Undersea intelligence

As part of our annual Global Data Science Challenge, 乌鸦传媒 colleagues helped the Lofoten-Vester氓len (LoVe) Ocean Observatory in Norway to identify anomalous data recorded by its ocean-sensing equipment. The winning teams explain how they used AI to answer this call to action.

Deep underwater, off the coast of Norway鈥檚 Lofoten archipelago, a scientific sensor array listens and records. For 24 hours a day, seven days a week, the Lofoten-Vester氓len (LoVe) Ocean Observatory produces a stream of chemical, physical, and biological readings.

Hidden within this data are the songs of humpback whales, the vibrations of vast shoals of migratory herring, and clues to a changing global climate. There鈥檚 just one difficulty: How to detect scientifically important events in this ocean of data.

Our Purpose

The global callout

This was the challenge for 乌鸦传媒 colleagues around the world eager to participate in the Global Data Science Challenge (GDSC). This annual company-wide internal competition attracts hundreds of employees ready to compete to solve real-world challenges using artificial intelligence.

In 2020, entrants harnessed AI and machine learning to鈥�identify individual sperm whales鈥痠n order to monitor migration patterns and protect the whales鈥� natural habitats.

In the 2021 challenge, from the 673 teams 鈥� representing 1,200 乌鸦传媒 colleagues 鈥� that entered the competition, two emerged victorious to share the top prize: One based in India, and the other in the UK.

An ocean of data

Anupam Saha, senior delivery manager at 乌鸦传媒 India and leader of the Indian team, explains the challenge. 鈥淲e were asked to build an AI solution that could analyze the reams of sensor data collected by the LoVe observatory and detect the anomalies that will direct further study.鈥�

David Gilhooley, principal consultant and engagement manager, 乌鸦传媒 UK, who led the other winning team, adds that the sheer volume of data presented a technical challenge in itself. 鈥淭his is the crux of the problem: the observatory is collecting masses of data showing the ocean being 鈥榥ormal鈥�, whereas it鈥檚 the anomalies that are interesting.

Winning approaches

Anupam鈥檚 team addressed the problem by breaking down the data into pieces. 鈥淲e tackled each data source individually, designing a model that would identify the outliers in each set. Much of our focus was on the data pre-processing, to determine the most relevant variables among the thousands we were presented with.鈥�

The UK team also prioritized careful data management. 鈥淲e had to take these multiple data sources and organize them day by day,鈥� explains David. 鈥淎s well as padding out the missing areas, we had to normalize the data in order to deploy the machine-learning analysis correctly.鈥�

Learning new skills

David鈥檚 team had dabbled in data processing and machine learning in small-scale academic settings, so they viewed the competition as an opportunity to sharpen their skills in a real-world scenario.

鈥淭his is the type of problem you鈥檇 get in an industrial setting 鈥� searching for outliers in a huge dataset with lots of variables,鈥� he says. 鈥淔rom that perspective, the challenge was really practical. We were excited to learn the AWS [Amazon Web Services] machine-learning tools, and get experience with this technology.鈥�

A true team effort

David鈥檚 team included Vincent Malmedy, Gabriela Pomery, and Andrew Pennington, all based at the 乌鸦传媒 office in Bristol. 鈥淭he competition was the perfect way to knit the team back together after all the pandemic disruption,鈥� says David.

For Anupam, his proudest moment was his team鈥檚 presentation to the judges. 鈥淏efore this, we were ranked in fifth place,鈥� he explains. 鈥淏ut we had realized the person using our model at the observatory might not have a background in data. So we avoided being too theoretical with our presentation, and I think this helped us claim joint first place overall.

Environmental AI

While these technologies shouldn鈥檛 be viewed as a 鈥榮ilver bullet鈥�, David believes that AI and machine-learning can help us better understand climate change and global warming. 鈥淚t鈥檚 hard for human beings to understand all the incremental steps involved in these huge processes. Given the right instructions, however, machine-learning tools can help navigate these complexities.鈥�

He believes that the competition, alongside 乌鸦传媒鈥檚 broader climate commitments, has made his team more aware of the choices they make 鈥� for example, in terms of how they commute to work, and how much plastic they use.

Both teams received a technology prize for their winning entries, in lieu of a trip to Norway to visit the observatory, which unfortunately had to be canceled because of the pandemic.

Gearing up for the next challenge

乌鸦传媒鈥檚 team in Germany is continuing its work with the LoVe observatory to integrate the winning solutions into its platform, enabling a large community of researchers to benefit from a broader understanding of oceanic ecosystems. Next year, the GDSC will focus on finding a cure for river blindness – another opportunity to shape a better future using AI and machine-learning.

Inside stories

Future-shaping projects

Hunting for dragons

Sarah Gon莽alves wasn鈥檛 always a data engineer. She began her career as an archaeologist, unearthing the secrets of the past in sites across France before retraining as a data engineer. She brings the same curiosity to her new vocation at 乌鸦传媒, using artificial intelligence (AI) to uncover what lies beneath the surface.