AI and ML Applications in Science
AI and ML accelerate research by helping us find solutions in vast problem spaces, contributing to tools that allow us to build upon past findings, and allowing us to see patterns that are challenging with simpler statistics or the mere human eye. We partner with organizations in research, academia, and industry to further discoveries across various domains.
AI Accelerating Research
We aim to explore applied machine learning in chemistry, materials science, physics, and environmental science in order make meaningful contributions across scientific disciplines and in our daily lives. Some applications include image classification for novel lithium battery development, using LLM's in materials science and chemistry, and small molecule property prediction.
Our Projects
Machine Learning for Grid Energy Storage
Our team is engaged with the Department of Energy’s (DOE) Rapid Operation Validation Initiative (ROVI) to hasten the expansion of energy storage systems with data and machine learning technologies. We and our partners DOE laboratories are building tools which will enhance new deployments of grid-storage batteries with tools that help their operators learn to better operate and maintain new technologies. Our team will be developing AI tools to learn from a swell of data from the grid, integrating them with advanced modeling techniques, and backing those tools with the advanced data and compute infrastructure available through the DOE.
Understanding Nuclear Reactor Degradation through Nano-scale Videos
Materials in nuclear reactors degrade as radiation creates nanoscale voids in the materials - a process we can see with Transmission Electron Microscopy (TEM) but cannot yet understand. The Accelerate team is part of a multi-year effort led out of the Intermediate Voltage Electron Microscope (IVEM) Facility at Argonne to turn videos of void formation into knowledge and models that explain when voids form. Our team has built several computer vision tools which detect voids and is now employing them to track how individual nanovoids interact.
Intelligent Computational Campaigns with Colmena
Getting the most out of a supercomputer requires more than just running massive numbers of computations at once; one must also allow each computation to interact. Our team has created Colmena to simplify bringing such cooperation by leveraging AIs to guide campaigns of otherwise disconnected tasks. Colmena lets scientists express sophisticated patterns for how the AI is used and trained as simple Python functions then deploy those policies on massive supercomputers or across multiple computing systems. Colmena lets our partners design molecules or proteins faster, and ties in well with Globus’s compute and data tools.
Removing Humans from Automated Laboratories
Robots and other types of automation have big shoes to fill if they are to replace humans in some laboratories. Our team has been contributing in many small but important ways to extend the capabilities of robots. For instance, we have created tools which analyze combustion experiments then plan the next one with Bayesian learning and are finding which chemicals degrade faster by sifting through weeks of NMR data. The Accelerate team works to find ways, subtle or stately, to turn buildings full of disconnected instruments into independent partners for human scientists.