BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T000714Z
LOCATION:DEF Concourse
DTSTART;TZID=America/Denver:20231116T100000
DTEND;TZID=America/Denver:20231116T170000
UID:submissions.supercomputing.org_SC23_sess300@linklings.com
SUMMARY:ACM Student Research Competition Posters Display
DESCRIPTION:ProxyStreams: Leveraging Lightweight Proxies for Portable Stre
 ams\n\nA novel streaming approach is introduced for Python, leveraging the
  ProxyStore system to facilitate the exchange of stream references across 
 distributed systems. This approach utilizes generators to efficiently publ
 ish and consume messages from streams. The extensible backend connector in
 terface of ...\n\n\nNazanin Mahmoudi (Wayne State University) and Valerie 
 Hayot (University of Chicago)\n---------------------\nCray EX40 Cluster In
 trusion Detection System\n\nAnalysis of a High-Performance Computing clust
 er’s external network traffic provides the opportunity to identify securit
 y issues, cluster misuse, or configuration problems without reducing perfo
 rmance. This project captured the external network traffic to and from a C
 ray EX40 cluster over thre...\n\n\nDaniel Wild (Los Alamos National Labora
 tory (LANL), University of New Mexico)\n---------------------\nBetter Data
  Splits for Machine Learning with Astartes\n\nMachine Learning (ML) has be
 come an increasingly popular tool to accelerate traditional workflows. Cri
 tical to the use of ML is the process of splitting datasets into training,
  validation, and testing subsets to develop and evaluate models. Common pr
 actice is to assign these subsets randomly. Althou...\n\n\nJackson Burns (
 Massachusetts Institute of Technology (MIT))\n---------------------\nDynam
 ic and First-Class Priorities\n\nInteractive parallel programs have varyin
 g responsiveness requirements for tasks of differing urgency, which has be
 en met with the solution of thread priorities to determine the tasks' allo
 cation of processor time. Previous priority-based language models limit th
 e span of entire threads to a single ...\n\n\nMarelle León (Illinois Insti
 tute of Technology)\n---------------------\nA Reinforcement Learning-Based
  Backfilling Strategy for HPC Batch Jobs\n\nHigh Performance Computing (HP
 C) systems are essential for various scientific fields, and effective job 
 scheduling is crucial for their performance. Traditional backfilling techn
 iques, such as EASY-backfilling, rely on user-submitted runtime estimates,
  which can be inaccurate and lead to suboptimal ...\n\n\nElliot Kolker-Hic
 ks (University of North Carolina, Charlotte)\n---------------------\nHow M
 uch Noise Is Enough:  On Privacy, Security, and Accuracy Trade-Offs in Dif
 ferentially Private Federated Learning\n\nCentralized machine learning tec
 hniques have caused privacy concerns for users. Federated Learning~(FL) mi
 tigates this as a decentralized training system where no raw data are comm
 unicated across the network to a centralized server. Instead, the machine 
 learning model is trained locally on each devi...\n\n\nAdhishree Kathikar 
 (Indiana University)\n---------------------\nFast Operations on Compressed
  Arrays without Decompression\n\nIn modern scientific computing and machin
 e learning systems, data movement has overtaken compute as the performance
  bottleneck, thus motivating the wider adoption of lossy data compression.
  Unfortunately, state-of-the-art floating-point array compressors such as 
 SZ and ZFP require decompression befo...\n\n\nHarvey Dam (University of Ut
 ah)\n---------------------\nA Heterogeneous, In Transit Approach for Large
  Scale Cellular Modeling\n\nThe field of in silico cellular modeling has m
 ade notable strides in number of cells that can be simultaneously modeled.
