Abstract:

The rapid improvement of phenotyping capability, accuracy and throughput have greatly increased the volume and diversity of phenomics data. We present Temporal Emerging Phenomenon Finder (TEP-Finder), aiming to improve our understanding of the quantitative variation of complex phenotypes and to attribute gene functions. An emerging phenomenon is a group of subjects who exhibit a coherent phenotype pattern during a relatively short time. Emerging phenomena are highly transient and diverse, and are dependent in complex ways on both environmental conditions and development. Identifying emerging phenomena will help biologists examine potential relationships among phenotypes and genotypes in a genetically diverse population and associate such relationships with the change of environments or development.

Abstract:

It is common knowledge that drugs have polypharmacological properties that can be explored for new insights in drug discovery. That is a given drug interacts with many different proteins and a given protein interacts with multiple drugs. The polypharmacological, promiscuous nature of pharmaceuticals can have both beneficial and detrimental consequences. This attribute can be exploited to improve drug efficacy and prevent drug resistance. In addition to the ability of chemical compounds to interact with an array of protein targets, many diseases have multiple genetic determinants, and individual genetic determinants may be involved in multiple diseases. Furthermore, protein function and expression are controlled by a regulatory network of other proteins. When targeted therapies work initially patients often develop resistance due to secondary mutations or compensation from other parts of the underlying biological network. This illustrates the potential benefits of establishing computational polypharmacology methods – discovering drugs that intentionally target multiple proteins for a beneficial therapeutic result. Many adverse drug reactions (ADRs) result from drugs interacting with non-therapeutic off-targets (unintended interactions). Animal studies during preclinical trial are not always a good indication of these adverse interactions in humans, and such adverse effects are generally not discovered until a drug has reached clinical trial or is already on the market. With the number of different proteins in humans and the genetic variations observable in the population, a full understanding of all possible interactions through experiments and clinical testing alone is not feasible, making computational investigations particularly useful and relevant.

In our digitalized, data-driven world, there is a wealth of knowledge available that is beyond the processing power of an individual researcher or even team of researchers. The abundance of available biomedical data combined with the massive computing power we have available today with leadership class supercomputers provides great opportunities to advance computational drug research. A tool that reliably predicts protein and drug binding would revolutionize the pharmaceutical industry. An accurate representation of polypharmacological networks would provide a wealth of knowledge and insights on drug repurposing, side-effect prediction, and drug efficacy. This would lead the way to personalized polypharmacological networks including individual’s genetic variations resulting in a breakthrough for precision medicine. However, there are still many obstacles to overcome when it comes to utilizing massive computational power and ensuring accuracy of our predictions.

Using machine learning to score potential drug candidates may offer an advantage over traditional imprecise scoring functions because the parameters and model structure can be learned from the data. However, models may lack interpretability, are often overfit to the data, and are not generalizable to drug targets and chemotypes not in the training data. Benchmark datasets are prone to artificial enrichment and analogue bias due to the overrepresentation of certain scaffolds in experimentally determined active sets. Datasets can be evaluated using spatial statistics to quantify the dataset topology and better understand potential biases. Dataset clumping comprises a combination of self-similarity of actives and separation from decoys in chemical space and is associated with overoptimistic virtual screening results. This talk explores data, methods, and potential data biases relevant to computational drug binding predictions.

Bio:

Dr. Sally Ellingson is a computational scientist working at the intersection of computational biology and high-performance computing. She has undergraduate degrees in computer science and mathematics from Florida Institute of Technology. She obtained her doctoral degree at the University of Tennessee and Oak Ridge National Laboratory under a fellowship funded by the National Science Foundation in computational biology. She is an Assistant Professor in the Division of Biomedical Informatics in the College of Medicine. Her primary appointment is with the Markey Cancer Center’s Cancer Research Informatics Shared Resource Facility. In this role, she facilitates high throughput genomics and big data processing for cancer research and precision medicine. Dr. Ellingson was awarded as a KL2 scholar by the University of Kentucky’s Center for Clinical and Translational Sciences and is a Research Affiliate at Lawrence Berkeley National Laboratory. She has taken University of Kentucky students with her to further develop her research at Lawrence Berkeley National Laboratory for two past summers. This summer she has expanded her national lab collaborations as a visiting faculty scholar at Lawrence Livermore National Lab.

Dr. Ellingson also engages in mentoring and outreach, especially for underrepresented groups in computational sciences. Since graduation she has been actively engaged in the organization of the student programs at Supercomputing, the largest conference in high-performance computing. She organized the student volunteer program last year and will organize the conference wide mentoring program this year, and keeps the diversity of program participants a priority.

Abstract:

We are living in a time of rapid data growth from an ever increasing number and diversity of sources. In a recent Cisco report it is estimated that there will be 28 billion connected devices by 2022 generating 43 trillion gigabytes of electronic data annually. In the medical world, connected devices range from distributions of low-power sensors and personal fitness devices, to large data generators such as medical imaging and genomic sequencing devices. While the scale and diversity of computational and data exchange challenges has increased rapidly, data acquisition, processing techniques and, supporting infrastructures have not maintained pace. Data management in highly regulated areas such as clinical diagnostics and medical research is especially impacted through stringent and often antiquated governance requirements. It is common practice to transfer in an unaltered form raw data acquired from its source of generation to remote locations, such as a laptops, servers, and computational clusters for processing and storage. In this centralized (even clusters) computational paradigm the majority of applications are developed within the context of the platform, with operating assumptions such as direct binary-level access to data storage or computational accelerators. As a byproduct of what some refer to as data gravity, data and associated applications tend to stay within the system they are stored or generated. Given the increased diversity of computational needs, the availability of practically limitless resources in pubic clouds, and the need to manage computational pipelines from data generation, through processing, and resulting, new methods must be developed and employed. We argue that computational inference, filtering, aggregation, and other function must be able to take place at any point in the computational pipeline, including near sources of data generation. Such capabilities impact not only operational efficiencies, but are also relevant to the enforcement of data governance and the preservation of privacy.

