Julio Cesar Duarte

This study analyzes, via a systematic review, how the integration of BIM and ML can improve time and risk management in construction projects. The PRISMA protocol was applied to searches in Scopus and CAPES Journals, starting from 677 records and yielding 54 articles aligned with BIM, ML, schedule/time management, and risk management. Findings show a predominance of supervised and unsupervised models, Deep Learning, Natural Language Processing (NLP), optimization algorithms, and hybrid approaches applied to schedule automation, deadline prediction, progress monitoring, and risk assessment. Identified benefits include improved accuracy, automation, and decision support, but challenges persist—interoperability, data quality, and training—alongside a lack of empirical validation in real-world contexts. We conclude that BIM functions as an integrating core within a data-driven ecosystem (AI, ML, and emerging digital technologies) to reduce uncertainty across the project lifecycle.

Este trabalho aborda o desafio da detecção e classificação de armas de fogo, pelo som emitido no disparo, utilizando uma rede neural convolucional (CNN). A pesquisa foi motivada pela sua aplicabilidade em segurança pública, forças armadas e perícia criminal, em que a rápida detecção e identificação do armamento podem representar um diferencial importante, permitindo uma rápida e eficaz resposta das autoridades. Para isso, foram criadas 4 arquiteturas CNN para executar as tarefas de detecção de disparos, classificação por tipo e por calibre e identificação da arma. Os resultados alcançados no trabalho mostram-se promissores, alcançando acurácia de 100 % na detecção, 95,03 % para classificação por tipo, 77,64 % na classificação por calibre e 47,58 % na classificação por armamento.

Generating data for radar applications, such as range-Doppler maps, is often expensive and time-consuming, and available datasets are typically small, which limits the effectiveness of deep learning models and can significantly hinder performance in tasks like target classification. In this study, we analyzed the feasibility of using Generative Adversarial Networks (GANs), particularly a WGAN-GP with spectral normalization, to generate synthetic range-Doppler map images and expand data availability. Despite our efforts, the results showed that the WGAN-GP struggled to capture the dataset’s complexity, with the discriminator being excessively stronger than the generator. This imbalance led to low visual quality in the generated images and poor performance in the Geometric Score (GS) and SSIM metrics. These findings highlight the challenges of training GANs in complex scenarios and the need for more robust methodologies for future data augmentation applications in radar systems.

A small language model (SLM)-based workflow was developed to analyze over 5,000 Brazilian Symposium of Theoretical Chemistry (SBQT) one-page abstracts (1981-2023). Brazilian theoretical and computational chemistry expanded strongly until the 2000s, then declined and stabilized in the 2010s and early 2020s. The number of abstracts and participants has stabilized at around 300 since 1997. Research funding has been dominated by the Brazilian agencies National Council for Scientific and Technological Development (CNPq) and Agency for Support and Evaluation of Graduate Education (CAPES), but state foundations have made relevant contributions. The most popular theoretical methods over time include configuration interaction (CI), molecular dynamics (MD), and density functional theory (DFT). The initial applications of DFT followed the rest of the world, but investigations using the most advanced exchange-correlation functionals and dispersion corrections lag by 5-8 years in some groups. Local development of methods and software is very incipient. It was identified significant delays in Brazilian theoretical and computational chemistry, relative to the rest of the world, in machine learning techniques, automated workflows, high-performance graphics processing unit (GPU) architectures, and benchmarking. As most studies focus on molecular electronic structure calculations, the most widely used software is for this type of research. The most studied subjects are intra- and inter‑proton, electron, and charge-transfer chemical phenomena. Hydrogen bonding and van der Waals interactions have also been widely studied, with water and benzene being the most investigated molecules. These findings highlight possible pathways for Brazilian theoretical chemistry to increase its reach. In particular, data and code repositories that enable reproducible workflows, which are largely absent, would increase reach and comply with the highly desirable findable, accessible, interoperable, and reusable (FAIR) practices.

Accurately localizing drones in complex environments remains a significant challenge, with important implications for defense, law enforcement, and autonomous systems. This study addresses the problem of estimating drone localization in environments characterized by strong reflections and noise. We employ Time Difference of Arrival (TDOA) techniques for localization estimation and compare them with a specialized machine learning regression model. While previous works have considered neural networks for audio source localization, they often suffer from limited generalization across different environments. To address this, we propose a novel method that enhances the TDOA vector by incorporating both primary and secondary peaks of the cross-correlation, guided by the Zero Cyclic Sum condition. Additionally, we introduce optimization strategies that selectively reduce the number of TDOA inputs based on a least-squares cost function. We present a comparative analysis of TDOA-based optimization techniques with a machine learning method that utilizes the environment's reverberation fingerprint as input features for training. Experimental results demonstrate that the proposed TDOA-based method achieves a localization accuracy of 0.55±0.35 meters, showcasing its effectiveness and practical applicability in challenging acoustic environments.

Advances in multimedia systems and computer vision have enabled each time more sophisticated methods that are able to analyze complex visual patterns in industrial contexts. The inspection of failures in metallic alloys is important for sectors such as aerospace and defense, yet it faces limitations due to reliance on manual methods and data scarcity. This research presents a systematic approach for selecting convolutional neural network architectures by integrating data augmentation, image segmentation, and self-supervised learning. In particular, it uses the DINO technique to generate interpretable attention maps, enhancing model transparency and performance in defect detection tasks. The method was validated using three architectures, ResNetInceptionV2, DenseNet, and Xception, with DenseNet combined with DINO achieving the best results. Key contributions include the release of a hybrid dataset (real and synthetic), effective strategies for handling class imbalance, and a comparative study of architectures, offering insights for future industrial applications and deep learning research applied to pattern detection on metallic surfaces.

The field of narratology has long worked on different ways to classify characters according to their narrative role and importance; meanwhile, studies on character design have observed the existence of unclear patterns that correlate specific visual features and those narrative classifications. Therefore, it becomes worth considering if modern image classification methods can be used to not only achieve satisfactory performance on this task, but also help us learn about the patterns that shape it. The task of classifying characters carries a vast array of practical uses, such as creating recommender systems based on specific visual-narrative correlations, automatic annotation for media preservation, adaptive interactive storytelling based on how a character is perceived, and the creation of education tools on how those patterns change depending on culture. In this work, we construct a dataset customized to assessing both performance and visual–narrative correlations within a specific medium, conduct an anonymous survey to evaluate human performance and tendencies, and compare the results with those of a traditional CNN classifier and a Foundation Model–augmented approach. Both models outperform the survey participants, with the latter achieving 76.93% accuracy, 18.11% higher than human performance, while also providing insights into the visual cues most closely tied to the given narrative roles.