RUN
Repositório Institucional da Universidade NOVA de Lisboa
Entradas recentes
When Pixels Lie
Publication . Hovakimyan, Gurgen; Bravo, Jorge Miguel; Information Management Research Center (MagIC) - NOVA Information Management School; NOVA Information Management School (NOVA IMS)
Concept drift poses a major challenge to the reliability and transparency of computer vision systems. This study provides a systematic analysis of how four drift types, such as noise, blur, rotation, and brightness, affect both model performance and Grad-CAM/Grad-CAM++ explainability across CIFAR-10 and Tiny ImageNet datasets using three CNN architectures (ResNet-18, DenseNet-121, ShuffleNet v2). Following recent robustness literature, we treat these controlled test-time transformations as synthetic covariate-shift proxies for the input-distribution component of drift rather than as temporal concept drift in the strict sense. For CIFAR-10, accuracy declines from a baseline of 79–86% to as low as 24% under strong noise drift, while SSIM and IoU between baseline and drifted heatmaps fall from nearly 100% to as low as 18% and 25%, respectively. On Tiny ImageNet, the degradation is even more pronounced: top-1 accuracy collapses from 40–60% to below 4% under the strongest noise drift, with SSIM and IoU dropping to as low as 10% (per-architecture values are reported in the Results and Discussion section). Rotation and brightness prove markedly milder than noise and blur, indicating that high-frequency corruptions are the principal threat to both accuracy and attribution stability. Image-level paired tests (Wilcoxon signed-rank and paired t-test) with effect sizes and bootstrap confidence intervals, complemented by pixel-level distributional diagnostics (KS, Anderson-Darling, Wilcoxon rank-sum), confirm significant shifts in heatmap distributions, revealing that drift alters not only model outputs but also internal attribution patterns. The results demonstrate that different drift mechanisms degrade interpretability in distinct ways and that explainability metrics are often more sensitive to drift than accuracy itself. This work provides, to the best of our knowledge, one of the first unified, quantitative evaluations of explainability degradation under controlled visual drift, highlighting the need for adaptive, drift-aware XAI pipelines in dynamic real-world environments.
A conceptual framework for ESG–OPEX integration
Publication . Carneiro, F.; Nóvoa, H.; Carvalho, M.; DEMI - Departamento de Engenharia Mecânica e Industrial; UNIDEMI - Unidade de Investigação e Desenvolvimento em Engenharia Mecânica e Industrial; Elsevier
Environmental, Social, and Governance (ESG) principles and Operational Excellence (OPEX) systems have evolved from distinct managerial traditions: one focused on sustainability and the other on efficiency and performance improvement. Despite their conceptual complementarity, their integration remains limited and insufficiently structured. This study addresses this gap by proposing an integrated conceptual framework that aligns ESG objectives with OPEX capabilities in an operationally viable way. The framework was developed through a structured conceptual literature review and thematic synthesis and is organised around three interconnected domains: Strategic Alignment, Integration Mechanisms, and Execution & Outcomes. The Strategic Alignment domain links ESG strategic priorities with OPEX enablers and capabilities through mechanisms such as Hoshin Kanri, digital leadership, and ESG–OPEX trade-off resolution. The Integration Mechanisms domain translates these priorities into practice through managerial mechanisms, including ESG-aligned KPIs, project selection, change management and frontline engagement, as well as digital/AI capabilities, such as predictive analytics, data governance, digital twins, and AI-enabled supply chain adaptability. The Execution & Outcomes domain identifies adaptive ESG–OPEX improvement actions, information-driven strategic resilience, and measurable outcomes, including stakeholder trust, innovation, enhanced ESG performance, organisational capability building, and sustainable value creation. An exploratory case-based assessment in the pulp and paper sector was conducted to examine the model’s operational plausibility. The study contributes to ESG implementation literature and supports organisations in embedding ESG into continuous improvement systems.
Energy-Gap Topology for Mapping Coupling Architectures in Lanthanide Co-Doped Photonic Materials
Publication . Vasconcelos, Helena Cristina; Meirelles, Maria Gabriela; LIBPhys-UNL; MDPI - Multidisciplinary Digital Publishing Institute
Lanthanide co-doped photonic materials are commonly interpreted through selected resonant transitions between assigned 4f multiplet states. Here, we introduce an energy-gap topology framework that compares Ln3+ ion pairs from the complete set of internal separations within their 4f manifolds. Reported multiplet-centre energies for a common LaF3 spectroscopic reference were reduced to energy-only manifolds, from which normalized gap distributions were constructed. A symmetric descriptor, (Formula presented.), quantifies global similarity between complete gap distributions, whereas a directional descriptor, (Formula presented.), measures the inclusion of the gaps of ion (Formula presented.) within the occupied support of ion (Formula presented.). When applied to 11 trivalent lanthanide ions, this reveals distinct coupling architectures. Tb3+–Ho3+, Sm3+–Dy3+, Dy3+–Ho3+, Nd3+–Ho3+, and Er3+–Ho3+ exhibit high symmetric overlap, defining a broad manifold–manifold topology signature. In contrast, Yb3+-containing pairs show low global compatibility but high directional inclusion of the Yb3+ gap within the landscapes of candidate acceptor ions, such as Er3+, Tm3+, and Nd3+, consistent with sparse-to-rich sensitizer-like architectures. The pairwise organization remains stable for bin widths between 250 and 1000 cm−1. The descriptors provide a pre-spectroscopic screening map of energetic architecture; they do not predict the transfer efficiency, dominant mechanism, or final optical performance and should not be interpreted as host-independent constants.