  While computational capabilities have grown exponentially, I/O performanc
 e has lagged behind. To address this issue, we present an in-transit appro
 ach to enable in situ visualizat...\n\n\nAyman Yousef (Duke University)\n-
 --------------------\nJob Level Communication-Avoiding Detection and Corre
 ction of Silent Data Corruption in HPC Applications\n\nDetecting and corre
 cting Silent Data Corruption (SDC) is of high interest for many HPC applic
 ations due to the dramatic consequences such undetected computation errors
  can have. Additionally, going into the exascale era of computing, SDC err
 or rates are only increasing with growing system sizes. Sta...\n\n\nLaslo 
 Hunhold (University of Cologne)\n---------------------\nCase Study for Per
 formance Portability of GPU Programming Frameworks for Hemodynamic Simulat
 ions\n\nPreparing for the deployment of large scientific and engineering c
 odes on GPU-dense exascale systems is made challenging by the unprecedente
 d diversity of vendor hardware and programming model alternatives for offl
 oad acceleration. To leverage the exaflops of GPUs from Frontier (AMD) and
  Aurora (Int...\n\n\nAristotle Martin (Duke University)\n-----------------
 ----\nIncremental Graph Clustering in Parallel\n\nWe develop a distributed
  memory graph clustering algorithm to find clusters in a graph where new n
 odes and edges are being added incrementally. At each stage of the algorit
 hm, we maintain a summary of the clustered graph computed from all increme
 ntal batches received thus far. As we receive a new ba...\n\n\nMd Taufique
  Hussain (Indiana University)\n---------------------\nEnabling Transparent
 , High-Throughput Data Movement for Scientific Workflows on HPC Systems\n\
 nThis poster presents the DYnamic and Asynchronous Data Streamliner (DYAD)
  middleware that provides an efficient and transparent method for data mov
 ement in scientific workflows based on the producer-consumer paradigm. We 
 develop DYAD on top of Flux, a fully hierarchical HPC workload manager, an
 d Uni...\n\n\nIan Lumsden (University of Tennessee)\n---------------------
 \nUtilizing Large Language Models for Disease Phenotyping in Obstructive S
 leep Apnea\n\nObstructive sleep apnea (OSA) impacts millions, linking to s
 evere complications yet understanding its influence on comorbidities lags.
  Complications can be avoided by using expensive continuous positive airwa
 y pressure (CPAP) machines, but physicians cannot identify those at risk. 
 Large language mod...\n\n\nIfrah Khurram (San Juan Bautista School of Medi
 cine, Lawrence Berkeley National Laboratory (LBNL))\n---------------------
 \nFast Checkpointing of Large Language Models with TensorStore CHFS\n\nThe
  frequency of checkpoint creation in large language models is limited by t
 he write bandwidth to a parallel file system.  In this study, we aim to re
 duce the checkpoint creation time by writing to the Intel Optane Persisten
 t Memory installed on the compute nodes.\n\nWe propose TensorStore CHFS, a
  st...\n\n\nSohei Koyama (University of Tsukuba)\n---------------------\nC
 loud Computing at Scale:  Tracking 4.5 Million Heartbeats of 3D Coronary F
 low via the Longitudinal Hemodynamic Mapping Framework\n\nTracking hemodyn
 amic responses to treatment and stimuli for long periods is a grand challe
 nge. Moving from established single-heartbeat technology to longitudinal p
 rofiles would require continuous data reflecting a patient's evolving stat
 e, methods to extend the temporal domain that could be feasibl...\n\n\nCyr
 us Tanade (Duke University)\n---------------------\nScaling Infrastructure
  to Support Multi-Trillion Parameter LLM Training\n\nThis poster discusses
  efficient system designs for Large Language Model (LLM) scaling to up to 
 128 trillion parameters. We use a comprehensive analytical performance mod
 el to analyze how such models could be trained on current systems while ma
 intaining 75% Model FLOPS Utilization (MFU). We first sho...\n\n\nMikhail 
 Isaev (Georgia Institute of Technology)\n---------------------\nAccelerati
 ng CRUD with Chrono Dilation for Time-Series Storage Systems\n\nIn recent 
 years, we have seen an un-precedented growth of data in our daily lives ra
 nging from health data from an Apple Watch, financial stock price data, vo
 latile crypto-currency data, to diagnostic data of nuclear/rocket simulati
 ons. The increase in high-precision, high-sample-rate time-series da...\n\
 n\nLan Nguyen (Illinois Institute of Technology)\n---------------------\nS
 eeing the Trees for the Forest:  Describing HPC Filesystem Trees with the 
 Grand Unified File-Index (GUFI)\n\nHigh performance computing (HPC) filesy
 stems are extremely large, complex, and difficult to manage with existing 
 tools. It is challenging for HPC administrators to describe the current st
 ructure of their filesystems, predict how they will change over time, and 
 the requirements for future filesystems...\n\n\nJenna Kline (Ohio State Un
 iversity, Los Alamos National Laboratory (LANL))\n---------------------\nR
 oad To Reliability:  Optimizing Self-Driving Consistency With Real-Time Sp
 eed Data\n\nSelf-driving cars can potentially improve transportation effic
 iency and reduce human fatalities – provided they have access to significa
 nt processing power and large amounts of data. One popular approach for ac
 tualizing autonomous vehicles is using end-to-end learning, in which a mac
 hine learn...\n\n\nWilliam Fowler (Tufts University)\n--------------------
 -\nUsing Deep Neural Networks to Classify Hot-Cold Data Storage\n\nThe Sci
 entific Data and Computing Center (SDCC) at Brookhaven National Laboratory
  manages a data storage system with millions of files totaling petabytes o
 f data. To optimize costs, they use a multi-tiered storage approach based 
 on data temperature, storing infrequently accessed ("cold") data on che...
