BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Drupal//recurring_events_ical//2.0//EN BEGIN:VEVENT UID:54a3796f-ef9b-4786-9b7e-f6ae60e343bc@support.access-ci.org DTSTAMP:20260331T132652Z DTSTART:20260415T180000Z DTEND:20260415T190000Z SUMMARY:Neocortex Seminar Series: Physics-Aware AI at Scale: Neural Compres sion and Vision Transformers for Simulation Data DESCRIPTION:We are pleased to invite you to a new delivery of the Neocortex Seminar Series, highlighting research and applications enabled by the spe cialized accelerator featured in the NSF-funded Neocortex project, the Cer ebras Wafer Scale Engine. This series showcases user experiences, best pra ctices, and emerging workflows from the community.Talk details:Title: Phys ics-Aware AI at Scale: Neural Compression and Vision Transformers for Simu lation DataSpeaker: Jessica Ezemba, Carnegie Mellon University Date: Apri l 15th, 2026Time: 2:00PM, ESTLocation / Format: ZoomAbstract:Modern scient ific computing generates terabyte-scale simulation data across physics dom ains, yet researchers lack efficient tools for storage, retrieval, and ana lysis. This work presents two complementary approaches to addressing this bottleneck, both leveraging wafer-scale computing on the Cerebras CS-3. Fi rst, SINCPS (Semantic-aware Implicit Neural Compression for Physics Simula tions) compresses physics simulation data by 150× to 25,000× using impli cit neural representations, reducing per-dataset training time to 2 to 3 h ours while preserving physics-critical conservation laws across 22 dataset s from The Well benchmark. Second, PhySiViT, a domain-specific Vision Tran sformer trained on approximately 7 million physics simulation images from The Well in just 22 hours, produces embeddings with distinct physics-infor med structure that outperforms general-purpose models like CLIP and DINOv2 on physics-specific tasks, achieving 43% better temporal forecasting (R² = 0.33 vs. 0.23) and superior physics clustering (silhouette score = 0.23 vs. 0.20). Together, these systems form a pipeline for transforming large simulation archives into compact, queryable representations suitable for downstream machine learning workflows, demonstrating that domain-focused m odels trained efficiently on specialized hardware can outperform general-p urpose counterparts on scientific tasks.Speaker Bio:Jessica Ezemba is a Ph .D. candidate in Mechanical Engineering at Carnegie Mellon University, whe re she is advised by Dr. Christopher McComb and Dr. Conrad Tucker. Her res earch sits at the intersection of artificial intelligence and engineering design, with a focus on making the design process faster and less prone to errors. She is particularly interested in engineering simulations, a crit ical yet time-consuming and expertise-dependent stage of design where prod ucts are tested before manufacturing. Her work investigates how AI can acc elerate simulation interpretation by enabling faster integration of multid isciplinary expertise. Through developing benchmarks, foundation models, a nd surrogate modeling approaches, Jessica has demonstrated that current AI tools struggle to understand engineering simulations, motivating her ongo ing work on alternative paradigms, including agentic workflows that keep h umans in the design loop while enabling automated understanding. She has l everaged wafer-scale computing on the Cerebras CS-3 to develop physics-foc used foundation models and neural compression methods for large-scale simu lation data, including PhySiViT, a domain-specific Vision Transformer for physics simulations, and SINCPS, an implicit neural compression framework achieving up to 25,000× compression on terabyte-scale simulation archives . She has collaborated with industry partners including Ansys, and her res earch has been published in venues including the ASME Journal of Mechanica l Design, and the ACM/IEEE Supercomputing Conference (SC). Registration:R egister here: https://cmu.zoom.us/meeting/register/s6v2-Zu_QxCeM6eq46VR6wA fter registering, you will receive a confirmation email with details on jo ining the session.For questions about this event or the Neocortex Seminar Series, please contact us at neocortex@psc.edu.About the Neocortex Seminar Series:The Neocortex Seminar Series showcases talks from researchers and practitioners leveraging the specialized accelerator featured in the NSF-f unded Neocortex project (the Cerebras Wafer-Scale Engine) to advance AI, m achine learning, and data-intensive workloads. The series shares practical experiences, fosters community engagement, and encourages researchers, en gineers, and members of the academic community to explore and adopt these cutting-edge capabilities.-- The Neocortex team. Links of interest:Neoco rtex Slack User SupportNeocortex PortalCalendly AI Office-hoursCalendly SD K Office-hours URL:https://support.access-ci.org/events/9033 END:VEVENT END:VCALENDAR