DATA SCIENCE & Ai
KEY-ENABLING TECHNOLOGIES
The Data Science research programme focuses on the development of data-driven technologies to solve both classical and new problems in the maritime domain.
APPROACH
With the increasing amount of data, compute power and algorithms available, possibilities to analyse data in new ways emerge rapidly. This results in a couple of very interesting use cases.
The data science research maintains a fundamental approach towards the solutions that are developed. This is to ensure transferability and applicability beyond current use cases and to prepare for future developments.
The data science research maintains a fundamental approach towards the solutions that are developed. This is to ensure transferability and applicability beyond current use cases and to prepare for future developments.
Contact
Bart Mak
Senior Researcher
sub programmes
Data Driven Predictions
Many applications require an interpretation of (measured) data that is not straight forward. If physics models are not available or if computations become too expensive, a data driven model can be constructed, if enough data is available. This approach is useful in the Digital Twin context, but can also be applied to CFD, instrumentation calibration or behaviour mimicking agents.
Data Driven Predictions
Many applications require an interpretation of (measured) data that is not straight forward. If physics models are not available or if computations become too expensive, a data driven model can be constructed, if enough data is available. This approach is useful in the Digital Twin context, but can also be applied to CFD, instrumentation calibration or behaviour mimicking agents.
Data Driven Design
Searching for optimal design parameters can be very expensive, if the entire search effort is accompanied by full simulations. To mitigate this, surrogates are used to interpolate over the search domain in order to find near-optimal parameters with little effort, requiring only a few more simulations to finalize the optimization. Sometimes, however, the intended interpolation is more complex than what traditional surrogates can handle. MARIN investigates the use of machine learning approaches for this problem with specific focus on correctly capturing the physics while minimizing the required number of simulations.
Data Driven Design
Searching for optimal design parameters can be very expensive, if the entire search effort is accompanied by full simulations. To mitigate this, surrogates are used to interpolate over the search domain in order to find near-optimal parameters with little effort, requiring only a few more simulations to finalize the optimization. Sometimes, however, the intended interpolation is more complex than what traditional surrogates can handle. MARIN investigates the use of machine learning approaches for this problem with specific focus on correctly capturing the physics while minimizing the required number of simulations.
Interaction Optimisation
Choosing actions in an interactive environment expands the problem of learning or optimisation to predicting long term revenue. This revenue is not only based directly on the choice of actions, but also on the response of the environment. This response may be non-deterministic, requiring a good estimation of what to expect in the future, only partly based on the choice of actions.
Optimising the so-called policy that chooses the actions can be done with Reinforcement Learning, a set of techniques that optimises for overall performance while at the same time taking advantage of local opportunities. The range of applications spans from actuator control to collision avoidance and route planning.
Optimising the so-called policy that chooses the actions can be done with Reinforcement Learning, a set of techniques that optimises for overall performance while at the same time taking advantage of local opportunities. The range of applications spans from actuator control to collision avoidance and route planning.
Interaction Optimisation
Choosing actions in an interactive environment expands the problem of learning or optimisation to predicting long term revenue. This revenue is not only based directly on the choice of actions, but also on the response of the environment. This response may be non-deterministic, requiring a good estimation of what to expect in the future, only partly based on the choice of actions.
Optimising the so-called policy that chooses the actions can be done with Reinforcement Learning, a set of techniques that optimises for overall performance while at the same time taking advantage of local opportunities. The range of applications spans from actuator control to collision avoidance and route planning.
Optimising the so-called policy that chooses the actions can be done with Reinforcement Learning, a set of techniques that optimises for overall performance while at the same time taking advantage of local opportunities. The range of applications spans from actuator control to collision avoidance and route planning.
Artificial Intelligence (AI)
At MARIN we explore the pivotal role of artificial intelligence (AI) in the maritime industry. AI stands on the brink of transforming a vast range of maritime activities, offering sophisticated data analytics, machine learning algorithms, and autonomous systems. For this we offer a diverse array of AI applications.
Artificial Intelligence (AI)
At MARIN we explore the pivotal role of artificial intelligence (AI) in the maritime industry. AI stands on the brink of transforming a vast range of maritime activities, offering sophisticated data analytics, machine learning algorithms, and autonomous systems. For this we offer a diverse array of AI applications.
Relevant LINKS