BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Chicago
X-LIC-LOCATION:America/Chicago
BEGIN:DAYLIGHT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
TZNAME:CDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
TZNAME:CST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20211207T055400Z
LOCATION:Second Floor Atrium
DTSTART;TZID=America/Chicago:20211118T083000
DTEND;TZID=America/Chicago:20211118T170000
UID:submissions.supercomputing.org_SC21_sess243@linklings.com
SUMMARY:ACM Student Research Competition Posters Display
DESCRIPTION:ACM Student Research Competition: Graduate Poster, ACM Student
  Research Competition: Undergraduate Poster, Posters\n\ncuSZ(+): Optimizin
 g Error-Bounded Lossy Compression for Scientific Data on Modern GPUs\n\nTi
 an\n\nError-bounded lossy compression is a critical technique for signific
 antly reducing scientific data volumes. With ever-emerging heterogeneous h
 igh-performance computing (HPC) architecture, GPU-accelerated error-bounde
 d compressors (such as cuSZ and cuZFP) have been developed. However, they 
 suffer fro...\n\n---------------------\nDelving Into the Abyss: A Distribu
 ted Decompression System for Indexing Compressed Repositories\n\nWong\n\nD
 iscovery and use of scientific data is dependent on descriptive metadata. 
 Unfortunately, data lakes often contain compressed data, which are difficu
 lt to index and automatically extract metadata from due to storage and I/O
  constraints when inflating and processing recursively compressed data. He
 re ...\n\n---------------------\nIn-Situ Data Reduction for AMR-Based Cosm
 ology Simulations\n\nWang\n\nLarge-scale cosmology simulations generate la
 rge amounts of data for post-analysis, resulting in I/O and storage bottle
 necks. This work investigates an effective in-situ error-bounded lossy com
 pression for Nyx, an adaptive mesh refinement (AMR) based cosmology applic
 ation. Our contribution is threef...\n\n---------------------\nOne Fish, T
 wo Fish: Choosing Optimal Edge Topologies for Real-Time Autonomous Fish Su
 rveys\n\nTsen\n\nThe oceans represent more than seventy percent of the Ear
 th's surface, and marine ecosystems are central to many global challenges.
  However, monitoring the environment through state-of-the-art devices that
  require the use of GPUs presents unique challenges for the development of
  computer vision, art...\n\n---------------------\nEfficient Parallel Algo
 rithm for Shortest Path Updates in Dynamic Networks at Scale\n\nKhanda\n\n
 The application of graphs (networks) are versatile, and they are applied t
 o various fields, including but not limited to social network analysis, tr
 ansportation logistics, biological and genetic interaction study, IP traff
 ic routing, and resource allocation in IoT networks. All these domains dea
 l wi...\n\n---------------------\nEmbeddings Are All You Need: Transfer Le
 arning in Convolutional Neural Networks Using Word Embeddings\n\nBaughman\
 n\nRecent advances in efficient neural networks and relational learning us
 ing word embeddings as prediction targets for image classification indicat
 e the combination of these two  concepts offers promise for efficient tran
 sfer learning. Given the properties of word embeddings to represent inform
 ation-d...\n\n---------------------\nAnalyzing Performance of File Aggrega
 tion in VELOC\n\nGossman\n\nAs High-Performance Computing (HPC) systems an
 d applications continue to grow in size and complexity, the process of che
 ckpointing to stable, external storage often results in I/O contention and
  degraded performance. Multi-level asynchronous checkpointing strategies l
 ike VELOC (Very Low Overhead Che...\n\n---------------------\nMissing the 
 Trees for the Branches: Graphical-Scripting Interaction with Large-Scale C
 alling Context Trees\n\nScully-Allison\n\nIn large scale performance analy
 sis of HPC code, there is a tension between the visualizations provided by
  automated tools and the flexibility afforded by custom scripting. For cal
 ling context trees (CCT), node-and-link visualizations give users a view o
 f performance data that corresponds to their m...\n\n---------------------
 \nOptimizing Deep Learning Material Interface Reconstruction\n\nYamamoto\n
 \nCurrent material interface reconstruction methods provide inaccurate rec
 onstructions; a neural network provides more accurate reconstructions. The
  initial model architecture was too large to provide the required throughp
 ut. Reducing the size of the model shows a smaller model could be used and
  provi...\n\n---------------------\nPerformance Prediction of Large Data T
 ransfers\n\nCheung\n\nScientific facilities around the world transfer tera
 bytes of data to Berkeley Lab’s National Energy Research Scientific Comput
 ing Center (NERSC) for processing. These large data transfers can cause co
 ngestion on the computer network. To better manage these large transfers, 
 we plan to predict their e...\n\n---------------------\nAnalysis of Schedu
 ling Policies for Next-Generation Rabbit Architecture\n\nBateman\n\nThe Li
 vermore El Capitan supercomputer is planned to have a brand-new architectu
 re, with Rabbit nodes containing SSDs placed at the top of the racks. This
  will allow SSDs to be accessed either directly through PCIe connection or
  through the network fabric, which creates opportunities for new schedul..