We present the use of agent-based methodologies in support of three biomedical pipeline applications within the domains of genomic processing, medical image management, and clinical laboratory analysis. In this paradigm computational units represent nodes and data exchange is represented as edges, forming a graph or computational pipeline. We find it convenient to represent these distributed pipelines a graphs, allowing the use of a number of computational models for resource placement, scheduling, and policy enforcement. We show how various functions within a genomic processing pipeline can be acquired from sequencers, distributed to public clouds, processing routed across a diverse set of resources, and results reported in a protected environment. Likewise, we show how digital images can be acquired and de-identified at (protected) sources of generation, submitted to remote vendor services for AI interpretation, and results merged with clinical systems. Finally, we will show methods for capture, enrichment, and processing of data from a mass spectrometer devices, with results routing to laboratory information systems and raw data transferred and aggregated to a repository for use in machine learning.

Bio:

Dr. Bumgardner spent the first 20 years of his career IT formerly serving as Director of Enterprise Systems and Development, Chief Technology Architect, and Director of Research Computing. Currently, he serves as Assistant Professor and Chief of the Division of Pathology Informatics in the Department of Pathology and Laboratory Medicine at the University of Kentucky.

Abstract:

Will discuss strategies for tackling big genomes and lessons learned from the Axolotl, which has a genome that is 10X larger than the human genome, including new large-scale variants and the characterization of mutant genes that have eluded characterization for decades.

Abstract:

Hand labeling large datasets to create training examples for supervised machine learning maybe impractical or cost prohibitive in some scenarios. In this talk, neural transfer learning and domain adaptation applications are discussed to augment existing, potentially smaller, datasets for a (target) task with datasets previously created for related (source) tasks. The end goal is to create better models for text mining applications.

Abstract:

This talk will provide a computational fluid dynamics perspective on how high-performance computing is paving the way for the development of new generation engineering designs. Current and future advancements in aerospace technologies are heavily relying on the use of state-of-the-art computational tools and these leaps in technology are not possible without the use large computing infrastructure. While it is well-recognized that the use of high-fidelity simulation-based design substantially lowers the development cost by reducing the number of extensive hardware tests its potentially even greater benefit is to provide a test bed for real world conditions which are impractical to test in ground facilities. This talk will present some examples where high-fidelity computational fluid dynamics was employed for NASA and DoD applications to support the development of novel aerospace engineering designs.

Abstract:

Many materials express structural randomness at a sufficiently small lengthscales. Increasing interest in nanomaterials and materials by design is driving a need for modeling approaches able to account for randomness—something with which computers struggle. We have constructed a toolkit capable of generating sets of random structures derived from geometric ‘seed’ descriptions that allows the high-throughput calculation of distributions of materials properties. This toolkit will allow the computational determination of effective or expected properties, ranges and probabilities of observed properties as a function of length scale, determination of sensitivities to geometric features, and prototyping of designed structures. As demonstration of these capabilities, this talk will focus on new insights into the structure-property relationships governing the mechanical properties nanoporous materials culminating in a proposed revision to the widely applied Gibson-Ashby model.

Abstract:

Simulating quantum dynamics on classical computers is a difficult task which requires exponential computational resources. I will review several particular strategies leveraging current HPC infrastructure which can be used to improve efficiency of quantum mechanical simulations.

Abstract:

One of the major challenges in space exploration is developing Thermal Protection Systems (TPS) that allows vehicles to handle the extreme conditions they encounter. TPS typically fall within two categories: ablative and non-ablative. Non-ablative TPS are often used for reusable or multi-use vehicles. These vehicles include missile tips, hypersonic jets or the thermal title blankets, nose caps and leading edges. Ablative TPS is often used when the vehicle is classified as single-use. These materials are designed to turn kinetic energy into chemical energy through several types of reactions. These reactions cause the material to recess.

Read more ...

Abstract:

The accurate prediction of gas­material interactions is crucial to improving the reliability of designing materials that address critical needs in space exploration, national defense, and energy. Gas­material interactions at the atomistic and microscale scale influences the macroscale behavior. In my talk, I will discuss the multiphysics computational capabilities available in my lab to investigate systems operating under diverse length and time scales, and methodologies employed to build a macroscale understanding of gas­material interactions.

Abstract:

Company financial reports are about 80% words and 20% numbers. But most research examines only numbers in understanding how companies “speak” to markets. Recent, innovative research that examines the words in financial reports does so using: (1) simple, isolated counts of word lists (e.g., positive or negative sentiment), i.e., the “bags of words” approach of analyzing dictionary lists of words in financial reports, and, (2) traditional linear models (e.g., regression). We use emerging AI and machine learning methods to analyze the text content of financial reports to explore the ability of these disclosures to predict company risk, earnings, and returns. The results help articulate how the deep analysis of language “speaks” to markets, and the effects of these communications on market reactions, i.e., to learn what the market “listens to” in company’s text disclosures.

