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:20260422T000623Z LOCATION:501-502 DTSTART;TZID=America/Denver:20231113T090000 DTEND;TZID=America/Denver:20231113T173000 UID:submissions.supercomputing.org_SC23_sess440@linklings.com SUMMARY:Workshop on Artificial Intelligence and Machine Learning for Scien tific Applications (AI4S) DESCRIPTION:AI4S – Invited Talk\n\nAnimashree (Anima) Anandkumar (Californ ia Institute of Technology)\n---------------------\nElastic Deep Learning through Resilient Collective Operations\n\nA robust solution that incorpor ates fault tolerance and elastic scaling capabilities for distributed deep learning. Taking advantage of MPI resilient capabilities, aka. User-Level Failure Mitigation (ULFM), this novel approach promotes efficient and lig htweight failure management and encourages smoo...\n\n\nJiali Li, George B osilca, and Aurelien Bouteiller (University of Tennessee) and Bogdan Nicol ae (Argonne National Laboratory (ANL))\n---------------------\nAcceleratin g Particle and Fluid Simulations with Differentiable and Interpretable Gra ph Networks for Solving Forward and Inverse Problems\n\nWe leverage physic s-embedded differentiable graph network simulators (GNS) to accelerate par ticulate and fluid simulations to solve forward and inverse problems. GNS represents the domain as a graph with particles as nodes and learned inter actions as edges, improving generalization to new environmen...\n\n\nKrish na Kumar (University of Texas System) and Yonjin Choi (University of Texas )\n---------------------\nEntropy-Driven Optimal Sub-Sampling of Fluid Dyn amics for Developing Machine-Learned Surrogates\n\nOptimal sub-sampling of large datasets from fluid dynamics simulations is essential for training reduced-order machine learned models. A method using Shannon entropy was d eveloped to weight flow features according to their level of information c ontent, such that the most informative features can be ...\n\n\nWesley Bre wer (Oak Ridge National Laboratory (ORNL)); Daniel Martinez (Science and T echnology Corporation); Muralikrishnan Gopalakrishnan Meena, Aditya Kashi, Katarzyna Borowiec, and Siyan Liu (Oak Ridge National Laboratory (ORNL)); and Christopher Pilmaier, Greg Burgreen, and Shanti Bhushan (Mississippi State University)\n---------------------\nToward Foundation Models for Mat erials Science: The Open MatSci ML Toolkit\n\nArtificial intelligence and machine learning have shown great promise in their ability to accelerate novel materials discovery. As researchers and domain scientists seek to un ify and consolidate chemical knowledge, the case for models with potential to generalize across different tasks within materi...\n\n\nKin Long Kelvi n Lee (Intel Corporation), Carmelo Gonzales (Intel Labs), Matthew Spelling s (Vector Institute), Mikhail Galkin and Santiago Miret (Intel Labs), and Nalini Kumar (Intel Corporation)\n---------------------\nToward Rapid Auto nomous Electron Microscopy with Active Meta-Learning\n\nIn this work, we d eveloped a method to accelerate computational steering of microscopy exper iments by active meta-learning. Before this work, a tailored AI model was trained specifically for every experiment by active learning to reconstruc t spectrum and uncover regions of interest by sampling just ...\n\n\nGayat hri Saranathan, Martin Foltin, and Aalap Tripathy (Hewlett Packard Enterpr ise (HPE)); Maxim Ziatdinov (Oak Ridge National Laboratory (ORNL)); Ann Ma ry Justine Koomthanam and Suparna Bhattacharya (Hewlett Packard Enterprise (HPE)); Ayana Ghosh and Kevin Roccapriore (Oak Ridge National Laboratory (ORNL)); and Sreenivas Rangan Sukumar and Paolo Faraboschi (Hewlett Packar d Enterprise (HPE))\n---------------------\nMachine Learning Applied to Si ngle-Molecule Activity Prediction\n\nCatalytic processes are used in about 1/3 of US manufacturing, from the field of chemical engineering to renewa ble energy. Assessing the activity of single-molecules, or individual mole cules, is necessary to the development of efficient catalysts. Their heter ogeneity structure leads to particle-spec...\n\n\nKendric Hood and Qiang G uan (Kent State University)\n---------------------\nTournament-Based Pretr aining to Accelerate Federated Learning\n\nAdvances in hardware, prolifera tion of compute at the edge, and data creation at unprecedented scales hav e made federated learning (FL) necessary for the next leap forward in perv asive machine learning. For privacy and network reasons, large volumes of data remain stranded on endpoints located in ge...