DeepSea Technologies, the award-winning marine technology and energy efficiency expert company led by Al, has published a new study outlining a pioneering way to verify the accuracy – and therefore the usefulness – of the AI-generated model of a ship in real conditions. The new approach was developed by seven of 13 DeepSea researchers, led by Dr. Antonis Nikitakis, and presented at the HullPic 2022 conference in Tullamore, Ireland in May 2022.
Because the reality of ship-at-sea data is highly variable, most model accuracy figures reported in publications and marketing materials do not correspond to the actual usefulness of these models in real-world use cases. . DeepSea has been researching approaches to solve the technical challenge of enhancing the ability of models to understand unseen (“out of domain”) conditions for years. However, until today, there was no benchmark to assess this type of skill within a ship model.
This is a crucial step in AI modeling because the more accurate the virtual model, the more efficient a ship can be, and vice versa. The few models that currently provide an estimate of their accuracy all do so based on testing with data obtained from the same distribution (i.e. representative of similar conditions and containing similar biases) as the data. used to train the model. For example, if the model is trained on data from the historical behavior of the ship, within a narrow range of well-known wind speeds or drafts, it is also tested on data with those speeds and drafts. Thus, the tests performed cannot tell whether the model reproduces the biases in the training data – and whether it will perform as well under different, never-before-seen conditions.
Following the publication of this body of work on proving the real utility of AI approaches, DeepSea will focus on promoting common and transparent standards for the real utility of AI in industry.
Commenting on the research, Dr Nikitakis, Director of AI Research at DeepSea, said: “This research is an important step in helping our customers and the wider maritime community understand the true power of an AI-based approach. on AI while mitigating its limitations. Coupled with the real day-to-day impact we see on fuel mileage and CII ratings, we believe this type of insight is critical to popularizing this amazing technology across the industry.
“As serious researchers working to popularize the AI approach continue to search for rigorous methods to prove the real value of what they create, we can also expect that end users looking to use this type of technology – in shipping and other sectors – require proof of its usefulness.
Dr Konstantinos Kyriakopoulos, CTO and co-founder of DeepSea, said: “We have designed our AI framework for the direct benefit of the consumer – shipowners, cargo owners and all other members of the industry who cover the costs. fuel. I’m excited to see how, through this research, we can fight the hype and support our honest approach to AI with such compelling evidence. This is exactly what DeepSea was founded to do and what, every day, has a growing impact on our clients’ results and the sustainable future of the planet. DeepSea has released details of its approach in the hope that global researchers can use it themselves and catalyze greater transparency and common standards across the industry.
Source: DeepSea Technologies