Saving seaweed with machine learning

Last year, Charlene Xia '17, SM '20 recovered herself astatine a crossroads. She was finishing up her master's grade successful media arts and sciences from the MIT Media Lab and had conscionable submitted applications to doctoral grade programs. All Xia could bash was beryllium and wait. In the meantime, she narrowed down her vocation options, careless of whether she was accepted to immoderate program.

"I had 2 thoughts: I'm either going to get a PhD to enactment connected a task that protects our planet, oregon I'm going to commencement a restaurant," recalls Xia.

Xia poured implicit her extended cookbook collection, researching planetary cuisines arsenic she anxiously awaited connection astir her postgraduate schoolhouse applications. She adjacent looked into the outgo of a nutrient motortruck licence successful the Boston area. Just arsenic she started hatching plans to unfastened a plant-based skewer restaurant, Xia received connection that she had been accepted into the mechanical engineering postgraduate programme astatine MIT.

Shortly aft starting her doctoral studies, Xia's advisor, Professor David Wallace, approached her with an absorbing opportunity. MathWorks, a bundle institution known for processing the MATLAB computing platform, had announced a caller effect backing programme successful MIT's Department of Mechanical Engineering. The programme encouraged collaborative probe projects focused connected the wellness of the planet.

"I saw this arsenic a super-fun accidental to harvester my passionateness for food, my method expertise successful water engineering, and my involvement successful sustainably helping our planet," says Xia.

Wallace knew Xia would beryllium up to the task of taking an interdisciplinary attack to lick an contented related to the wellness of the planet. "Charlene is simply a singular pupil with bonzer endowment and heavy thoughtfulness. She is beauteous overmuch fearless, embracing challenges successful astir immoderate domain with the well-founded content that, with effort, she volition go a master," says Wallace.

Alongside Wallace and Associate Professor Stefanie Mueller, Xia projected a task to foretell and forestall the dispersed of diseases successful aquaculture. The squad focused connected seaweed farms successful particular.

Already fashionable successful East Asian cuisines, seaweed holds tremendous imaginable arsenic a sustainable nutrient root for the world's ever-growing population. In summation to its nutritive value, seaweed combats assorted biology threats. It helps combat clime alteration by absorbing excess c dioxide successful the atmosphere, and tin besides sorb fertilizer run-off, keeping coasts cleaner.

As with truthful overmuch of marine life, seaweed is threatened by the precise happening it helps mitigate against: clime change. Climate stressors similar lukewarm temperatures oregon minimal sunlight promote the maturation of harmful bacteria specified arsenic ice-ice disease. Within days, full seaweed farms are decimated by unchecked bacterial growth.

"Saving Seaweed with Machine Learning". Credit: MIT

To lick this problem, Xia turned to the microbiota contiguous successful these seaweed farms arsenic a predictive indicator of immoderate menace to the seaweed oregon livestock. "Our task is to make a low-cost instrumentality that tin observe and forestall diseases earlier they impact seaweed oregon livestock by monitoring the microbiome of the environment," says Xia.

The squad pairs aged exertion with the latest successful computing. Using a submersible integer holographic microscope, they instrumentality a 2D image. They past usage a instrumentality learning strategy known arsenic a neural web to person the 2D representation into a practice of the microbiome contiguous successful the 3D environment.

"Using a instrumentality learning network, you tin instrumentality a 2D representation and reconstruct it astir successful existent clip to get an thought of what the microbiome looks similar successful a 3D space," says Xia.

The bundle tin beryllium tally successful a tiny Raspberry Pi that could beryllium attached to the holographic microscope. To fig retired however to pass these information backmost to the probe team, Xia drew upon her master's grade research.

In that work, nether the guidance of Professor Allan Adams and Professor Joseph Paradiso successful the Media Lab, Xia focused connected processing tiny underwater connection devices that tin relay information astir the water backmost to researchers. Rather than the accustomed $4,000, these devices were designed to outgo little than $100, helping little the outgo obstruction for those funny successful uncovering the galore mysteries of our oceans. The connection devices tin beryllium utilized to relay information astir the water situation from the instrumentality learning algorithms.

By combining these low-cost connection devices on with microscopic images and instrumentality learning, Xia hopes to plan a low-cost, real-time monitoring strategy that tin beryllium scaled to screen full seaweed farms.

"It's astir similar having the 'internet of things' underwater," adds Xia. "I'm processing this full underwater camera strategy alongside the wireless connection I developed that tin springiness maine the information portion I'm sitting connected adust land."

Armed with these information astir the microbiome, Xia and her squad tin observe whether oregon not a illness is astir to onslaught and jeopardize seaweed oregon livestock earlier it is excessively late.

While Xia inactive daydreams astir opening a restaurant, she hopes the seaweed task volition punctual radical to rethink however they see nutrient accumulation successful general.

"We should deliberation astir farming and nutrient accumulation successful presumption of the full ecosystem," she says. "My meta-goal for this task would beryllium to get radical to deliberation astir nutrient accumulation successful a much holistic and earthy way."

---

---

This communicative is republished courtesy of MIT News (web.mit.edu/newsoffice/), a fashionable tract that covers quality astir MIT research, innovation and teaching.