Abstract
Education is vital in fostering economic growth and societal development, particularly in developing countries like Timor-Leste. As technology has revolutionised education in the digital transformation era, the concept of a smart university, driven by advanced technologies and data analytics, has gained prominence globally. Timor-Leste, amid its progress in institutional structures and public infrastructure, is also exploring integrating smart technologies in higher education. This underscores a commitment of The East Timor National Education Strategic Plan (NESP) 2011-2030 to meet national and international standards, positioning the country at the forefront of educational innovation. This study aims to assess the feasibility of implementing a Smart University in Timor-Leste to evaluate the readiness of the country to embrace digital technologies and integrate them into higher education practices. The research employs a Design Science Research methodology where qualitative and quantitative data are gathered through interviews, surveys, and document analysis. Design artefacts, including system architecture and an evaluation framework, are developed to comprehensively understand the technological and informatics aspects of implementing a Smart University in Timor-Leste. The findings will contribute to decision-making and inform the implementation plan, offering valuable insights into stakeholders' perspectives and perceptions, and will support the advancement of the educational landscape in Timor Leste by integrating smart technologies and innovative practices in higher education.

Abstract
Post-hoc explanation methods have often been criticised for abstracting away the decision-making process of deep neural networks. In this work, we would like to provide natural language descriptions for what different layers of a vision backbone have learned. Our DeViL method generates textual descriptions of visual features at different layers of the network as well as highlights the attribution locations of learned concepts. We train a transformer network to translate individual image features of any vision layer into a prompt that a separate off-the-shelf language model decodes into natural language. By employing dropout both per-layer and per-spatial-location, our model can generalize training on image-text pairs to generate localized explanations. As it uses a pre-trained language model, our approach is fast to train and can be applied to any vision backbone. Moreover, DeViL can create open-vocabulary attribution maps corresponding to words or phrases even outside the training scope of the vision model. We demonstrate that DeViL generates textual descriptions relevant to the image content on CC3M, surpassing previous lightweight captioning models and attribution maps, uncovering the learned concepts of the vision backbone. Further, we analyze fine-grained descriptions of layers as well as specific spatial locations and show that DeViL outperforms the current state-of-the-art on the neuron-wise descriptions of the MILANNOTATIONS dataset.

Abstract
Climate change has intensified the need for robust fire prevention strategies. Sustainable forest fuel management is crucial in mitigating the occurrence and rapid spread of forest fires. This study assessed the impact of vegetation clearing and/or grazing over a three-year period in the herbaceous and shrub parts of a Mediterranean oak forest. Using high-resolution multispectral data from an unmanned aerial vehicle (UAV), four flight surveys were conducted from 2019 (pre- and post-clearing) to 2021. These data were used to evaluate different scenarios: combined vegetation clearing and grazing, the individual application of each method, and a control scenario that was neither cleared nor purposely grazed. The UAV data allowed for the detailed monitoring of vegetation dynamics, enabling the classification into arboreal, shrubs, herbaceous, and soil categories. Grazing pressure was estimated through GPS collars on the sheep flock. Additionally, a good correlation (r = 0.91) was observed between UAV-derived vegetation volume estimates and field measurements. These practices proved to be efficient in fuel management, with cleared and grazed areas showing a lower vegetation regrowth, followed by areas only subjected to vegetation clearing. On the other hand, areas not subjected to any of these treatments presented rapid vegetation growth.

Abstract
Large language models (LLMs) have substantially pushed artificial intelligence (AI) research and applications in the last few years. They are currently able to achieve high effectiveness in different natural language processing (NLP) tasks, such as machine translation, named entity recognition, text classification, question answering, or text summarization. Recently, significant attention has been drawn to OpenAI's GPT models' capabilities and extremely accessible interface. LLMs are nowadays routinely used and studied for downstream tasks and specific applications with great success, pushing forward the state of the art in almost all of them. However, they also exhibit impressive inference capabilities when used off the shelf without further training. In this paper, we aim to study the behavior of pre-trained language models (PLMs) in some inference tasks they were not initially trained for. Therefore, we focus our attention on very recent research works related to the inference capabilities of PLMs in some selected tasks such as factual probing and common-sense reasoning. We highlight relevant achievements made by these models, as well as some of their current limitations that open opportunities for further research.This article is categorized under:Fundamental Concepts of Data and Knowledge > Key Design Issues in DataMiningTechnologies > Artificial Intelligence

Abstract
Angle assessment is crucial in rehabilitation and significantly influences physiotherapists' decision-making. Although visual inspection is commonly used, it is known to be approximate. This work aims to be a preliminary study about using the AI image-based to assess upper limb joint angles. Two main frameworks were evaluated: MediaPipe and Yolo v7. The study was performed with 28 participants performing four upper limb movements. The results showed that Yolo v7 achieved greater estimation accuracy than Mediapipe, with MAEs of around 5 degrees and 17 degrees, respectively. However, even with better results, Yolo v7 showed some limitations, including the point of detection in only a 2D plane, the higher computational power required to enable detection, and the difficulty of performing movements requiring more than one degree of Freedom (DOF). Nevertheless, this study highlights the detection capabilities of AI approaches, showing be a promising approach for measuring angles in rehabilitation activities, representing a cost-effective and easy-to-implement solution.

Abstract
Hybrid Energy Storage Systems (HESS) have attracted attention in recent years, promising to outperform single batteries in some applications. This can be in decreasing the total cost of ownership, extending the combined lifetime, having higher versatility in providing multiple services, and reducing the physical hosting location. The sizing of hybrid systems in such a way that proves to optimally replace a single battery is a challenging task. This is particularly true if such a tool is expected to be a practical one, applicable to different inputs and which can provide a range of optimal solutions for decision makers as a support. This article provides exactly that, presenting a technology -independent sizing model for Hybrid Energy Storage Systems. The model introduces a three-step algorithm: the first block employs a clustering of time series using Dynamic Time Warping (DTW), to analyze the most recurring pattern. The second block optimizes the battery dispatch using Linear Programming (LP). Lastly, the third block identifies an optimal hybridization area for battery size configuration (H indicator), and offers practical insights for commercial technology selection. The model is applied to a real dataset from an office building to verify the tool and provides viable and non-viable hybridization sizing examples. For validation, the tool was compared to a full optimization approach and results are consistent both for the single battery sizing, as well as for confirming the hybrid combination dimensioning. The optimal solution potential (H) in the example provided is 0.13 and the algorithm takes a total of 30s to run a full year of data. The model is a Pythonbased tool, which is openly accessible on GitHub, to support and encourage further developments and use.