The Van Trump Report

Iowa State Researchers Developing AI-Powered Pest Identifier Win “Supercomputing” Time

A way to reliably identify and receive recommended controls for agricultural pests is at the top of nearly every farmer’s wishlist. While there are countless identification apps online, most of the digital bug and weed identifiers are limited in scope, difficult to use, and often unreliable. But with the increasing power of artificial intelligence (AI), this could be about to change. Iowa State University researchers developing large, vision-based artificial intelligence tools to identify agricultural pests are among the first teams to be awarded time on one of the fastest supercomputers in the US.

The award will provide the Iowa State researchers with one million “node hours” of supercomputing time on the $60 million Frontera supercomputer at the Texas Advanced Computing Center at the University of Texas at Austin. Frontera’s power comes from more than 8,000 compute nodes, each containing 56 core processors. It’s the fastest supercomputer based on a US university campus.

The Iowa State researchers will use Frontera to help them train an ensemble of large machine-learning models that can analyze photos to quickly identify agricultural pests, including insects and weeds. The models will be packaged into an app platform, which is designed to be deployed around the world to help farmers protect their crops. The researchers aim to ultimately connect the vision models with large language models in order to provide a conversational tool that can suggest pest-control strategies.

Baskar Ganapathysubramanian, the Joseph and Elizabeth Anderlik Professor in Engineering at Iowa State, is leading the computing project along with Aditya Balu, a data scientist at Iowa State’s Translational AI Center. Ganapathysubramanian said researchers have been working on their approach for about two years with support from two federally funded efforts on campus. Building on preliminary work with InsectNet, a model trained on a large dataset of insect images, the team will utilize an expanded dataset of approximately 100 million images.

According to Ganapathysubramanian, the original idea – conceptualized by Iowa State’s Soynomics Team, a research team founded in 2014 that applies technology and data science to improve agriculture – was to build AI tools to identify diseases in soybeans. The idea evolved to include identification of insects harmful to soybeans, then agricultural pests across Iowa, then agricultural pests around the world.

But as the scope of the project has been slowly broadening, “so have the computing requirements to develop our model,” said Ganapathysubramanian. That’s why the team submitted a proposal for one of the first awards from the national AI resource program. And, Ganapathysubramanian said, ever-growing computing requirements is why the team will make more proposals to the program.

The award comes through the National Artificial Intelligence Research Resource Pilot, a two-year pilot program led by the National Science Foundation (NSF) and the Department of Energy. A total of 35 projects won computational time in this first round. The initial call for applicants was issued in January 2024. In tandem with the announcement of initial awards, the NAIRR Pilot opened the next opportunity for researchers and educators to apply for access to resources that support AI research. Researchers and educators can apply for access to these resources and view descriptions of the first cohort projects HERE.

Leave a Comment

Your email address will not be published. Required fields are marked *