Title | Description/More Information |
Imperfect Recognition: A Study of OCR Limitations in the Context of Scientific Documents | Chinmay Sahasrabudhe, Yang Ho, Nick Winovich, and Siva Rajamanickam submitted a publication and gave talks at the IEEE International Parallel & Distributed Processing Symposium May 29th, 2025 and the Trillion Parameter Consortium meeting July 31st, 2025. |
Scaling AI for Scientific Discovery: Faster Kernels, Efficient Models, Better Physics | Chinmay Sahasrabudhe from Sandia National Laboratories gave this talk as part of our series, BANYAN Guest Seminars June 16th, 2025. Description: The talk discussed common limitations discovered in evaluations of prominent optical character recognition (OCR) tools in the context of scientific document analysis, as well as automatic evaluation of extraction output and document elements to guide banyan-ingest, a general-purpose document processing tool. |
Data Flow and GenAI Research: Convergence of HPC and AI Simulations | Siva Rajamanickam from Sandia National Laboratories gave this invited talk at Oak Ridge National Laboratory July 15th, 2025. |
Banyan-ingest – Challenges of Processing Documents for LLMs | Yang Ho, from Sandia National Laboratories gave this talk at Sandia 2025 Machine Learning/Deep Learning Workshop August 14th,2025. Description: The talk went over banyan-ingest the tool as well as highlighting document processing results. |
Using Open Source and Proprietary models for Parallel Code Generation | Siva Rajamanickam from Sandia National Laboratories gave this talk at the 2024 Chesapeake Large-Scale Analytics Conference. |
Artificial Intelligence Safety Institute Red Teaming Exercise | A joint Sandia and Lawrence Livermore team successfully performed an artificial intelligence large language model red teaming pilot. The pilot focused on biosafety with the AI Safety Institute of the Department of Commerce, as requested in an October 2024 National Security Memo. |
Patent submitted 2/7/2025 | Large language models (LLMs) are prone to producing incorrect information, which makes them difficult to use for high consequence decision making. The team developed methods that use knowledge graphs to provide LLMs with trusted information and to automatically check that the answers that the LLMs produce are correct. |
Computing Platform Engineering: Creating a Converged HPC & Cloud Computing Environment for the Future | Kevin Pedretti from Sandia National Laboratories gave this invited talk at the CEA NNSA Computational Science Collaboration Meeting May 5th-8th, 2025. |