BEGIN:VCALENDAR VERSION:2.0 PRODID:Linklings LLC BEGIN:VTIMEZONE TZID:Europe/Stockholm X-LIC-LOCATION:Europe/Stockholm BEGIN:DAYLIGHT TZOFFSETFROM:+0100 TZOFFSETTO:+0200 TZNAME:CEST DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:+0200 TZOFFSETTO:+0100 TZNAME:CET DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20260625T133339Z LOCATION:Bldg. 6 - 001 - Plenary Room DTSTART;TZID=Europe/Stockholm:20260630T121500 DTEND;TZID=Europe/Stockholm:20260630T124500 UID:submissions.pasc-conference.org_PASC26_sess129@linklings.com SUMMARY:Flash Poster Session - Part II DESCRIPTION:P41 - Using Generative Machine Learning to Produce High-Resolu tion Weather Data\n\nGenerative machine learning techniques show promise for performing atmospheric downscaling (super-resolution for meteorologica l data) to produce high-resolution weather and climate simulations. Previo us work has evaluated the quality of these models using standard error sco res such as RMSE, absolut...\n\n\nPetar Stamenkovic and Mary McGlohon (Met eoSwiss, ETH Zurich); David Leutwyler and Xavier Lapillonne (MeteoSwiss); Fabian Bösch, Lukas Drescher, and Henrique Mendonça (ETH Zurich / CSCS); S ebastian Schemm (University of Cambridge); Siddhartha Mishra (ETH Zurich); and Oliver Fuhrer (MeteoSwiss)\n---------------------\nP36 - Stencil Comp utation on Tenstorrent Wormhole\n\nThe rapid ascent of large language mode ls (LLMs) has prioritized domain-specific accelerators (DSAs) optimized fo r dense matrix-based deep learning. However, the suitability of these arch itectures for traditional high-performance computing (HPC) kernels, like s tencil-based partial differential equat...\n\n\nLorenzo Piarulli and Danie le De Sensi (Sapienza University)\n---------------------\nACMP11 - Mosaic: Automatic Categorization of I/O Patterns from Scientific Applications\n\n While newly deployed High-Performance Computing (HPC) systems embeds more compute capabilities than the generations before, parallel file systems (P FS) are struggling to keep up with this same trend, increasing the gap bet ween computing power and I/O performance. If new paradigms and technologie s ar...\n\n\nThéo Jolivel (INRIA, Universite de Rennes)\n----------------- ----\nP29 - Perspectives on Teamwork and AI in Scientific Computing\n\nThe development and use of high-quality software—a primary mechanism for sust ained collaboration and progress in scientific computing—is undergoing pro found change, driven by increasing complexity in scientific drivers and co mputing architectures and the rapid adoption of AI, placing dem...\n\n\nOl ivia B. Newton (University of Montana), Anshu Dubey (Argonne National Labo ratory), Denice Ward Hood (University of Illinois Urbana-Champaign), Lois Curfman McInnes (Argonne National Laboratory), and Santiago Ospina Tabares (University of Illinois Urbana-Champaign)\n---------------------\nP31 - P redictive Alerts and Atmospheric Data for Airport Windshear by CPAS 200m W eather Model\n\nLow-level wind shear (LLWS) at Hong Kong International Air port (HKIA) poses significant aviation risks, primarily driven by terrain- induced turbulence from Lantau Island and sea-breeze interactions. While c urrent mitigation relies on real-time detection and short-term forecasting , for operational pu...\n\n\nSai Lun Tin, Chi Chiu Cheung, Ka Ki Ng, and W ai Pang Sze (ClusterTech Limited)\n---------------------\nP32 - Profile-Gu ided-Optimisation of Lattice QCD Contractions on CPU and GPU\n\nWe present a performance optimisation study of the 2+2 disconnected component bottle neck of Lattice QCD computation of the hadronic light-by-light contributio n to the muon's anomalous magnetic moment. Optimization on CPU and GPU arc hitectures were guided by popular profiling tools perf, valgrind, ns...\n\ n\nJingJing Li, Urs Wenger, and Roman Gruber (University of Bern)\n------- --------------\nP35 - Shedding Light on the Solar Dynamo Using Bayesian Da ta Science\n\nSolar magnetic activity exhibits an approximately 11-year cy cle that is far from stable, showing strong long-term variability and recu rrent episodes of strongly reduced activity known as Grand Minima. Underst anding the origin of these features remains a fundamental challenge in sol ar physics and is ...\n\n\nSimone Ulzega (Zurich University of Applied Sci ences, Institute of Computational Life Sciences) and Carlo Albert (Swiss F ederal Institute of Aquatic Science and Technology (EAWAG))\n------------- --------\nP33 - RDQ: A Zero-Copy Remote Data Queue for In-Situ Machine Lea rning in HPC\n\nThe integration of machine learning into the computational sciences is increasingly pursued to reduce time-to-solution, alleviate I/ O bottlenecks, and enable adaptive analysis during simulation. We present RDQ (Remote Data Queue), a library for coupling HPC simulations and machin e learning training ...