Tiago Sousa

Mr. Tiago Sousa

Department of Computer Science, University of Luxembourg, Luxembourg, Luxembourg
orcid

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
22 January 2025 in Information
Verified
Journal Article
Open access
PDF file
Tiago SousaTiago Sousa
Tiago Sousa
BRBenoît Ries
Benoît Ries
Nicolas GuelfiNicolas Guelfi
Nicolas Guelfi
High-quality Earth Observation (EO) imagery is essential for accurate analysis and informed decision making across sectors. However, data scarcity caused by atmospheric conditions, seasonal variations, and limited geographical coverage hinders the effective application of Artificial Intelligence (AI) in EO. Traditional data augmentation techniques, which rely on basic parameterized image transformations, often fail to introduce sufficient diversity across key semantic axes. These axes include natural changes such as snow and floods, human impacts like urbanization and roads, and disasters such as wildfires and storms, which limits the accuracy of AI models in EO applications. To address this, we propose a four-stage data augmentation approach that integrates diffusion models to enhance semantic diversity. Our method employs meta-prompts for instruction generation, vision–language models for rich captioning, EO-specific diffusion model fine-tuning, and iterative data augmentation. Extensive experiments using four augmentation techniques demonstrate that our approach consistently outperforms established methods, generating semantically diverse EO images and improving AI model performance.

1 October 2023
Verified
Conference Paper
Closed access
Tiago SousaTiago Sousa
Tiago Sousa
Since the Stockholm Declaration on the human environment in 1972, there has been a growing recognition of the impact of human activities on Earth's ecosystems. This has created an increasing need for modeling and predicting the resilience of ecosystems, which is crucial not only for understanding ecosystem patterns and processes but also for addressing climate change and implementing effective conservation and management strategies. Despite the importance of this issue, the intrinsic complexity of ecosystems and the lack of sufficient data present considerable challenges. To address these challenges, we propose an approach that combines model-driven engineering and artificial intelligence. Specifically, we propose a formalization for modeling and verifying ecosystem requirements, a method for synthesizing heterogeneous ecosystem resilience data, and a product line of neural network architectures adaptable to diverse properties and types of ecosystem scenarios to study. Additionally, we propose a model-driven process specification detailing the different artifacts, stakeholder roles, tasks, and model transformations of the proposed approach. This paper outlines the problem and preliminary work, presents the proposed approach, which is the current focus of an ongoing Ph.D. thesis, and discusses the future research contributions.

31 May 2024 in arXiv
Verified
Preprint
Open access
PDF file
Tiago SousaTiago Sousa
Tiago Sousa
BRBenoît Ries
Benoît Ries
NGNicolas Guelfi
Nicolas Guelfi
United Nations have declared the current decade (2021-2030) as the "UN Decade on Ecosystem Restoration" to join R\&D forces to fight against the ongoing environmental crisis. Given the ongoing degradation of earth ecosystems and the related crucial services that they offer to the human society, ecosystem restoration has become a major society-critical issue. It is required to develop rigorously software applications managing ecosystem restoration. Reliable models of ecosystems and restoration goals are necessary. This paper proposes a rigorous approach for ecosystem requirements modeling using formal methods from a model-driven software engineering point of view. The authors describe the main concepts at stake with a metamodel in UML and introduce a formalization of this metamodel in Alloy. The formal model is executed with Alloy Analyzer, and safety and liveness properties are checked against it. This approach helps ensuring that ecosystem specifications are reliable and that the specified ecosystem meets the desired restoration goals, seen in our approach as liveness and safety properties. The concepts and activities of the approach are illustrated with CRESTO, a real-world running example of a restored Costa Rican ecosystem.

22 September 2024
Verified
Conference Paper
Open access
Tiago SousaTiago Sousa
Tiago Sousa
BRBenoît Ries
Benoît Ries
Nicolas GuelfiNicolas Guelfi
Nicolas Guelfi
Achieving the Sustainable Development Goals (SDGs) set by the United Nations requires innovative solutions to address the related complex and interconnected challenges. The application of AI has demonstrated the potential to significantly contribute to these efforts by providing advanced analytics and decision-making capabilities. However, integrating AI into sustainability initiatives faces several challenges, including the need for flexible and reusable solutions that can be adapted to diverse and evolving SDG contexts, as well as the challenge of making these technologies accessible to nonexpert stakeholders. This paper proposes an integrated approach that combines Model-Driven Engineering (MDE) with Software Product Line Engineering (SPLE) to address these challenges. The proposed process includes key activities such as domain analysis, metamodel-driven requirements specification, product derivation, and AI model training. This approach aims to automate the derivation of flexible and reusable AI architectures tailored to specific SDG contexts, thus reducing the development time of AI-based software solutions for sustainability efforts.