Skip to main content

AIS-Based Vessel Classification for Maritime Security

Embarking on a research voyage, we delve into the expansive sea, utilizing artificial intelligence as our compass to tackle the formidable challenge of preventing abuse and illegal activities. Our focus centers on the Ionian-Adriatic Sea, where the innovative application of a deep Long Short-Term Memory (LSTM) model, operating on AIS data of varying quality, emerges as a powerful tool.

During the EU ANDROMEDA H2020 project's trial period in the Adriatic-Ionian region, our model stands as a sentinel, proficiently identifying nine common vessel classes. Beyond conventional classifications, it excels in detecting vessels engaged in unexpected behaviors, presenting a significant leap forward in bolstering maritime security.

This impactful research, recognized and published in esteemed scientific and maritime journals, has garnered acclaim by winning the prestigious Best Paper Award. Beyond demonstrating the prowess of artificial intelligence in navigating the complexities of the sea, this achievement signifies a tangible contribution to the evolving landscape of maritime security. Stay tuned for further insights into our journey as we continue to innovate and safeguard our waters.


Stay curious with me.

Comments

Post a Comment

Popular posts from this blog

Ensuring Responsible Integration of LLMs in Data Science Workflows

It is crucial to exercise caution when incorporating LLMs into data science workflows, particularly with previously unseen data or in unfamiliar domains. Data science at its core involves a comprehensive understanding of data within its unique context. LLMs, while capable of generating functional code and providing insightful suggestions, do not inherently comprehend the underlying implications of the data they process. This disconnect can introduce biases in methodology and potentially lead to a misinterpretation of the problem. To mitigate these risks, it is advisable to limit the use of LLMs to labor-intensive, basic data wrangling tasks. Even then, sensitive data should not be directly fed into the LLM. Instead, use dummy examples or limited portions of the data to guide the LLM without compromising data integrity. This approach depends heavily on understanding the intricate relationships between features and adhering to data protection requirements. However, be aware that strippin...
Post a Comment
Read more

Enhancing DNN Explainability in BCI

This project tackled the challenge posed by the opaque nature of Deep Neural Network (DNN) models, especially crucial in safety-driven fields like Brain-Computer Interface (BCI) applications. The primary aim was to unleash the full potential of DNN models, specifically in analyzing EEG signals. In this initial study, I highlighted the importance of not just relying on quantitative performance but also considering qualitative evaluation. The investigation delved into a key paper in the field, providing a comprehensive analysis as a starting point. I then compared the model against common alternatives, tweaking various hyper-parameters to validate its performance. To demystify the "black-box," I utilized the Layer-wise Relevance Propagation (LRP) method, revealing insightful details about each model's reliability. LRP proved invaluable in explaining the inner workings of the DNN, making it a useful tool in interpreting complex neural network decisions and enhancing transpar...
Post a Comment
Read more

Dynamic Learning Tool for Medical Education

Embarking on a transformative journey in medical education, I present a cutting-edge self-learning and self-evaluating platform. Originally built for the aspiring minds of medical students at Sapienza University, this innovation takes the form of a tree-structured questionnaire, introducing a dynamic approach to learning. How It Works Designed to adapt to individual progress, our platform customizes the question flow based on previously selected answers. This dynamic feature not only provides instant feedback but serves as a motivational guide for learners. Users engage with a series of thought-provoking questions, scenarios, and answer choices, witnessing results and explanations unfold in real-time. Active Recall Technique At the heart of this revolutionary platform is an underlying algorithm that employs the powerful Active Recall technique. This technique enhances memory retention by prompting users to actively recall information, contributing to a more robust and enduring understa...
Post a Comment
Read more