When frogs in the rainforest go “ribbit,” Ankita Shukla listens.
The Computer Science & Engineering assistant professor developed an artificial intelligence (AI) learning model that can identify different frog species in the Brazilian rainforest by the sounds they make. This work helped Limelight Rainforest, a multi-institutional academic team, win the $10 million XPRIZE Rainforest competition in November.
“It was amazing,” Shukla said, regarding the win. “It felt great to see the dedication and hard work the team put in getting recognized. Personally, I find it very motivating to work on developing AI-based solutions that could assist in eco-conservation efforts.”
The win, announced at the G20 Summit in Rio de Janeiro, Brazil, comes with a $5 million first-place prize for Limelight Rainforest, a team led by Colorado Mesa University Biology Professor Thomas Walla.
“She really came through for the team when she helped us develop a supervised model (a type of algorithm trained on labeled data) for amphibian detection,” Walla said. “The work was fast paced and she was always working under a deadline but she really came through for us at the end.”
The Limelight team — consisting of about 60 engineers, scientists and indigenous people — also included 推荐杏吧原创 doctoral candidate Ari Grele from the College of Science.
Shukla and Grele both were on Limelight’s sub-team focused on machine learning — a type of artificial intelligence that learns and improves from data without direct programming — but each worked on different aspects of the competition, and had different experiences.
And for Shukla, the competition experience was remote and took place during the dark of night.
Birds have many moods
The XPRIZE Rainforest project launched in 2019 with a goal to develop novel technologies to survey biodiversity and produce information to inform conservation efforts. Basically, teams each were tasked with developing an autonomous biodiversity sampling device that would be tested in competition.
At that time, Shukla was still earning her doctorate at the Indraprastha Institute of Information Technology Delhi (IIIT-Delhi) in New Delhi, India. Her intrinsic love of animals, coupled with her expertise in machine learning, led her to work on projects assessing endangered and vulnerable animal populations such as tigers and beluga whales.
Shukla joined the Rainforest team — formerly called Team Waponi — in 2022 while working at Arizona State University as a postdoctoral researcher for Professor Pavan Turaga.
Turaga, who also was on the team, got Shukla involved.
Her work for the team initially consisted of collecting and identifying bird sounds, converting that data into a format that could be processed by a machine learning model and then building and training the model to identify different birds by the sounds they make. This is not an easy task.
Differences in sounds made by different bird groups can be subtle. Even within a single group, there can be different calls to indicate different moods.
“I didn’t know they had so many moods,” she laughed.
Her model identifying bird sounds was tested in the 2023 semifinal competition at a rainforest preserve in Singapore. Basically, teams had 24 hours to collect data in the rainforest via the autonomous systems they built, followed by 48 hours of data analysis to identify flora and fauna. Out of the 13 teams to compete in the semifinals, Shukla’s team was one of the six to qualify for the finals in Brazil.
Did she celebrate in the famous Southeast Asian metropolis? Nope — because of her exchange scholar visa status and concerns about re-entering the United States after the competition, Shukla participated from home. Teammates sent her the audio files that were collected, and she spent the next 48 hours running the data through her AI model and preparing the results.
It was an experience she repeated this past summer, during the final competition in Rio de Janeiro.
Night shift
In 2024, Shukla joined the faculty in the University’s Computer Science & Engineering Department, but continued her work for the Limelight Rainforest team, which was headed for the competition finals in July in Brazil. She built another AI model trained to recognize the sounds of Brazilian frogs, based in part on frog sound data from Anuraset, an openly accessible dataset created by researchers primarily from Brazilian institutions. She was ready to identify some frogs.
“I was very much excited,” she said.
Because Rio de Janeiro is about five hours ahead of Reno, when the data-collecting portion of the competition closed for the evening in Brazil, it was it was the middle of the night in Nevada.
“I was here in the night, working, at 2 a.m.,” she said. “I’m waiting for the data coming in. I tied my hair back, and it was like, alright, it’s my time to work.”
Again, teammates shared their collected data in batches over a Google drive while she ran the AI model.
“I was available for the time, and was checking and in communication with the team on the ground about my findings,” she said.
AI for science, AI for social good
Shukla said she and her teammates will get some of the prize money, but her work is driven mainly by two concepts: AI for science and AI for social good. Both concepts refer to the use of AI to advance science and to address pressing societal challenges, respectively.
“I like to work in these fields, because I find them important,” Shukla said.
She plans to continue working with Walla, the Limelight Rainforest team leader from Colorado Mesa University.
“He has interesting ideas,” she said, “and my passion for conservation is not going anywhere.”