\n\n\nMatt Baughman (Uni versity of Chicago); Nathaniel Hudson (University of Chicago, Argonne Nati onal Laboratory (ANL)); Ryan Chard (Argonne National Laboratory (ANL)); An dre Bauer (University of Chicago); Ian Foster (Argonne National Laboratory (ANL)); and Kyle Chard (University of Chicago, Argonne National Laborator y (ANL))\n---------------------\nAutotuning Apache TVM-Based Scientific Ap plications Using Bayesian Optimization\n\nApache TVM (Tensor Virtual Machi ne), an open source machine learning compiler framework designed to optimi ze computations across various hardware platforms, provides an opportunity to improve the performance of dense matrix factorizations such as LU (Low er Upper) decomposition and Cholesky decomposi...\n\n\nXingfu Wu (Argonne National Laboratory (ANL), University of Chicago); Praveen Paramasivam (Un iversity of South Dakota); and Valerie Taylor (Argonne National Laboratory (ANL))\n---------------------\nAI4S – Afternoon Break\n------------------ ---\nEnhancing Heterogeneous Federated Learning with Knowledge Extraction and Multi-Model Fusion\n\nConcerned with user data privacy, this paper pre sents a new federated learning (FL) method that trains machine learning mo dels on edge devices without accessing sensitive data. Traditional FL meth ods, although privacy-protective, fail to manage model heterogeneity and i ncur high communication costs ...\n\n\nDuy Phuong Nguyen and Sixing Yu (Io wa State University), J. Pablo Muñoz (Intel Corporation), and Ali Jannesar i (Iowa State University)\n---------------------\nA Comparison of Mesh-Fre e Differentiable Programming and Data-Driven Strategies for Optimal Contro l under PDE Constraints\n\nThe field of Optimal Control under Partial Diff erential Equations (PDE) constraints is rapidly changing under the influen ce of Deep Learning and the accompanying automatic differentiation librari es. Novel techniques like Physics-Informed Neural Networks (PINNs) and Dif ferentiable Programming (DP) ar...\n\n\nRoussel Desmond Nzoyem Ngueguin, D avid A.W. Barton, and Tom Deakin (University of Bristol)\n---------------- -----\nProtein Generation via Genome-Scale Language Models with Bio-Physic al Scoring\n\nLarge language models (LLMs) trained on vast biological data sets can learn biological motifs and correlations across the evolutionary landscape of natural proteins. LLMs can then be used for de novo design of novel proteins with specific structures, functions, and physicochemical p roperties. We empl...\n\n\nGautham Dharuman, Logan Ward, Heng Ma, and Priy anka V. Setty (Argonne National Laboratory (ANL)); Ozan Gokdemir (Universi ty of Chicago); Sam Foreman, Murali Emani, Kyle Hippe, and Alexander Brace (Argonne National Laboratory (ANL)); Kristopher Keipert and Thomas Gibbs (NVIDIA Corporation); Ian Foster (Argonne National Laboratory (ANL)); Anim a Anandkumar (California Institute of Technology); and Venkatram Vishwanat h and Arvind Ramanathan (Argonne National Laboratory (ANL))\n------------- --------\nTencoder: Tensor-Product Encoder-Decoder Architecture for Predi cting Solutions of PDEs with Variable Boundary Data\n\nIt is widely hoped that artificial intelligence will boost data-driven surrogate models in sc ience and engineering. However, fundamental spatial aspects of AI surrogat e models remain under-studied. We investigate the ability of neural-networ k surrogate models to predict solutions to PDEs under varia...\n\n\nAditya Kashi (Oak Ridge National Laboratory (ORNL))\n---------------------\nAI4S – Morning Break\n---------------------\nEnabling Performant Thermal Condu ctivity Modeling with DeePMD and LAMMPS on CPUs\n\nThe ability to retain D FT-level accuracy and reduce the high computational costs has been made po ssible using Deep Potential models which allow accurate prediction of inte ratomic force and energy distributions, when trained on DFT data. DeePMD-k it is a Python/C++ package which implements such a mode...\n\n\nNariman Pi roozan and Nalini Kumar (Intel Corporation)\n---------------------\nAI4S – Keynote\n\nRick Stevens (Argonne National Laboratory (ANL))\n------------ ---------\nAI4S – Lunch Break\n\nTag: Artificial Intelligence/Machine Lear ning\n\nRegistration Category: Workshop Reg Pass\n\nSession Chairs: Murali Emani (Argonne National Laboratory (ANL)); Gokcen Kestor (Barcelona Super computing Center (BSC); University of California, Merced); and Dong Li (Un iversity of California, Merced) END:VEVENT END:VCALENDAR