Return to this PI's <a href="https://www.msi.umn.edu/pi/b2973ba8bf6971da78fcc75845368b84/31200">main page</a>.</p>
<h3 class="red">Impact of Alcohol and Drug Use on the Development of Neural Connections During Adolescence and Young Adulthood </h3><p>The primary aims of this ongoing longitudinal study are to conduct a comprehensive investigation of brain development during adolescence and early adulthood and to determine how brain development is altered when individuals begin to use alcohol (as well as other drugs, such as marijuana) during this period. The researchers employ an extensive two-day data collection protocol at each study time point, consisting of behavioral assessments (interviews, questionnaires, computerized testing), brain magnetic resonance imaging (MRI; high resolution anatomical scans, several types of diffusion scans, spectroscopy, resting functional scans), and electroencephalography (EEG). They also have a one-time collection of genetic data (single-nucleotide polymorphisms). Data collection waves occur at two-year intervals and currently the researchers are completing their fifth assessment. In their analyses within and across these types of data, the researchers investigate the refinement of brain network connectivity during normal adolescent development and identify alterations due to alcohol and drug use. MSI resources are heavily used to achieve this “connectivity” aspect of this brain-behavior research, which relates directly to the goals of the Human Connectome Project. For example, using high-resolution anatomical MRI scans, the researchers extract complete representations of the cortical surface in both brain hemispheres; using diffusion MRI scans they compute measures of the microstructural organization of neural fibers that connect brain regions, and then conduct a “virtual dissection” of these fibers using probabilistic tractography; using resting functional MRI scans they measure neurophysiological activity across multiple overlapping brain networks; using EEG recordings they identify the coordinated synchronization of electrophysiological activity within brain networks in response to external stimuli; and so on. MSI resources are used in all of these analyses, for both data preprocessing and permutation-based statistics. </p><p>Return to this PI's <a href="https://www.msi.umn.edu/pi/394e59f70b9225cca8c6a42339669d38/34374">main page</a>.</p>
Python is a modern general purpose programming language that is popular in scientific computing for its readable syntax and extremely rich ecosystem of scientific and mathematical modules.