
Apresenta-se um modelo de Rede Neural Artificial (RNA) do tipo Perceptron Multicamadas (MLP - Multilayer Perceptron) para um problema de classificação binária, visando predizer a probabilidade de um adolescente entre 12 e 16 anos apresentarem Transtorno do Espectro Autista (TEA). O modelo foi desenvolvido com técnicas de aprendizado profundo (Deep Learning) e treinado com 70% de um conjunto de dados contendo 248 participantes, cada um com 17 atributos. Os dados de entrada referem-se a características comportamentais extraídas do questionário AQ-10 paraadolescentes — obtido do repositório de Machine Learning da Universidade da Califórnia em Irvine (UCI) — acrescidas desete caraterísticas sociodemográficas. A implementação do algoritmo foi realizada em Python, utilizando a API Keras da biblioteca TensorFlow. Adicionalmente, com o objetivo de ampliar o acesso, o modelo foi implantado em um aplicativo web de acesso público denominado TRIA-TEA Adolescente, que funciona como uma ferramenta de triagem inicial para identificação de possíveis traços de TEA. Desenvolvido em português, o aplicativo visa facilitar o uso por pais, educadorese profissionais de saúde.
The Traveling Salesman Problem (TSP) has been documented in the history of combinatorial optimization since the 18th century. TSP is a well-known NP-hard problem that has been extensively studied, where a salesman aims to visit each of a set of cities exactly once and return to the starting city with minimal distance traveled. The significance of the TSP lies in its applicability to many real-life scenarios. Machine Learning (ML) addresses the question of how tobuild computers that improve automatically through experience. In the last two decades, ML has been applied in many areas. The scope of this paper is to bridge this gap by providing a comprehensive survey of ML usage in TSP. The paper is organized around the following topics: an introduction to the subject, TSP definition, variations and some related works, ML definition and methods, the research methodology used, and results. Additionally, we discuss certain open problems and potential research directions.
This article presents a literature review on the identification and mitigation of non-line-of-sight conditions in UWBnetworks, with a focus on edge computing. The difference from previous studies lies in the emphasis on solutionsthat can be executed directly on embedded devices, considering computational resource constraints. The search wasconducted in March 2025 in the IEEE Xplore, ScienceDirect, Scopus, Taylor & Francis, Web of Science, and Wiley OnlineLibrary databases. Initially, 1,050 articles were identified. After removing duplicates, screening abstracts, and readingthe full text, 17 studies aligned with the scope of the review remained. The results indicate the predominance of researchaimed at simplifying features and reduced models. Several studies mention suitability for edge computing, but do notpresent evaluation on restricted hardware devices. The overview obtained highlights the need to move towards solutionsthat close the loop between detection and action, evaluating latency, consumption, and memory directly at the edge.
This study develops an enhanced anomaly detection framework that integrates state-of-the-art deep learning techniques. The system detects hate speech, malicious websites, vulnerabilities in open ports, and suspicious processes. It optimizes detection accuracy and reliability by leveraging tailored datasets, advanced model architectures, and refined preprocessing strategies. To ensure robustness, it employs comprehensive validation metrics, including the Polygon Area Metric (PAM). Empirical results show significant improvements in generalization and prediction accuracy over previous methodologies. The hate speech detection model achieved an accuracy of 88%, with precision of 75%, recall of 62%, F1 score of 62%, and ROC AUC of 92%. These results deliver a proactive monitoring solution suited for modern enterprise security challenges.
Pre-trained contextual language models have demonstrated excellent performance in training with data from new languages and tasks. For this purpose, additional pre-training is necessary, since the lack of vocabulary of the language always tends to degrade the results. In this article, we present a procedure to deal with and treat unknown languages or languages without available resources such as Kicongo, a Bantu matrix language, commonly spoken in the northern regions of Angola, with greater incidence in rural regions and in the countries of central Africa (Democratic Republic of Congo, Republic of Congo and Gabon) and throughout the world. With the extension of Natural Language Processing (NLP) models based on the Bidirectional Encoder Representations from Transformers (BERT) architecture (BERT, RoBERTa and DistilBERT), the performance evaluation of the models was carried out using metrics such as accuracy, precision, recall and f1-score, achieving a performance rate higher than 97%.