A team of four Malaysians recently won runner-up in an international ecological hackathon, the PlankThon Challenge, launched by a French transnational company in conjunction with World Oceans Day
On 26 and 27 June, a team from 42 Kuala Lumpur (42KL) — a tuition-free computer programming school based in Sunway — took on a 48-hour challenge to create a machine learning programme to identify different types of plankton to be used by scientists at sea.
The competition, organised by French resource management company Veolia and non-profit environmental organisation Tara Ocean Foundation, wanted data science students to come up with a model that could classify 130,000 photos of plankton taken by the foundation's current expedition in the Atlantic Ocean.
Eventually, the best model will be used on board the schooner Tara for future scientific expeditions – capturing images of plankton in water samples and studying their role as the base of the oceanic food chain.
Suresh Chetty, Hans Hazairi, Gates Tan, and Bunyod Shamsiddinov — all coming from different backgrounds around Malaysia — came ahead of 35 teams from around the world
Together, not only did they manage to secure a seat in the Top Eight finals, but they were also the only team to successfully develop a machine learning model with plankton identification accuracy higher than the current system used onboard the schooner.
"The challenge was immense as none of us had any prior exposure to machine learning, and I had never explored Python programming language before. So, we initially just took it as an opportunity for us to speed-learn machine learning models," said 24-year-old Tan, who was a medical graduate prior to picking up programming at 42KL.
Meanwhile, 19-year-old Bunyod, who is one of the youngest students at 42KL, said, "Right after we registered for the event, we studied various machine learning videos and tutorials and we were confident before the hackathon."
"But when it started, we quickly realised that everything that we had studied before was useless and we had to develop something much bigger than just a beginner model. So, there was a lot of going back to the drawing board for us."
However, through their experience of working together, the team managed to come up with an approach to build an algorithm and a way to visualise the data
"Learning about machine learning itself in 48 hours, complete with its jargon and concepts, was the biggest challenge. So, what we did was use a pre-trained model and retrained it by tweaking some parameters," said 26-year-old Hans, a former civil engineer who resigned from his previous role to upskill himself.
Meanwhile, Suresh, a 46-year-old former freelance programmer, said it was a team effort that got them that far: "It was definitely collaborative. You bring what you know to the party, nobody needed to stand out and we saw everything fall into place."
Making it to the finals, the team — named 'HawaiianBobtailGreen7', after their initials — was invited to present their model in front of a panel of judges composed of Veolia and Tara Ocean Foundation employees through a video conference
While the team had the most accurate machine learning model, they lost the top spot due to the energy consumption needed to operate their model in the middle of the ocean with no Internet access.
"They were very transparent with us on where we did not do as well as the top placed team, but we were just surprised that we got this far," said the team.
Coming in second place, the team won EUR6,000 (RM29,840) worth of computer hardware, low-tech equipment, and Raspberry Pi computer accessories.
They just barely missed out on the first place prize of getting invited to Lorient, France to meet the schooner's scientific team when it returns from its two-year expedition on marine microbiome in October 2022.
The four of them are now currently focused on improving their machine-learning model in hopes that it can be applied in other industries
They are most likely looking at agriculture, as food security is a major concern in Malaysia.
"In the future, we are hoping to apply this model and learning to be used for other types of identification, such as the health of our plants and vegetables," said the team, eyeing a possible future project with Sunway FutureX Farms, Malaysia's first urban farming innovation hub.