\n\n\nMaximilian Sander (TU Dresden) and Jens Domke (RIKEN)\n---------------------\nACMP14 - Stability and Accuracy of the r² SCAN Functional for Group-IV Elemental Solids in a Cost-Aware Workflow Per spective\n\nDensity-functional theory (DFT) is a central tool in computati onal materials science, and its predictive power depends critically on the choice of exchange-correlation functional. Here we benchmark the meta-GGA r²SCAN in a study of the group-IV elemental solids C, Si, Ge, and Sn, foc using on stabili...\n\n\nAdonis Haxhijaj (EPFL, ETH Zurich)\n------------- --------\nP38 - Tracking Mechanistic Evolution Across Brain Tissues and Ce ll Types Using Multiplex Networks\n\nTracking the evolution of biological function remains a major challenge in computational biology, as existing a pproaches are often limited to sequence conservation, gene presence, or pr edefined pathways. These methods can fail to identify conserved functional mechanisms even though constituent genes...\n\n\nKenneth Smith (Oak Ridge National Laboratory); Matthew Lane (University of Tennessee); Alice Towns end and Jean Merlet (Oak Ridge National Laboratory, University of Tennesse e); Anna Vlot and Alana Wells (Oak Ridge National Laboratory); and Daniel Jacobson (Oak Ridge National Laboratory, University of Tennessee)\n------- --------------\nACMP03 - Deep Reinforcement Learning for Algorithm Selecti on in Molecular Dynamics Simulations\n\nShort-range particle simulations a re crucial in physics and chemistry, requiring efficient neighbor search a nd interaction algorithms. The C++ library AutoPas supports more than 100 algorithms, but no single algorithm is universally optimal. Traditional ma nual tuning is impractical due to dynamic si...\n\n\nPatrick Metscher (Tec hnische Universität München)\n---------------------\nACMP13 - Solver-Integ rated Lossy and Lossless Compression for Scalable Flow Simulations\n\nLarg e-scale computational fluid dynamics (CFD) simulations on modern GPU-accel erated supercomputers generate terabytes of data per run, making checkpoin ting, storage, and post-hoc analysis increasingly I/O-bound. This bottlene ck limits data retention and hinders downstream workflows such as visualiz ...\n\n\nViral Sudip Shah (University of Illinois Urbana-Champaign)\n----- ----------------\nP37 - Structured Reinforcement Learning for Loop Transfo rmation in MLIR\n\nOptimizing polyhedral kernels for modern multicore arch itectures is a high-dimensional, non-convex problem where small structural changes often yield orders-of-magnitude runtime variation. While traditio nal compilers rely on rigid static heuristics and autotuners require prohi bitive search times, th...\n\n\nAbrar Hossain (University of Toledo)\n---- -----------------\nACMP10 - Mixed Precision Acceleration of Light Matter D ynamics: Enabling HPC Co-Design for Quantum Material Discovery\n\nLight-ma tter dynamics in topological quantum materials holds the promise for a sus tainable society with ubiquitous and power-hungry artificial intelligence (AI) by enabling ultralow-power (attojoule) and ultrafast (petahertz) comp uting and sensing devices. A challenge is simulating multiple field an...\ n\n\nTaufeq Mohammed Razakh (University of Southern California)\n--------- ------------\nACMP05 - Hypergraph Partitioning for Sparse Matrix Reorderin g\n\nFill-in during sparse matrix factorization remains a critical bottlen eck in scientific computing. We present an efficient hypergraph partitioni ng approach for sparse matrix reordering based on the Clique-Node Hypergra ph (CNH) representation, building on prior work by Çatalyurek et al. and S elvitopi ...\n\n\nRitvik Ranjan (ETH Zurich)\n---------------------\nP34 - Scaling Linear Algebra: Eigenvalue Solvers and Performance Trends on Cont emporary HPC Systems\n\nLarge-scale eigenvalue problems are a fundamental component of modern scientific simulations in fields such as materials sci ence, computational chemistry, and theoretical physics, where they often r epresent a dominant computational bottleneck. The efficiency and scalabili ty of linear algebra and eig...\n\n\nMaria Montagna, Sergio Orlandini, and Fabio Affinito (CINECA)\n---------------------\nACMP09 - Medulla: Cluster - and Application-level I/O Performance Diagnosis with LLMs\n\nModern High -Performance Computing (HPC) I/O systems offer immense bandwidth but are n otoriously difficult to utilize effectively.\nIdentifying critical bottlen ecks is a challenge that persists for both\ndomain scientists and cluster administrators. Existing analysis tools\ncan identify common bottlenec...