 .\n\n---------------------\nProductive, Performant, and Parallel Generic L
 ossy Data Compression with LibPressio\n\nMalvoso\n\nData compression is vi
 tal in scientific applications because it reduces the size of files. This 
 is important because running experiments on the data and storing it requir
 es significantly less time and memory. In an effort to create more  effici
 ency, LibPressio was created as a single interface to ut...\n\n-----------
 ----------\nApplication-Based Fault Tolerance for Numerical Linear Algebra
  at Large Scale\n\nTorres Gonzalez\n\nLarge-scale architectures provide us
  with high computing power, but as the size of the systems grows, computat
 ion units are more likely to fail. Fault-tolerant mechanisms have arisen i
 n parallel computing to face the challenge of dealing with errors that may
  occur at any moment during the execution ...\n\n---------------------\nMi
 tigating the Metadata Mess:  Autonomous Metadata Extraction Pipelines for 
 Large-Scale Data Repositories\n\nChen, Hsu\n\nMany scientific repositories
  are rendered useless due to their enormous size (exceeding petabytes of d
 ata across billions of files) and lack of descriptive metadata to aid disc
 overy, understanding, and use. Building on a distributed metadata extracti
 on service, Xtract, we propose a scheduler designe...\n\n-----------------
 ----\nYin and Yang: Balancing Cloud Computing and HTC Workloads\n\nKang, Z
 hen, Keahey\n\nWith the help of proper preemption policies and proactive r
 esource schedulers, combining academic cloud and High Throughput Computing
 (HTC) systems through preemptible instances would help increase the utiliz
 ation rate of the clouds and reduce the energy cost at the same time. We p
 ropose a data-drive...\n\n---------------------\nRegulating Traffic in a C
 rowded Cache:  Overcoming the Container Explosion Problem\n\nGao\n\nMulti-
 user interactive computing services, such as Binder, dynamically create an
 d deploy software containers to provide customized execution environments 
 with required system and language dependencies. Unfortunately, container c
 reation can be slow, making services unresponsive to users, while cachin..
 .\n\n---------------------\nUsing IU Jetstream for OpenMP Offloading and O
 penACC Testsuites\n\nJarmusch, Baker\n\nResearchers often run large scient
 ific applications, which can take days to complete, from weather models to
  arithmetic calculations. Using OpenMP or OpenACC is one way of accelerati
 ng applications on large scale systems, thus reducing the time for complet
 ion. The OMP SOLLVE & OpenACC validation and...\n\n---------------------\n
 Datastore Design for Analysis of Police Broadcast Audio at Scale\n\nAhmad\
 n\nWith policing coming under greater scrutiny in recent years, researcher
 s have begun to more thoroughly study the effects of contact between polic
 e and minority communities. Despite data archives of hundreds of thousands
  of recorded Broadcast Police Communications (BPC) being openly available 
 to the ...\n\n---------------------\nAnalyzing Software Cache Configuratio
 n for In-line Data Compression\n\nRanjan\n\nIn order to compute on or anal
 yze larger data sets, application need access to large amounts of DRAM mem
 ory. To increase the size of memory requires a costly hardware upgrade. Co
 mpressing data structures stored in memory does not require hardware upgra
 des. Inline compression compresses and decompres...\n\n-------------------
 --\nIncreasing Mini-Batch Size While Preserving Accuracy for Distributed D
 eep Learning\n\nMcCord\n\nData-parallel distributed training is an effecti
 ve way to reduce training times for complex deep learning models and large
 -scale datasets. In distributed training, each worker requires enough work
  to justify the communication overhead and maintain a balanced computation
 -to-communication ratio. This ...\n\n---------------------\nParallel Algor
 ithms and Generalized Frameworks for Learning Large-Scale Bayesian Network
 s\n\nSrivastava\n\nBayesian networks (BNs) are an important subclass of gr
 aphical machine learning (ML) models that enable probabilistic reasoning a
 bout interactions between variables of interest. Their interpretability ma
 kes them an ideal model for making high-stakes decisions in fields where e
 xplainability is desira...\n\n---------------------\nToward Access Pattern
  Aware Checkpointing for Kokkos Applications\n\nTan\n\nThe common checkpoi
 nt philosophy, checkpoint everything as frequently as possible, is becomin
 g ineffective as we progress towards exascale machines, facing shrinking t
 ime between failures. This makes portability and resilience vital for the 
 future of HPC. This poster demonstrates the need and forms ...\n\n--------
 -------------\nSYMBIOMON: A High Performance, Composable Monitoring Servic
 e for Online Application Introspection and Adaptation\n\nRamesh\n\nThe con
 struction of high-performance scientific software has shifted from a tradi
 tional monolithic message-passing interface (MPI) executable model to a co
 upled, services-style model in which simulations run alongside a host of d
 istributed HPC data services within the same batch job allocation. Perf...
 \n\n\nTag: In-Person Only\n\nRegistration Category: Tech Program Reg Pass,
  Exhibit Hall Only
END:VEVENT
END:VCALENDAR