Abstract:

The electronic and optical characteristics of organic semiconductors are critically dependent on local and long range molecular packings throughout the materials. While chemists have at their disposal the vast tool chest of organic synthetic chemistry to design π-conjugated molecules and polymers with precise redox and optical characteristics, the development of materials from these building blocks remains Edisonian, as there is limited knowledge among the intimate relationships that connect chemical composition and molecular architecture, materials processing, and the solid-state packing arrangements that define the underlying physicochemical processes that determine material performance. Here we will discuss challenges associated with finding connections among these characteristics, and how computational and theoretical approaches can aid to find solutions to these questions at different stages of the design-development-deployment pipeline.

Abstract:

This paper examines media coverage of the US Supreme Court in an increasingly diverse and fragmented media landscape. We beta-test a web application that allows for the automated collection of Supreme Court news coverage from a wide variety of mainstream and non-traditional news sources online, in real time. We catalogue important features of this coverage, finding coverage differs systematically in new media sources, portraying the Court in a positive tone, yet more ideologically polarized and less complex. We then use a structural topic model to identify how new media differ systematically in their portrayal of the Court.

Abstract:

Over the last four years Facilities Information Services has collected technology and support data about the devices they manage to make better data driven decisions. This presentation will discuss the behind the scenes processes, classifications, standardizations and decision-making algorithms created that have become the base for a continuously evolving technology lifecycle management system. A system that is currently revealing being proactive with technology lifecycles, replacing technology on a data driven schedule, is more cost effective than a reactive approach of replacing devices when failing or when purchasing budgets are available.

Abstract:

I will describe an emerging “AI for Science” initiative at Argonne National Laboratory to advance the concept of Artificial Intelligence (AI) aimed at addressing challenge problems in science. We aim in this initiative to: (1 identify those scientific problems where existing AI and machine learning methods can have an immediate impact (and organize teams and efforts to realize that impact); (2) identify areas of where new AI methods are needed to meet the unique needs of science research (frame the problems, develop test cases, and outline work needed to make progress); and (3) develop the means to automate scientific experiments, observations, and data generation to accelerate the overall scientific enterprise. Science offers plenty of hard problems to motivate and drive AI research, from complex multimodal data analysis to integration of symbolic and data intensive methods, to coupling large-scale simulation and machine learning to drive improved training to control and accelerate simulations. A major sub-theme is the idea of working toward the automation of scientific discovery through integration of machine learning (active learning and reinforcement learning) with simulation and automated high-throughput experimental laboratories. I will provide some examples of projects underway and lay out a set of long-term driver problems.

Bio:

Ian Foster is Senior Scientist and Distinguished Fellow, and also director of the Data Science and Learning Division, at Argonne National Laboratory, and the Arthur Holly Compton Distinguished Service Professor of Computer Science at the University of Chicago. Ian received a BSc (Hons I) degree from the University of Canterbury, New Zealand, and a PhD from Imperial College, United Kingdom, both in computer science. His research deals with distributed, parallel, and data-intensive computing technologies, and innovative applications of those technologies to scientific problems in such domains as materials science, climate change, and biomedicine. His Globus software is widely used in national and international cyberinfrastructures. Foster is a fellow of the American Association for the Advancement of Science, Association for Computing Machinery, and British Computer Society. His awards include the British Computer Society Lovelace Medal, IEEE Kanai award, and Charles Babbage Award.

Abstract:

The American Linguistic Atlas Project (LAP) began in 1929, when Hans Kurath decided to follow the lead of European dialectologists in Germany and France and to create a Linguistic Atlas of American English. Now, the LAP is the largest single survey of regional and social differences in spoken American English. Regional surveys extend from New England to California; some field work was conducted as early as the 1930s, some continues today.

The two earliest surveys are the Linguistic Atlas of New England (LANE) and the Linguistic Atlas of the Middle and South Atlantic States (LAMSAS), for which data was collected by face-to-face interview and transcribed on the spot in fine phonetic notation. These interviews relied on a questionnaire of 104 pages with seven or eight items per page, designed to reveal regional and social differences in everyday vocabulary, grammar, and pronunciation. Over time, the LAP interviews have changed (the questionnaire has been adjusted and more recent interviews have been digitally recorded), but the goal of capturing the regional language use of American speakers has remained. At present, the LAP is comprised of data from over 5000 individual interviews, resulting in a massive amount of linguistic data, all of it connected to specific social and geographic information.

The LAP is currently processing and curating its historical data in an attempt to get it all into a (somewhat) uniform and usable format, thus creating ‘big data’ from historical sources. This talk will give an overview of where we are in that process and will address specifically the ‘big data’ issues of volume, velocity, and variety.

In addition to outlining the challenges and opportunities that the LAP affords, this talk will also address what conceiving of the LAP data as ‘big data’ has already revealed: that the recurring pattern of variation found in such a large amount of data suggests that language functions as a complex system. Characteristic of a complex system is the emergence of scalable, nonlinear patterns and, indeed, at every level of examination, one finds that nonlinear frequency profiles (A-curves) characterize the LAP data. A-curves emerge for entire datasets and also for every subgroup in the data; the order of elements in the profile is likely to differ between groups, and between any group and the overall dataset, and it is these differences in order that effectively create what we refer to as ‘dialects’.

Bio:

As a linguist who specializes in sociolinguistics, my research centers around language variation and change, a subject that I approach from both a large-scale historical perspective using data from the Linguistic Atlas Project and a smaller-scale perspective that focuses on identity-making within individual interactions. My books Language and Material Culture (2015) and Language and Classification (2018) reflect my ongoing interest in the relationship between variation in language and material culture, while the textbook that I co-authored, Exploring Linguistic Science (2018), advocates a view of language as a complex system, with an emphasis on interaction and emergence.