\ n\n\nAnish Sathyanarayanan (BITS Pilani K K Birla Goa Campus)\n----------- ----------\nACMP06 - Improving Deep Learning Based Seismic Inversion with Online Augmentation\n\nThis study investigates the application of data aug mentation techniques to optimize Deep Learning-based Seismic Inversion (DL I), aiming to overcome the scarcity of labeled datasets in the industry. U sing the OpenFWI benchmark, the study evaluates four incremental strategie s: horizontal flipping, syn...\n\n\nLucas Souza (Federal University of São Carlos)\n---------------------\nACMP12 - Semantic-Aware Implicit Neural C ompression for Physics Simulations\n\nMachine learning surrogates and data -driven scientific discovery require efficient access to simulation data, yet physics simulations generate terabyte-scale datasets. Traditional comp ression either achieves insufficient ratios or corrupts physics-critical f eatures like conservation laws. Implicit n...\n\n\nJessica Ezemba (Carnegi e Mellon University)\n---------------------\nACMP04 - Fine-Tuning Large La nguage Models for HPO-Term Recognition\n\nLarge language models (LLMs) off er a promising approach for extracting structured medical information from free-text clinical notes. This work investigates fine-tuning LLMs for Hum an Phenotype Ontology (HPO) term recognition, a core task in clinical gene tics that requires accurate identification and...\n\n\nSrinithi Krishnamoo rthy (Cornell University)\n---------------------\nP39 - Tuning the Perform ance of Three-Body Interactions in Molecular Dynamics\n\nMolecular Dynamic s (MD) simulations predict thermophysical properties, yet standard pair po tentials can lack the desired accuracy for certain applications. Introduci ng three-body potentials, such as the Axilrod-Teller-Muto model, improves results but poses significant computational challenges. This ...\n\n\nMark us Mühlhäußer, Samuel James Newcome, Fabio Gratl-Gaßner, Manish Kumar Mish ra, and Hans-Joachim Bungartz (Technical University of Munich)\n---------- -----------\nACMP08 - Matrix-Free vs. Matrix-Based Finite Element Solvers for 3D Advection–Diffusion–Reaction Equations\n\nHigh-order finite element methods are attractive for three-dimensional advection–diffusion–reaction (ADR) problems, but their efficiency on distributed-memory systems is oft en limited by memory traffic and communication in large linear solves. Thi s work compares matrix-based (MAT) and ma...\n\n\nZhaohui Song (Politecnic o di Milano)\n---------------------\nACMP01 - Adaptive Multidimensional Qu adrature on Multi-GPU Systems\n\nThe present work introduces a distributed adaptive deterministic quadrature framework for high-dimensional integrat ion on multi-GPU systems, enabling accurate evaluation of integrals arisin g in applications such as radiative transfer and probabilistic design. The method is formulated as a hierarchic...\n\n\nMelanie Tonarelli (Universit à della Svizzera Italiana)\n---------------------\nACMP02 - Data Augmentat ion to Improve the Performance of Deep Learning-Based Seismic Inversion\n\ nDeep Learning-based Seismic Inversion (DLI) is a promising alternative to Full Waveform Inversion (FWI), but struggles with data scarcity. This wor k evaluates data augmentation techniques adapted from computer vision to a ddress this bottleneck, and demonstrates that generating synthetic earth m odels...\n\n\nCarlos Gomes de Carvalho Junior (Universidade Federal de São Carlos)\n---------------------\nP28 - Performance-Portable and Highly Sca lable Spectral Transforms with ecTrans\n\nThe continued increase in the sk ill of weather forecasts observed over the past decades depends crucially on the efficient exploitation of the next generation of high-performance c omputers by Earth-system models. The European Centre for Medium-Range Weat her Forecast (ECMWF)'s model, the IFS, is one ...\n\n\nSam Hatfield (ECMWF )\n---------------------\nACMP07 - Mapping the Productivity-To-Energy Trad e-Off in Memory-Bound HPC via DVFS, Core Scaling, and C-State Control on R epurposed Hardware\n\nHigh-performance computing (HPC) systems in resource -constrained environments, such as Africa, often rely on repurposed hardwa re, shifting the primary financial burden from capital expenditure to oper ational energy costs. This work aims to identify realistic and reproducibl e performance–energy...\n\n\nSuné Toerien (University of the Witwatersrand ), Vele Nefale (UniverUniversity of the Witwatersrand), and Ntandoyenkosi Memela and Mubeen Dewan (University of the Witwatersrand)\n\nSession Chair : Tobias Hodel (University of Bern, Switzerland) END:VEVENT END:VCALENDAR