 \n\n\nKeene Lu (Northwestern University) and Ai Kagawa (Brookhaven Nationa
 l Laboratory)\n---------------------\nSensitivity of Black-Box Statistical
  Prediction of Lossy Compression Ratios for 3D Scientific Data\n\nCompress
 ion ratio estimation is an important optimization of I/O workflows process
 ing terabytes of data. Applications such as compression auto-tuning or los
 sy compressor selection require a high-throughput, accurate estimation. Pr
 ior works that utilize sampling are fast but inaccurate, while approac...\
 n\n\nAlexandra Poulos (Clemson University)\n---------------------\nSimulta
 neous Evaluation of Mindful Fault Checking across the CPU and GPU\n\nThis 
 work comprehensively analyzes the overhead when implementing fault-checkin
 g algorithms for sparse preconditioned conjugate gradient (PCG) solvers on
  many-core and GPU-accelerated systems. Our objective is to selectively ut
 ilize GPUs for duplicate calculations based on the numerical properties o.
 ..\n\n\nHayden Estes (University of Alabama, Huntsville)\n----------------
 -----\nLossy and Lossless Compression for BioFilm Optical Coherence Tomogr
 aphy (OCT)\n\nOptical Coherence Tomography (OCT) can be used as a fast and
  non-destructive technology for bacterial biofilm imaging.  However, OCT g
 enerates approximately 100 GB per flow cell, which complicates storage and
  data sharing. Data reduction can reduce data complications by reducing th
 e overhead and the...\n\n\nMax Faykus (Clemson University)\n--------------
 -------\nNetCDFaster: A Geospatial Cyberinfrastructure for Multi-Dimension
 al Scientific Datasets Full-Stack I/O and Visualization\n\nNetCDF's origin
 al design included a portable file format and an intuitive application pro
 gramming interface (API). However, the current NetCDF framework and its de
 rived libraries lack efficient support for querying and visualizing data s
 ubsets with low memory use and time cost. Therefore, a full-sta...\n\n\nZh
 enlei Song (Texas A&M University)\n---------------------\nScaling Studies 
 for Efficient Parameter Search and Parallelism for Large Language Model Pr
 etraining\n\nAI accelerator processing and memory constraints largely dict
 ate the scale in which machine learning workloads (training and inference)
  can be executed within a desirable time frame. Training a transformer-bas
 ed model requires the utilization of HPC harnessed through inherent parall
 elism embedded in...\n\n\nChris Pierre Paul (Oak Ridge Institute For Scien
 ce And Education, Florida State University) and Leo Phan (Oak Ridge Instit
 ute For Science And Education, George Washington University)\n------------
 ---------\nA Comparison of Deep and Shallow Residual Networks for Medical 
 Imaging Classification\n\nThe complexity and parameters of mainstream larg
 e models are increasing rapidly. For example, the increasingly popular lar
 ge language models (e.g., ChatGPT) have billions of parameters. While this
  has led to performance improvements, the performance gains for simple tas
 ks may be unacceptable for the...\n\n\nElaine Lu (Columbia University)\n\n
 Registration Category: Tech Program Reg Pass, Exhibits Reg Pass
END:VEVENT
END:VCALENDAR