Abstract:

Every day billions of images are uploaded to the Internet and trillions of pixels are gathered by satellites. Together they provide many perspectives of the world, ranging from what someone had for dinner to where new buildings have been constructed. This talk describes my work in extracting geospatial attributes of places from this imagery using deep convolutional neural networks. I will focus on methods that combine overhead and ground-level imagery in novel ways, with applications including image localization, cross-modal hallucination, and fine-grained classification of buildings in urban areas.

Bio:

Nathan Jacobs earned a Ph.D. in Computer Science from Washington University in St. Louis (2010) and a B.S. in Computer Science from the University of Missouri. He is currently an Associate Professor of Computer Science at the University of Kentucky. Dr. Jacobs' research area is computer vision; his specialty is developing learning-based algorithms and systems for processing large-scale image collections. He is a recipient of an NSF CAREER award, and his research has been funded by DARPA, IARPA, NGA, Army Research Lab, Google, ARMY-SMDC, and NIH. His current focus is on developing techniques for mining information about people and the natural world from geotagged imagery, including images from social networks, publicly available outdoor webcams, and satellites.

Abstract:

In this talk, I cover the current state of standardization of the MPI-4 version of the Message Passing Interface (and beyond). For pre-exascale and exascale systems, MPI is being "upgraded." Several new areas are planned for standardization for MPI-4 (MPI-Next, circa 2020) and MPI-Next-next (circa 2023-24), in time for the nominal arrival of the first exascale computers. Areas of particular interest include Sessions technologies for managing large, complex and changing resources underlying the MPI "job." Persistent collective communication is designed to add performance to applications with temporal locality in their use of collective operations. Bleeding-edge area including endpoints, finepoints (partitioned communication), and fault-tolerance models are discussed. A number of other "proposals" and "concepts" will be discussed as well. MPI+X models are also discussed, such as MPI+OpenMP. Sessions and fault-tolerant models offer the potential for significant new application areas for MPI in large-scale cloud applications as well. Improvements to MPI that enable and enhance algorithmic development such as for machine learning are also covered.

Bio:

Anthony (Tony) Skjellum studied at Caltech (BS, MS, PhD). His PhD work emphasized portable, parallel software for large-scale dynamic simulation, with a specific emphasis on message-passing systems, parallel nonlinear and linear solvers, and massive parallelism. From 1990-93, he was a computer scientist at LLNL focusing on performance-portable message passing and portable parallel math libraries. From 1993-2003, he was on the faculty in Computer Science at Mississippi State University, where his group co-invented the MPICH implementation of the Message Passing Interface (MPI) together with colleagues at Argonne National Laboratory. From 2003-2013, he was professor and chair at the University of Alabama at Birmingham, Dept. of Computer and Information Sciences. In 2014, he joined Auburn University as Lead Cyber Scientist and led R&D in cyber and High-Performance Computing for over three years. In Summer 2017, he joined the University of Tennessee at Chattanooga as Professor of Computer Science, Chair of Excellence, and Director, SimCenter, where he continues work in HPC and Cybersecurity, with strong emphases on IoT and blockchain technologies. He is a senior member of ACM, IEEE, ASEE, and AIChE, and an Associate Member of the American Academy of Forensic Science (AAFS), Digital & Multimedia Sciences Division.

Abstract:

The next generation sequencing (NGS) technology has substantially advanced the understanding of cancer biology, identification of potential therapeutic targets, and development of personalized cancer treatment. To enhance NGS-based research at the Markey Cancer Center, we have developed data processing pipelines as well as customized downstream analysis strategies for whole-exome sequencing, RNAseq, ChIPseq, and bisulfite sequencing. Furthermore, we have developed new bioinformatics methods to better understand the mechanism of tumorigenesis and identify potential biomarkers for cancer therapies. Our methods have been applied to data from both cancer patients in Appalachia Kentucky and The Cancer Genome Atlas (TCGA).

Bio:

Chi Wang is an Associate Professor of Biostatistics in the College of Public Health, and the Associate Director of Bioinformatics in the Biostatistics and Bioinformatics Shared Resource Facility in the Markey Cancer Center at the University of Kentucky (UK). Prior to joining UK, he was an Assistant Professor in the Department of Statistics at the University of California, Riverside.

His primary research interest is to develop new statistical and bioinformatics methods for high-throughput genomic, epigenomic, transcriptomic, and metabolomic data and to apply the methods to cancer-related projects. He is also interested in causal inference, survival analysis, and Bayesian methods. He received a doctoral degree in biostatistics from the Johns Hopkins Bloomberg School of Public Health, and both a master’s and bachelor’s degree in statistics from Peking University. He has published over 60 peer-reviewed papers.

Abstract:

The rising usage of systems biology modeling to probe complex signaling pathways raises the potential for their application in computational drug design. Toward this end, we are developing a workflow that predicts the ability of a computationally-designed calcium handling protein to restore dysregulated calcium signaling in cardiac ventricular tissue. The workflow includes 1) a systems biological model of calcium signaling under disease-like conditions 2) sensitivity analysis to identify calcium-dependent pathways whose modulation could recover normal signaling and 3) machine learning to identify potential proteins and sequence mutations to controllably modulate a given calcium signaling pathway. Preliminary results will be discussed in this presentation.