Implementation and Impact of Nutritional Protocols in Oncology
Publication . Alpuim Costa, Diogo; Capela, Andreia; Afonso, Ana; Pinho, João Pedro; Irving, Susana Couto; Costa, Marta; Macedo, Cátia; Lopes, Ana Bagulho; Caseiro, Tânia; Alves, Paula; NOVA Medical School|Faculdade de Ciências Médicas (NMS|FCM); Elsevier
BACKGROUND AND AIM: Malnutrition is a prevalent yet under-recognized complication in oncology that adversely affects treatment outcomes and quality of life. Although existing guidelines recommend early nutritional screening and intervention, their consistent implementation in routine oncology care remains limited. This study aimed to address this gap by: (1) developing and implementing a nutritional screening protocol across cancer day hospitals, (2) assessing health care professionals' (HCPs) perceptions about the protocol's effectiveness, and (3) building expert consensus on approaches to better integrate nutritional care into routine oncology practice. METHODS: A standardized nutritional screening protocol for patients with cancer was implemented across 13 hospitals in Portugal. Subsequently, a mixed-methods approach was employed involving, in the first phase, a quantitative survey to assess HCPs' perceptions regarding the protocol's implementation and impact, and in the second phase of the study, an expert panel consensus. The survey's structured questionnaire evaluated implementation practices, perceived care improvements, and patient demographics. Surveys were distributed in two rounds (October 1 and October 21, 2024) and responses were collected over two months. Subsequently, the findings were presented to a multidisciplinary expert panel (May, 2025) to establish consensus on strategies for the integration of nutritional care into routine oncology practice. Descriptive statistics and qualitative synthesis were used for data analysis. RESULTS: Sixty HCPs (oncologists: 21, nurses: 23, nutritionists: 16) participated in the survey. Nutritional risk screening was conducted for 1,023 patients over an average period of four months; 55% of them were at nutritional risk as per the HCP survey. Screening was primarily conducted by nurses during initial appointments, using validated tools such as Nutritional Risk Screening (NRS)-2002. Key criteria for identifying risk included: nutrition impact symptoms, weight loss, and reduced intake. The most commonly implemented interventions were referral to a clinical nutritionist and the use of oral nutritional supplements. Post-implementation, nutritional consultations increased by 20.2% (p = 0.033), and waiting times for high-risk patients decreased by 43.2% (p = 0.001), allowing for compliance with the national directives. HCPs reported improved multidisciplinary coordination and training. The main implementation challenges were time constraints and insufficient staff. CONCLUSIONS: Implementation of nutritional screening protocols in routine oncology care is feasible and effective but requires systemic support for sustainability.
A Comparative Study of EO Data Fusion and Satellite Embeddings for Soil Organic Carbon Estimation in Mainland Portugal
Publication . Idrees, Abran; Cabral, Pedro; Jamal, Saad Ahmed; NOVA Information Management School (NOVA IMS); Information Management Research Center (MagIC) - NOVA Information Management School; Wiley
Reliable information on soil organic carbon (SOC) is essential for sustainable land management, climate change mitigation, and the conservation of soil biodiversity. However, conventional SOC assessment, which generally involves field based soil sampling followed by laboratory analyses such as dry combustion, wet oxidation, is labor intensive, time consuming, and costly, particularly when applied over large areas. The integration of Earth Observation (EO) data with Machine Learning (ML) algorithms has the potential to predict SOC on a large scale. This study estimates the SOC in mainland Portugal using two data-sets; Dataset-I incorporates Sentinel-1/2, with ancillary climate and topographic covariates while Dataset-II contains the Google Satellite Embeddings (GSE) dataset. The Recursive Feature Elimination with Cross Validation (RFECV) was used to identify and remove redundant variables. Three state-of-the-art ML algorithms, i.e., Random Forest (RF), Gradient Tree Boost (GBT)and Classification and Regression Tree (CART) were used for SOC prediction. The results indicate that RF achieved the highest predictive performance using Dataset-II, with an R2 of 0.34, RMSE of 19.2 gCkg−1, MAE of 13.2 gCkg−1, and RPIQ of 1.27 under spatial cross validation, while independent validation on the 43 withheld samples yielded an R2 of 0.44, RMSE of 19.89gCkg−1, MAE of 15.50 gCkg−1, and RPIQ of 1.18. The calibrated RF uncertainty was generally lower across central and southern Portugal, with greater spatial variability in the northern region. The findings contribute to the advancement of SOC measurements for sustainable development.