DeepSea Technologies, the award-winning AI-led maritime company, has recently announced its partnership with other industry-leading institutions and researchers in the field of AI to launch Shifts Project; an international effort to study the distributional shift in machine learning for real-world applications.
The most prominent example of distributional shift in maritime AI is where a vessel’s dataset “moves” over time, as a result of hull fouling. Marine fouling occurs when organisms attach themselves to underwater objects like boats, rope, pipes and building structures. Fouling can lead to various operational inefficiencies, thus understanding how a vessel’s dataset shifts over time is of crucial importance for vessel’s performance modeling.
More robust and interpretable vessel models could be the key for safety but also to unlocking shipping’s huge decarbonisation potential and minimising fuel waste.
The key question we are trying to address in this presentation is: How can we build and prove a vessel’s model is robust on shifting or even unseen conditions? The answer to such a question is of paramount importance when dealing with high-risk AI tasks like routing optimisation. DeepSea has established a benchmark evaluation framework that enables the systematic exploration of a model’s generalization capability on shifted or unseen data. With the proposed evaluation methodology, released within the Shift Project initiative, we aim to contribute towards the development of transparent and explainable methods that prove the value of AI but also promote the trust in such technology across the shipping industry.
To learn more about the Shifts Project initiative, visit shifts.ai.
About DeepSea Technologies
DeepSea is a maritime technology company specialising in vessel performance monitoring and optimisation. We combine many years of experience in the shipping industry with deep expertise in the fields of software, hardware, and Artificial Intelligence. We are passionate about creating products that help our clients make data-driven decisions to achieve their decarbonisation objectives and lead the industry forward. DeepSea has built a pioneering next-generation Performance Routing tool – Pythia – which offers shipowners an average of 8% savings in emissions and fuel consumption.
Visit our website www.deepsea.ai to find out more about how we help shipping companies become more efficient and reduce their CO2 emissions.
About the speaker
Antonis Nikitakis is the Director of the AI Research department of DeepSea Technologies. He is a electrical and computer engineer and a cognitive psychologist. He holds an Electrical and Computer Engineering diploma, a master's degree (DUTH & Tech. Uni of Crete) and a PhD in Embedded Design and Hardware Acceleration for Computer Vision Systems (Tech. Uni of Crete). He has more than 20 publications in international journals and conferences on AI, embedded/hardware design and computer vision. He has been involved for more than 10 years as a researcher in numerous European and national competing research programs (FP7, H2020) regarding the design, development and validation of state-of-the-art technologies in reconfigurable hardware, embedded design and AI.