Bio:

Peter Kekenes-Huskey, Ph.D. (PI) is an assistant professor at the University of Kentucky. He graduated summa cum laude with a B.S. in Chemistry from the University of North Carolina Asheville (2001), which was followed by a Fulbright Fellowship to Germany, whereafter he obtained his Ph.D. at the California Institute of Technology (2009). At UNC Asheville, he developed algorithms for the ab initio prediction of halogenated ethane decomposition rates as a Barry Goldwater Scholar with Bert Holmes and George Heard (UNCA). His Ph.D. research with William A. Goddard, III (Caltech) focused on developing Monte Carlo tools and molecular dynamics simulations. Peter complemented his studies with applied mathematics and computer science coursework from algorithm design to parallel computing, in fulfillment of National Science Foundation and Department of Energy Computional Science Graduate Fellowships. Following his Ph.D. work, he worked as a Staff Scientist at Arete Associates from 2007 to 2010, where he developed near-real time image processing and signal detection algorithms for defense applications. Returning to his passion for the physical sciences, he pursued postdoctoral studies with Professors Andy McCammon (Chemistry) and Andrew McCulloch (Bioengineering) at the University of California San Diego under support of American Heart Association and National Institutes of Health fellowships, to understand cardiac calcium signaling using molecular dynamics, partial differential equations and systems biology models. At UK, he is expanding multi-physics descriptions of molecular-driven events with macro-scale phenomena, with myriad applications in biological and nanoscale systems.

Abstract:

Advances in sensor and computer technology over the past generation have transformed many aspects of the earth sciences. At the global level, some of the world’s fastest multi-petaflop supercomputers are used for geophysical data processing and reservoir simulation in the energy industry and climate simulations by academic and government scientists. Closer to home, the 2018 completion of a statewide digital elevation model (DEM) based on airborne LiDAR (a type of laser scanning) offers exceptionally detailed documentation of Kentucky’s topography and insights into the geological processes that gave rise to it. The statewide LiDAR DEM comprises approximately 46 billion centimeter-accurate bare earth heights interpolated on a uniform 5 ft (1.5 m) grid. The underlying unstructured LiDAR point cloud, which includes non-ground returns and ancillary data such as laser return intensity, from which the DEM was created is estimated to comprise about 460 billion records. Subsequent LiDAR data acquisition campaigns, which will allow for detailed estimation of topographic change as they increase both storage and processing demands, are in the planning stages. The Kentucky Geological Survey has established its Digital Earth Analysis Lab—the KGS DEAL—to support development of LiDAR solutions beneficial to the commonwealth. Ongoing and potential research includes coupling physics-based slope stability models with real-time precipitation data to create a statewide landslide and mudflow hazard warning system; developing deterministic and AI-based approaches to sinkhole mapping in an effort to reduce costly damage associated with sudden ground collapse; and using convolutional neural networks to locate improperly abandoned oil and gas wells that may pose environmental and health risks. KGS also operates the Kentucky Seismic and Strong Motion Network, which streams nearly 2 Gb per day from 28 seismic stations across the commonwealth and is capable of detecting events ranging from imperceptible microtremors to major New Madrid seismic zone earthquakes (as well as noise from sources such as mine blasts, passing trains, and electronic interference). KGS is currently developing AI approaches to support automated event detection, particularly of small earthquakes that would be impossible to identify manually, as well as exploring options for stochastic simulations of earthquake effects that take into account the Kentucky’s complicated bedrock and surficial geology.

Bio:

Bill Haneberg is the state geologist and director of the Kentucky Geological Survey and a research professor in the UK Department of Earth & Environmental Sciences. His expertise includes digital terrain modeling using high resolution topographic and bathymetric data, computational geology, GIS applications, and geohazard and risk assessment. Before moving to Kentucky in 2016, he spent 17 years as a consulting geologist, first as a sole practitioner based mainly in the Seattle area and then with Fugro in Houston, where he led a team of geologists, geophysicists, and engineers working on quantitative geohazard and risk assessments for deep-water oil and gas projects around the world. He previously spent 11 years at New Mexico Tech, where he was a research geologist and assistant director of the New Mexico Bureau of Mines & Mineral Resources. The author or co-author of 160+ published papers and abstracts, co-editor of 3 monographs, and author of the AEG Claire P. Holdredge award winning book Computational Geosciences with Mathematica, Bill was also the 2011 AEG-GSA Richard H. Jahns Distinguished Lecturer in Engineering Geology. He earned a B.S. in geology from Bowling Green State University and his M.S. and Ph.D. in geology from the University of Cincinnati.

Abstract:

CCS, in partnership with the ITS Research Computing Group, has been developing infrastructure required to support new and growing demands of Data Science (aka Big Data) research across the campus. We have made considerable advances, improving UK's computational, data analytics, storage, and network capabilities. We will describe the new systems coming online, the capabilities they will bring to researchers, and procedures for usage and access.

Bio:

James Griffioen is a Professor in the Department of Computer Science at the University of Kentucky. He serves as the Director of the Center for Computational Sciences, and also as the Director of the Laboratory for Advanced Networking. His research interests include future internet architectures, cloud computing, software defined networks, protocol design, network monitoring, and testbed networks

Bio:

Dave Buchholz is a Principal Engineer & Worldwide Director for technical enabling for Intel. His focus is not only corporate solutions but capability and solution strategies around verticals like healthcare, education and hospitality. With over 28 years of experience leading client technologies, Dave specializes in Workplace Transformation, Consumerization, Enterprise Client Technology trends & usage models.

