NVIDIA and NAIRR Propel AI Research Forward
The U.S. National Science Foundation’s National Artificial Intelligence Research Resource (NAIRR) pilot program has significantly advanced research efforts across the country, supporting over 700 projects in fields ranging from protein prediction to infectious disease management. Launched two years ago, this initiative has received substantial backing from NVIDIA, which has provided cloud-based resources and technical support to researchers, enabling them to access powerful AI infrastructure.
Transformative Research Through AI Infrastructure
NVIDIA’s contribution to the NAIRR pilot includes access to a minimum of four NVIDIA DGX nodes for at least a month, along with ongoing technical assistance throughout various research projects. This robust infrastructure has allowed researchers to streamline their workflows and uncover innovative technologies poised to impact critical sectors such as healthcare, agriculture, and energy.
The potential for scientific exploration facilitated by NAIRR is immense. Below are highlights from several notable projects that exemplify the program’s reach and capabilities.
Polymathic AI’s Well Dataset Enhances Physical Simulations
In various industries, simulation-to-real pipelines are increasingly utilized as an effective means of deploying technologies safely and cost-efficiently. Polymathic AI—a collaboration involving scientists from institutions like the Flatiron Institute, Cambridge University, and Lawrence Berkeley National Lab—has leveraged NVIDIA GPUs and NVLink interconnect technology to enhance physical fluid-like simulations using a large-scale dataset known as the “Well.”
This dataset aims to train the most comprehensive foundation model for fluid-like behavior yet developed. Named Walrus, this model is publicly accessible along with its data, code, and pre-trained weights. Polymathic AI’s approach builds upon previous advancements in physics pretraining environments while addressing existing limitations in scale and diversity. The team plans to investigate scaling laws further to accelerate the development of more powerful models for scientific applications.
University of Michigan’s Fusion Model for Energy Storage
Energy storage and conversion are critical components of modern society, necessitating innovative materials that can meet future demands. Researchers at the University of Michigan are spearheading a model-fusion framework under the guidance of Professor Venkat Viswanathan from the Department of Aerospace Engineering. This framework integrates domain-specific molecular AI with general-purpose large language models (LLMs) to facilitate exploration within chemical space.
The Molecular Insight SMILES Transformers (MIST) family of molecular foundation models was designed specifically for discovery across chemical landscapes. Pretrained on extensive unlabeled molecular datasets, MIST employs a novel tokenizer called Smirk that captures essential information about molecular representations. Fine-tuned on over 400 structure-property relationships, MIST models demonstrate state-of-the-art performance in various domains including electrochemistry and quantum chemistry.
Developed using a 40-GPU NVIDIA DGX cluster allocated through NAIRR and supplemented by 200,000 GPU hours on ALCF’s Polaris cluster, MIST enables accurate quantum-chemical calculations while expediting the design of energy systems crucial for electrifying transportation sectors such as aviation.
Boston University’s BEACON AI Pipeline for Infectious Disease Detection
The rapid spread of infectious diseases poses significant challenges globally. Boston University’s Hariri Institute for Computing and Center on Emerging Infectious Diseases are working on an AI pipeline named BEACON—Biothreats Emergence, Analysis and Communications Network—to enhance outbreak monitoring capabilities. This initiative involves training an LLM using NVIDIA accelerated computing resources.
The LLM is being developed with a comprehensive corpus focused on infectious diseases and epidemic-prone pathogens to assist field experts in outbreak analysis. BEACON will analyze online content related to emerging disease outbreaks globally, extracting key features for categorization and prioritization purposes. By processing signals from diverse sources—including global disease-tracking platforms like HealthMap—BEACON generates concise reports that can inform clinical guidelines for emerging infections.
Internationally deployed doctors and researchers have already begun utilizing BEACON to quickly identify and respond to infectious diseases. According to Ioannis Paschalidis, director of Boston University’s Hariri Institute, this new pipeline has drastically reduced report generation time from several hours down to approximately two minutes.
Broad Impact Across Institutions
The advancements made possible by NAIRR extend beyond these highlighted projects; numerous universities—including Harvard, Stanford, and Colorado State University—are also leveraging NAIRR resources alongside NVIDIA technology to drive scientific breakthroughs across various fields. The accessibility of AI tools combined with accelerated computing resources is fostering innovation aimed at creating a safer and healthier society.
What This Means
The collaboration between NAIRR and NVIDIA represents a significant leap forward in scientific research capabilities across multiple disciplines. By providing researchers with advanced computational resources, these initiatives not only expedite research timelines but also open new avenues for discovery that could lead to transformative solutions in healthcare, energy storage, environmental science, and more. As access to these technologies continues to grow, the potential for impactful innovations becomes increasingly tangible.
For more information, read the original report here.
