Page not found

Big Data, Model Combining, and Predictive Modeling of High Dimensional Data

<h3 class="red">Big Data, Model Combining, and Predictive Modeling of High Dimensional Data</h3><p>Big Data and the predictive modeling of high-dimensional datasets are of great interest to practitioners in many fields, such as finance, biology, and economics. These researchers are taking a methodology, model combination, that is widely and efficiently used for low-dimensional datasets and adapting it for high-dimensional situations. The project will develop a general risk bound for the group&#39;s methodology for high-dimensional predictive modeling, especially classification problems. Further, an efficient computing algorithm for the combination schemes will be developed and wrapped into a publicly available R package.</p><p>Many Big-Data sets (real data) will be analyzed by multiple high-dimensional classification methods using cross-validation. It will take about 10 million non-linear numerical optimizations for process. Besides working with real data, the researchers will perform various numerical experiments in order to have a better understanding of their methods. For different scenarios, they will compare their methods with between five and ten other popular methods and run large number of replicates to reduce the bias from the samplings. This will take about 10 million calculations.</p><p>Return to this PI&#39;s <a href="">main page</a>.</p>
Group name: 

Carbon Dioxide Levels Affect Grasses Differently Than Expected

MSI PI Peter Reich (professor, Forest Resources ; Fellow, Institute on the Environment ) was the lead researcher in a study, published this month in the prestigious journal Science , that showed unexpected growth performance of some types of grasses as carbon dioxide levels rise. This 20-year study...

High performance optimization for geographic optimization and feature detection

<h4>High Performance Optimization for Geographic Optimization and Feature Detection</h4><p><span style="color: rgb(51, 51, 51); font-size: 14px; background-color: rgb(255, 255, 255);">The purpose of this project is to design high-performance optimization algorithms for large-scale geographic and medical data with up to half a million feature vectors. These algorithms will make extensive use of parallel structures.</span></p>
Group name: 

Parallel Simulations for Computer Architecture and Computer Aided Design

<h3 class="red">Parallel Simulations for Computer Architecture and Computer Aided Design</h3><p>For several emerging applications such as wearables, internet of things, and sensor networks, energy efficiency is of utmost importance. While custom ASICs have higher energy efficiency, general-purpose embedded processors are the preferred solution for many such applications due to the evolving nature of these applications and the high costs of custom IC design. The Sartori group discovers and exploits new opportunities for improving energy efficiency in general purpose embedded processors. They are currently focusing on new opportunities for energy efficiency enabled by detailed co-analysis of the design-level description of a processor and an application binary. Traditionally, co-analysis of the low-level hardware and details for a system has not been performed due to prohibitive costs. However, this group has developed anutomated analysis tools that perform unique analyses and expose new opportunities for energy efficiency. A few of their ongoing projects in this area are described below.</p><ul><li>The group has created a tool that identifies the parts of a processor that can never be exercised by a particular application. As such, they can identify paths in a processor that can never be exercised for a particular workload. If the most critical paths in a processor are not exercised, then extra timing slack exists that can be exploited to reduce power or increase performance.</li><li>Knowing the parts of a processor that can never be exercised by an application or application phase also allows new opportunities for aggressive power gating. The researchers are developing techniques that allow the benefits of aggressive hardware-based power gating with costs similar to those of software power gating. Their techniques can provide guarantees that power gating decisions are safe without requiring hardware checking mechanisms and provide near-optimal power savings, compared to oracular control decisions.</li><li>Detailed activity analysis and guarantees for a hardware-software system can also allow researchers to more accurately and aggressively bound the peak power requirements of the system. The gap between conventional peak power rating and application-aware peak power rating can be exploited for reduced energy and area, improved performance and throughput, and greater efficiency.</li></ul><p>The new techniques this group is creating require detailed analysis of a system&#39;s hardware and software. This detailed analysis relies on high-throughput parallel simulation methodologies to be performed in a reasonable amount of time. As such, it relies on high-performance parallel computing resources.</p><p>Return to this PI&#39;s <a href="">main page</a>.</p><p>&nbsp;</p>
Group name: 

MSI Tours

Tour Information MSI has a limited capability for tours for prospective students and faculty, University student groups, and secondary-school groups. Please contact with questions about possible tours. Facilities MSI Computer Machine Room -- MSI enables interdisciplinary research...