Bio:

Paul Kassebaum works as the global team lead for physics at MathWorks, establishing and maintaining collaborations with large physics research teams in the domains of condensed matter, astrophysics, and particle physics

Zahra Ronaghi

Title: Accelerating Data Science Workflows with RAPIDS

Abstract:

Abstract: The open source RAPIDS project allows data scientists to GPU-accelerate their data science and data analytics applications from beginning to end, creating possibilities for drastic performance gains and techniques not available through traditional CPU-only workflows. Learn how to GPU-accelerate your data science applications by:

• Utilizing key RAPIDS libraries like cuDF (GPU-enabled Pandas-like dataframes) and cuML (GPU-accelerated machine learning algorithms)
• Learning techniques and approaches to end-to-end data science, made possible by rapid iteration cycles created by GPU acceleration

Bio:

Zahra Ronaghi is a senior solution architect at NVIDIA working with higher education and research on high performance computing (HPC), deep learning (DL) and machine learning (ML). Prior to joining NVIDIA, Zahra was a postdoctoral fellow at Lawrence Berkeley National Laboratory (NERSC), where she worked on performance optimization and portability on DOE supercomputers. She graduated from Clemson University with a Ph.D. in biomedical engineering and received her B.S. and M.S. degrees in electrical engineering.

Igor Alvarado

Title: Real-Time High-Performance Computing (RT-HPC) for Simulation and Control of Cyber-Physical Systems

Abstract:

Traditionally, High-Performance Computing (HPC) systems have been designed to perform offline simulations under “flexible” time constraints and with no data streaming from sensors or to actuators. In this work, we describe a new generation of Real-Time HPC (RT-HPC) systems that are capable of performing simulations in real-time within tight (strict) time constraints, in a deterministic manner and with extremely low jitter and latency. This type of systems can directly collect data from sensors, perform signal processing, classification/labeling and control tasks, and generate output signals in real- time. RT-HPC systems can also be used in hardware-in-the-loop (HIL) simulation applications, as well as in distributed, agent-based simulation systems where highly deterministic networks and clock distribution mechanisms can be used for synchronization purposes (e.g. time and frequency can be distributed over a fiber optic network within sub- nanosecond accuracy). For computation purposes, heterogeneous architectures that combine multi-core CPUs, GPUs and FPGAs are used. We will discuss how to leverage FPGAs for the implementation of task/application-specific processors and systolic arrays especially designed to perform specific mathematical and statistical functions. Also, FPGAs can have an important role on the implementation of adaptive computing strategies where processors are continuously optimized and adapted to the task at hand. These and other features are critical for simulating and controlling cyber-physical systems with thousands of nodes (agents) and hundreds of thousands of I/O signals from/to sensors and actuators, including but not limited to massive MIMO wireless communications systems for 5G and mmWave phased-array antennas, smart-grid monitoring and control systems, smart-transportation systems, smart cities, swarms of drones and autonomous agents, etc.

Bio:

Igor Alvarado is the Business Development Manager for Academic Research at National Instruments (www.ni.com) where he help to develop collaborations and strategic partnerships with leading universities in the U.S. in such areas as Cyber-Physical Systems, Smart Energy Systems, Medical Imaging/Devices, Advanced Manufacturing and RF/Wireless Communications to advance scientific research and accelerate innovation with support from NSF, NIH, DoD, DoE and other funding agencies. Mr. Alvarado is a Mechanical Engineer (Kansas State University, ’84) and has been with NI since 1999. He is a NSF Innovation-Corps mentor and has more than 30 years of practical experience in successfully developing and growing markets for high-technology products and services in the U.S. and Latin America. He has led the design, development and deployment of real- time, measurement and intelligent control systems that involve advanced numerical methods and algorithms using high-performance embedded platforms. He’s a member of IEEE, ASME, SIAM, IS, AUSA, NORDP, APS and AAAS. In 2017, Mr. Alvarado received the prestigious Electrical and Computer Engineering Department Heads Association’s (ECEDHA) Industry Award for his contributions to the ECE discipline in the U.S.

Abstract:

Accelerating discovery in computational science and high performance computing environments requires compute, network and storage to keep pace with technological innovations. Within a single organization, interdepartmental and multi-site sharing of assets has become more and more crucial to success. Furthermore, as the growth of data is constantly expanding, storage workflows are exceeding the capabilities of the traditional file system. For most organizations, facing the challenge of managing terabytes, petabytes and even exabytes of archive data for the first time can force the redesign of their entire storage strategy and infrastructure. Increasing scale, level of collaboration and diversity of workflows are driving users toward a new model for data storage.

In the past, data storage usage was defined by the technology leveraged to protect data using a pyramid structure, with the top of the pyramid designated for SSD to store ‘hot’ data,’ SATA HDDs used to store ‘warm’ data and tape used for the bottom of the pyramid to archive ‘cold’ data. Today, modern data centers have moved to a new two-tier storage architecture that replaces the aging pyramid model. The new two-tier paradigm focuses on the actual usage of data, rather than the technology on which it resides. The new two-tier paradigm combines a project tier that is file-based and a second or perpetual tier which is object based. The object based perpetual tier includes multiple storage media types, multi-site replication (sharing), cloud, and data management workflows. Data moves seamlessly between the two tiers as data is manipulated, analyzed, shared and protected – essentially creating yin and yang between the two storage tiers. Solutions designed to natively use the Perpetual Tier empower organizations to fully leverage their primary storage investments by reducing the overall strain on the Primary Tier, while at the same time, enabling data centers to realize numerous benefits of the Perpetual Tier that only increase as the amount of storage to manage increases.
The next logical question is how to manage data between the two tiers while maintaining user access and lowering overall administration burdens. Join us for a deeper look into the nuances of the two-tier system and data management between them. We will cover storage management software options; cloud vs. on-premise decisions; and using object storage to expand data access and create a highly effective storage architecture to break through data life cycle management barriers.

