image: Researchers led by Noah Cowan at Johns Hopkins University have secured NIH funding to probe how animals alternate between "explore" (sensing) and "exploit" (task-oriented) behaviors in uncertain environments, using the weakly electric glass knifefish (pictured) as a model. The team includes researchers from four universities who will integrate their expertise in neuroscience, math, engineering, and machine learning to build on 2023 findings in Nature Machine Intelligence that revealed the explore/exploit pattern across species from amoebas to humans. The project aims to decode decision triggers, with implications for robotics and medicine.
Credit: Noah Cowan
A glass knifefish darts back and forth in a short tube, its brain activity being recorded in real time. This small fish, alternating between swift bursts of sensing activity and slower, task-driven behaviors, is helping scientists understand how animals decide when to gather information about their environment versus act on it. A team of researchers is blending neuroscience, math, and engineering to decode these choices, with potential to guide robots in uncertain terrains or unlock secrets of the brain.
The team’s research has just been funded by the Collaborative Research in Computational Neuroscience (CRCNS) program—a joint initiative of the National Institutes of Health (NIH) and the National Science Foundation (NSF) that supports interdisciplinary research.
The CRCNS program emphasizes collaborative efforts to advance understanding of nervous system functions through computational tools. With the lead investigator at Johns Hopkins University and additional collaborators at the University of Maryland, Baltimore County (UMBC), New Jersey Institute of Technology (NJIT), and the University of Minnesota, the team for the newly funded project spans biology, engineering, mathematics, and computer science—a mix well-positioned to discover deeper insights into brain mechanisms.
‘Explore’ or ‘exploit’?
The new project builds on the same team’s prior research, published in 2023 in Nature Machine Intelligence, which revealed similar decision-making patterns across species, from amoebas to humans. In that work, the team analyzed the behavior of glass knifefish—weakly electric fish that navigate dark waters using self-generated electric fields—in experiments run by Noah Cowan, the lead investigator for the new grant. Then they compared their findings to the behavior of other species as described in the scientific literature, uncovering similar patterns in 11 species, including bats, mice, moths, and humans.
In the prior work, “We looked at velocity distributions, and we found that there were two modes of movement. We called them ‘explore’ and ‘exploit,’ but you could also describe them as ‘fast’ and ‘slow,’" explains Kathleen Hoffman, professor of mathematics and statistics at UMBC and a co-lead on the new grant. During experiments in narrow tubes, the fish alternated between two modes: rapid, exploratory movements to sense their surroundings ("explore") and slower, deliberate actions using the information they’d collected ("exploit").
That research challenged robotics norms, showing that animals don't constantly scan their environment, but rather burst into action when needed, a strategy the team showed is both more economical and more effective. The new project ramps up data collection—from 40 seconds per trial to 10 minutes—allowing the team to reveal subtler patterns, like burst lengths and correlations between the fish’s movement mode and its position in the tube.
Deciphering animal decisions
A primary goal is to uncover what prompts the mode switch. "How does it decide when to switch? And the hypothesis that we're considering is that it's based on some internal measure of uncertainty in the fish, meaning that if the fish isn't sure if it's inside the tube, it's going to move so it can gather sensory information," Hoffman says.
To test this, the team integrates several methods. At the University of Minnesota, engineers led by Andrew Lamperski will apply machine learning to map relationships between sensory inputs and behavioral outputs in the form of mathematical functions. Hoffman handles data analysis, starting with manual pattern-spotting before coding.
“I can't wait to get my hands on the data,” Hoffman says. She’ll start by simply printing out the velocity and position results and poring over them visually. “I don't think there's anything better than the human brain to see patterns, and mathematics is the study of patterns,” she adds. After observing what looks like a pattern, she’ll bounce her ideas off the rest of the team, and eventually “go write a program to automatically go through all the data and see if that pattern recurs.”
A boon for the project comes from NJIT, where biologist Eric Fortune will record neural activity via electrodes inserted into the fish's brains during the movement experiments—a technique unavailable in prior work. This will let the team compare brain signals with behavior in real time, and look for an underlying mechanism that drives the switch from “explore” to “exploit.”
A scientific ‘dream team’
This project's power lies in its teamwork. Hoffman coordinates from UMBC, analyzing data from all the collaborators. Cowan oversees behavioral tests on fish without brain probes, which allows for more complex experimental setups. Fortune at NJIT is handling the neural recordings, while Lamperski at Minnesota focuses on machine learning models that reflect what the others are seeing in the lab.
"What I love about this project is that all the components are necessary to elucidate the mechanism,” Hoffman reflects. “Nobody could do this completely on their own."
“I’m excited to have this dream team of mathematicians, engineers, and neuroscientists to assemble behind this problem,” Cowan said. “My lab at Hopkins has struggled to make sense of these movements for over a decade. This new team puts us on a path to finally decode the neural mechanisms animals use to switch gears between gathering task information, on the one hand, and getting the task done, on the other.”
‘My favorite kind of science’
This research could eventually transform robotics.
“If you want to build a robot that is going to mimic the motion of animals that exhibit this explore/exploit pattern for incorporating sensory information, you have to know how the animals do it,” Hoffman says. “This grant is focused on figuring out what that mechanism is.”
A robot that mimics natural intermittent sensing might navigate uncertain spaces, like disaster zones, more efficiently than constant-scanning models. The shared explore-exploit pattern also suggests broader relevance for the research, potentially informing understanding of neurological disorders—though Hoffman stresses those possibilities are further down the road.
The grant will also open doors for students: Hoffman plans to involve undergraduates in data visualization and analysis, offering hands-on experience in interdisciplinary research that demonstrates how together, diverse minds can unlock secrets of the brain—with ripple effects in tech and health.
“The one thing I'm really excited about in this grant is that it's completely multidisciplinary,” Hoffman says. “Everybody has a different perspective that helps us understand what's going on. This is my favorite kind of science.”