MSI Facilities Overview (Short)

Minnesota Supercomputing Institute (MSI) is the University of Minnesota's principle center for computational research. MSI also operates several data centers on campus. Its main data center is located in the basement of Walter Library (room B40) on the University Twin Cities campus. It has an IT raised floor surface of approximately 3700 sq.ft. and slightly over 1 MW of available power. The Institute operates two main supercomputing systems:

  • Mesabi, an HP heterogeneous Linux cluster with over 18,000 Haswell compute cores
  • Itasca, an HP Linux cluster with 1,134 HP ProLiant blade servers

In addition to the supercomputing systems, MSI supports interfaces and systems for advanced scientific visualization and interactive computing. MSI manages three large storage systems: a high performance parallel files system (6 PB), a CEPH/S3 tier 2 object storage system, and a 400 core, 400 TB Hadoop system. MSI hosts a SpectraLogic T950 tape library with expansion capabilities for over 20 TB of online storage, which is used to backup high value files. The data center is connected to the 100 Gbps campus research network via multiple 40 GbE connections. The University maintains 100 Gbps connections to our regional optical network, which in turn is connected to Internet2 and beyond. MSI provides the infrastructure and expertise to the greater Minnesota University system. In addition to the diverse systems, more than half of the MSI staff are available to provide expert consulting in areas such as research informatics, software development, and algorithm optimization.

Modeling and Simulation of Turbulent, Reacting, Multiphase Flows

<h3 class="red">Modeling and Simulation of Turbulent, Reacting, Multiphase Flows</h3><p>The Garrick group is&nbsp;interested in the modeling and simulation of multiphase reacting flows. These include particle formation and growth dynamics in laminar and turbulent flow systems, combustion problems, and spray dynamics. This research draws on fluid dynamics, computational fluid dynamics, aerosol dynamics, chemistry, and physics to develop computational tools to simulate atomization, particle formation, coagulation, coalescence, aggregation, break-up, and other physico-chemical processes. The underlying processes and dynamics are modeled in a fashion that render them amenable to simulation via high-performance computing. The group utilizes a variety of simulation techniques including DNS and LES to perform both scientific and engineering simulation and analysis.</p><p>Return to this PI&#39;s <a href="">main page</a>.</p>
Group name: 

Discovery of Compounds That Inhibit Select RNA Viruses

<h3 class="red">Discovery of Compounds That Inhibit Select RNA Viruses</h3><p>These researchers&#39; main interests are identifying anti-viral drug targets for major human pathogens and evaluating small molecules for their ability to inhibit those targets and virus replication. They employ four main approaches to uncover small molecules that inhibit the replication and pathogenesis of viruses:</p><ul><li>Conduct phenotypic screens to evaluate anti-viral activity of small molecules.</li><li>Collaborate with medicinal chemists, computational chemists, and structural biologists to rationally design small molecule inhibitors of established and novel viral drug targets.</li><li>Design and implement high throughput screening to identify small molecules that inhibit known and novel viral drug targets.</li><li>Understand mechanism of effect of small molecule inhibitors.</li></ul><p>Passage of virus in the presence of inhibitors will generate resistant virus and the researchers can sequence viral genomes to understand which viral gene products are important for inhibition and aspects of those proteins that interact with inibitors. They use MSI resources to help analyze large amounts of nucleic acid sequence data to understand how inhibitors function. In addition, the researchers are developing reagents to understand replication of RNA viruses such as coronaviruses and to screen for coronavirus inhibitors. They have used MSI resources previously to analyze the nucleic acid sequence data from a candidate viral repicon and are continuing that analysis.</p><p>Return to this <a href="">PI&#39;s main page</a>.</p>
Group name: 

Fast RF Calculations for High Field MRI

<p><strong>Fast RF Calculations for High Field MRI</strong></p> <p>Radio-frequency electromagnetic analysis of high field MRI coils is an indispensable task for the development of new coil designs, for optimization of existing coils through tuning and matching when they are loaded with human models, and to ensure safety through specific absorption rate (SAR) and temperature calculations. Finite difference time domain (FDTD)-based solvers have been the method of choice so far since they can easily handle the inhomogeneities in the human model and the utilization of low-cost GPU accelerators helped in reducing the simulation times. Nonetheless, these solvers still suffer long simulation times due to the resonant nature of MRI coils. Initial experiments and a recent publication show that frequency-domain hybrid methods, such as method of moments (MOM)-finite element method (FEM), are better suited to MRI experiments in terms of solution time. In addition, because they use surface-conformal tetrahedral meshes instead of cubical voxels, they are free from staircasing errors hence yield superior accuracy compared to FDTD. The cost of the FEM-MOM approach is that it needs more memory and high-performance computing clusters to handle high-resolution human models with low simulation times. MSI resources are therefore necessary for this work.</p>
Group name: 

Open Science Grid User School 2016

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...