Bio:

David Feller is an executive leader with a long history of delivering world changing consumer electronics products. As VP of Product Management and Solutions Engineering at Spectra Logic, he has ownership of the company’s strategic road map across all product lines. With 20+ years of experience in product management, strategic marketing, business development, operations, engineering and executive management, David has led nearly every aspect of small and fortune 100 business. He is especially passionate about product management and strategy – taking a market or customer requirement, evolving it into a concept, and molding an idea through development, launch, and mass production. Mr. Feller ran the product management/marketing group that invented wireless LAN, the group the pioneered putting mini hard drives in MP3/media players, and the group that made real-time security/surveillance remote access available home/business/worldwide.

Abstract:

The rapid emergence of AI, especially deep learning, presents tremendous new capabilities and possibilities for organizations and endeavors of all types: from government to education to research, and spanning all industries and consumer experiences. AI will enable the potential of the Internet of Things (IoT) to finally be realized, with smart devices at the edges making everything from smart buildings and factories to autonomous vehicles, drones and robots commonplace. Underpinning these transformations will be the advances in high performance computing (HPC), which will be increasingly needed to process streaming data from the “edges” and train more powerful AI models in data centers and clouds. The confluence of edge computing & IoT, advances in AI model accuracy, increasing scale of data centers and cloud computing, enterprise adoption of HPC, and even the global race to exascale computing will lead to integrations of these technologies that profoundly change business, government, research, national security, and society overall in the next decade.

Bio:

Jay Boisseau Vizias CEO Jay Boisseau is an experienced, recognized leader and strategist in advanced computing technologies, with over 20 years in the field. Jay has built and led many computing-focused programs, departments, and organizations. Jay’s is CEO and co-founder (June 2014) of Vizias. Vizias is a small team of talented, passionate technology professionals who are determined to change the world by using their advanced computing expertise and experience, especially in high performance computing (HPC), artificial intelligence (AI), and smart cities.

Jay’s current primary activity is working as the AI & HPC Technology Strategist for Dell EMC (since September 2014) to further develop its high-performance computing (HPC) and artificial intelligence (AI) strategies and new solutions for a broader use of HPC and AI by more companies, government organizations, and universities.

Through Vizias, Jay founded the Austin CityUP Consortium (July 2015) and currently serves as the executive director, with a vision of creating an integrated smart city fabric throughout Austin—leveraging mobile devices and IoT data collectors, as well as supercomputers and AI for predictive analytics and scenario simulation—in the years ahead to address city issues, empower city planning, and improve city life in general. Jay is also the executive director and founder of The Austin Forum on Technology & Society, which he started in 2006 and is now the leading monthly technology outreach and engagement event in Austin.

Abstract:

In the last 25 years the top performance of supercomputers has increased from teraflops to exaflops. This growth is explained by the exponential improvement in the cost/performance of commodity hardware embodied in Moore’s Law; and by the steady increase in the price of top supercomputers. On the other hand, the architectural evolution of commodity computers has had a negative impact on their performance.

These external engines of growth will not be available in the future: Moore’s Law is coming to an end; no alternative device technology seems ready to take over any time soon. As the cost of a top supercomputer approaches $1B, the scope for further price increases is limited.

This talk will elaborate on two claims:

1. Changes above the device level, in packaging, architecture, software and algorithms, can continue achieving growth in the sustained performance of top supercomputers in the next decade.
2. This growth will be achieved only if the business model of the supercomputing industry changes.

Bio:

Marc Snir is Michael Faiman Professor in the Department of Computer Science at the University of Illinois at Urbana-Champaign. He currently pursues research in parallel computing.

He was Director of the Mathematics and Computer Science Division at the Argonne National Laboratory from 2011 to 2016 and head of the Computer Science Department at Illinois from 2001 to 2007. Until 2001 he was a senior manager at the IBM T. J. Watson Research Center where he led the Scalable Parallel Systems research group that was responsible for major contributions to the IBM SP scalable parallel system and to the IBM Blue Gene system.

Marc Snir received a Ph.D. in Mathematics from the Hebrew University of Jerusalem in 1979, worked at NYU on the NYU Ultracomputer project in 1980-1982, and was at the Hebrew University of Jerusalem in 1982-1986, before joining IBM. Marc Snir was a major contributor to the design of the Message Passing Interface. He has published numerous papers and given many presentations on computational complexity, parallel algorithms, parallel architectures, interconnection networks, parallel languages and libraries and parallel programming environments.

Marc is AAAS Fellow, ACM Fellow and IEEE Fellow. He has Erdos number 2 and is a mathematical descendant of Jacques Salomon Hadamard. He won the IEEE Award for Excellence in Scalable Computing and the IEEE Seymour Cray Computer Engineering Award. He was recently awarded a Doctor Honoris Causa Degree from ENS, Lyon, France.

Abstract:

NSF's Office of Advanced Cyberinfrastructure (OAC) seeks to foster the advanced cyberinfrastructure (CI) that has become essential to progress of all areas of science and engineering research and education. OAC's investments in computing, data infrastructure, software, networking and cybersecurity, and people have enabled new innovations and discoveries. Yet, dramatic changes are occurring in the nature and requirements of science, in the scale and pervasiveness of research data, and in the landscape of technologies and resources. As will be described in this presentation, NSF has sought and received much community input on these changes - which are informing an evolution of OAC’s programs and investments in research CI and CI research to transform science in the 21st century.

