A report that used to take infectious disease experts several hours to compose now gets done in roughly two minutes. That's the headline from Boston University's BEACON AI pipeline, one of over 700 projects funded through the NSF's National Artificial Intelligence Research Resource (NAIRR) pilot program, supported by NVIDIA's DGX infrastructure.
BEACON Shrinks Outbreak Report Generation from Hours to Minutes
Boston University's Hariri Institute for Computing and the Center on Emerging Infectious Diseases trained a large language model on a huge corpus of infectious disease documents and epidemic-prone priority pathogens. The model, deployed on NVIDIA accelerated compute, ingests signals from HealthMap, news, social media, and community boards to generate concise outbreak reports. Ioannis Paschalidis, director of the Hariri Institute, says the pipeline cut report production time from several hours to two minutes. Internationally deployed doctors, government organizations, and academic researchers are already using BEACON to identify and treat infectious diseases faster.
Polymathic AI Open-Sources Walrus for Fluidlike Simulations
Polymathic AI, a coalition from Flatiron Institute, Cambridge University, and Lawrence Berkeley National Lab, leveraged NVIDIA GPUs and NVLink interconnect to build the "Well" dataset and train Walrus, the largest and most broadly applicable foundation model for fluidlike behavior. Walrus, along with its data, code, and pretrained weights, has been made publicly available. The team plans to explore scaling laws to accelerate development of even more powerful scientific foundation models. Simulation-to-real pipelines in industries like healthcare, agriculture, and energy should benefit directly.
University of Michigan's MIST Fuses Molecular AI with General-Purpose LLMs
Professor Venkat Viswanathan's group at the University of Michigan developed MIST (Molecular Insight SMILES Transformers), a family of molecular foundation models pretrained on large unlabeled datasets using a novel tokenizer called Smirk. MIST captures nuclear, electronic, geometric, isotopic, and stereochemical information from molecular representations. After fine-tuning on more than 400 structure-property relationships, MIST matches or exceeds state-of-the-art across benchmarks in electrochemistry, quantum chemistry, and physiology. The team trained on a 40-GPU NVIDIA DGX cluster allocated through NAIRR plus an additional 200,000 GPU hours on ALCF's Polaris cluster, using NVIDIA's NGC PyTorch container for reproducibility. Fusing MIST with general-purpose LLMs makes quantum-chemical calculations broadly accessible and accelerates design of energy storage and conversion systems for heavy-duty transportation and aviation electrification.
NAIRR's cloud-based resource guarantees researchers dedicated access to at least four NVIDIA DGX nodes for a month, plus technical support. With 700+ projects running, the pilot is demonstrating what focused AI infrastructure can unlock across scientific domains.
Source: NAIRR Science Program Reshapes Scientific Research, Powered by NVIDIA AI Infrastructure
Domain: blogs.nvidia.com
Comments load interactively on the live page.