DeepSea Technologies, the award-winning Al-led maritime technology company and energy efficiency experts, has announced that it is making a full vessel data set and DeepSea’s automatic model evaluation tool available for free to the shipping industry for the next 12 months. This is the first time that such tools have ever been made available to the public – and allows innovation, research and technical departments with strong data science capabilities to start exploring AI.
The announcement follows the publication of the winners of the Shifts Challenge 2023. The competition is an international collaboration of academic and industrial researchers, including Cambridge University, and is helping to make shipping a first-class citizen in AI research.
Overall, the Shifts Challenge received 175 submissions from institutional and academic research teams and individuals. The two areas of focus were ship power prediction and white matter multiple sclerosis lesion segmentation. The winning submission in the shipping track was made by a team from IBM Dublin, which proposed a novel approach harnessing learnings from the wind turbine sector in order to gain new insights from the data provided. Their solution could predict vessel power under a broad set of conditions, recording a 13% performance improvement against the reference model DeepSea provided as a baseline.
The full vessel data set from the challenge will now be provided for use to researchers and technical specialists in the maritime industry, and can be accessed via Zenodo (an open-source research database operated by CERN). DeepSea intends for it to help support the development of robust vessel models across the industry – something essential in enabling the sector to cut costs and cope with environmental regulations.
Commenting on the Shifts Challenge, Dr Nikitakis, DeepSea’s AI Director, said: “We were delighted to see the diverse set of solutions that the competitors provided. We’re also pleased to be opening this AI environment up to the whole of the shipping industry. We are always trying to “fight the hype” around AI, and there is no better way than giving the public the tools to explore – and evaluate – their own AI solutions. Perhaps later down the line we will put on a similar challenge for all AI providers in the industry – we hope they’ll take us on!”
Commenting on their winning submission, Seshu Tirupathi from the IBM Dublin team, said: “Challenges in making inferences on data with distribution shifts can be expected to gain increased importance with the exponential growth of data and real-time monitoring applications. This sort of research is extremely important, as it avoids unreliable predictions that may lead to life-threatening situations. Avenues like Shifts Challenge provide ground truth from real-world data that is essential to develop and validate robust algorithms in the space of distribution shifts and concept drifts.”
Dr. Konstantinos Kyriakopoulos, CEO and co-founder of DeepSea, said: “Though AI is still a relative newcomer to the shipping sector, it has already proven itself as an exceptional tool for those looking to optimise fuel consumption, emissions, and industry ratings. Like any other cutting-edge research area, this technology can only develop as part of an ecosystem, and the Shifts Project is (amongst other things) designed to help catalyse and support this movement. Industrial AI is a field where bullshit is rife, and it’s these sort of initiatives that help the industry to know who to trust.”
Background to The Shifts Project
The Shifts Project is an international effort involving multiple institutions alongside DeepSea, including the universities of Cambridge, Basel, Lausanne, and HES-SO Valais. The initiative, which aims to build a cross-disciplinary international community, brings together core machine learning (ML) researchers studying distributional shift with applied ML researchers, who work on tasks affected by distributional shift in the real world.
Handling distributional shift is one of the greatest obstacles to the widespread adoption – and impact – of AI across all industries. This is especially the case in shipping, and the world’s top experts are now collaborating to explore solutions. Maritime was one of two focuses in the competition – the other being distributional shift in relation to the treatment of the chronic condition, Multiple Sclerosis.
A great example of distributional shift is found in maritime – where the entire ship data set moves over time – as a result of hull fouling. Marine fouling occurs when organisms attach themselves to underwater objects, most notably the hull. This can lead to various operational inefficiencies and have a dramatic impact on vessel fuel consumption, emissions and CII – so understanding how to predict its effect using vessel data is a vital tool for the industry. Understanding how the entire ship data shifts over time is crucial to accurately modelling vessels, which is the key to unlocking shipping’s huge decarbonisation potential and minimising fuel waste.