Bio:

William Miller is the Science Advisor for the Office of Advanced Cyberinfrastructure (OAC) at the National Science Foundation. Bill leads strategic activities for OAC designed to catalyze new cyberinfrastructure (CI)-enabled pathways for scientific discovery, promote multi-disciplinary activities, and foster international engagement. He leads the Cyberinfrastructure for Emerging Science and Engineering Research (CESER) Program that supports early-stage science-CI collaborations. Current CESER foci are NSF priority investments including the NSF Big Ideas and NSF’s Major Multi-User Research Facilities. Bill has previously served as Acting Deputy of the NSF Office of Budget, Finance and Award Management, and in policy and oversight roles for large facility planning and construction. He has prior professional careers in space mission systems development and in experimental neuroscience, each conducted in the U.S. and internationally.

Bio:

Matt Koop is an architect in the Office of the CTO in the Server and Infrastructure Systems division of Dell EMC. Matt joined Dell EMC in 2016 after leading the global HPC architecture and engineering group for Schlumberger, a major commercial user of HPC. Prior to Schlumberger, Matt led a team providing technology transfer and application optimization expertise to the DoD HPC Modernization Program.

Matt holds a Ph.D. in Computer Science and Engineering from Ohio State University, an MBA from Georgetown University, and a B.S. in Computer Science from Calvin College.

Speaker Info:

Company: Spectra Logic
Position/title: VP of Product Management and Solutions Engineering
Email: davidfe@spectralogic.com
Phone: 303-449-6444,1436

Bio:

David Feller is an executive leader with a long history of delivering world changing consumer electronics products. As VP of Product Management and Solutions Engineering at Spectra Logic, he has ownership of the company’s strategic roadmap across all product lines. With 20+ years of experience in product management, strategic marketing, business development, operations, engineering and executive management, David has led nearly every aspect of small and fortune 100 business. He is especially passionate about product management and strategy – taking a market or customer requirement, evolving it into a concept, and molding an idea through development, launch, and mass production. Mr. Feller ran the product management/marketing group that invented wireless LAN, the group that pioneered putting mini hard drives in MP3/media players, and the group that made real-time security/surveillance remote access available home/business/worldwide.

Abstract: Considering Future Technologies in the World of Everlasting Data

Organizations across the globe have all faced the challenge of managing their unstructured data, which traditionally consists of 60%-90% of their total data. A major aspect of managing data is the ability for the data management software to tier your data, meaning it has the ability to move your inactive data to a more affordable tier of storage. This is where all solutions are not created equal, and organizations need to determine the best way to not only manage their data, but also protect the data that is being managed. How do organizations implement a tiered storage solution that protects their data for as long as they need to keep it? Many data management software products do a great job of identifying the data that should be offloaded from the user’s primary storage (tier 1 storage) but what happens to the data once its moved, and how protected is the data that is now essentially archived? How do you get the data back when you need it? Complete solutions should not only assist in the data management process but ensure that the inactive data being moved is securely protected and always available when its needed. This new focus on archive is not meant to replace backup solutions, but is necessary to enhance data protection for historically critical data. Join us to find out the newest data management strategies to ensure that your primary storage is free of inactive files, your inactive files are efficiently moved to a more affordable tier of storage and your data is protected so that it is forever available when it’s needed.

Bio:

Will has over 8 years in HPC, Big Data, & Artificial Intelligence workloads as well as how customers tailor these solutions towards their requirements. Will currently is the HPE Business Development & Technical Strategy Manager for Artificial Intelligence in North America. He has worked in previous roles in Hyperscale server product management, Big Data, & HPC Solutions. Will is working to enable edge to cloud to core designs with partners & customers to enable A.I.

Abstract:

Multidimensional streaming signals like audio, videos, and multi-spectral imagery are truly the biggest of the Big Data. Collected by sensors at the network edge, these data often contain highly sensitive information and their misuse can lead to significant invasion of privacy. For example, networks of surveillance cameras have been used by some countries in tracking their citizens, and pervasive use of IoT devices in smart homes opens door to recording private behaviors and interactions. It is highly challenging to apply traditional privacy enhancing technologies to handle these data. Cryptographic techniques like homomorphic encryption and garbled circuits are too computationally intensive for any practical applications. The more efficient differential private schemes, on the other hand, rely on additive noise and may not be able to provide adequate protection on semantic contents.

In this talk, I will discuss a number of techniques developed in my research group to process multidimensional signals while ensuring their privacy. Specifically, I will present our work in using generative adversarial network (GAN) and random neural networks to protect signal privacy in distributed deep learning. We use GAN to learn the statistics of sensitive facial data at local sites and generate privacy-preserving synthetic data for public centralized learning. While GAN at network edge can be computationally intensive, random neural network offers a much lighter weight privacy-enhancing protocol suitable for a broad range of IoT applications.

Bio:

Sen-ching “Samson” Cheung is a Professor of Electrical and Computer Engineering and the director of Multimedia Information Laboratory (Mialab) at University of Kentucky (UKY), Lexington, KY, USA. He is the endowed Blazie Family Professor of Engineering. Before joining UKY in 2004, he was a postdoctoral researcher with the Sapphire Scientific Data Mining Group at Lawrence Livermore National Laboratory that won the R&D 100 Award in 2006. He received his Ph.D. degree from University of California, Berkeley in 2002. He is a senior member of both IEEE and ACM. He has the fortune of working with a team of talented students and collaborators at Mialab in several areas in multimedia including video surveillance, privacy protection, encrypted domain signal processing, 3D data processing, virtual and augmented reality as well as computational multimedia for autism therapy. More details about current and past research projects at Mialab can be found at http://vis.uky.edu/mialab.