Doutorado em Ciência da Computação

URI Permanente para esta coleçãohttp://200.137.215.59/tede/handle/tede/9159

Navegar

Agora exibindo 1 - 20 de 50

Machine learning algorithms for detecting structured fraud and combating ICMS tax evasion
(Universidade Federal de Goiás, 2025-10-31) Silva, Douglas Bernardes; Silva, Nádia Félix Felipe da; http://lattes.cnpq.br/7864834001694765; Carvalho, Sérgio Teixeira de Carvalho; http://lattes.cnpq.br/2721053239592051; Carvalho, Sérgio Teixeira de; http://lattes.cnpq.br/2721053239592051; Silva, Nádia Félix Felipe da; http://lattes.cnpq.br/7864834001694765; Bernardini, Flavia Cristina; http://lattes.cnpq.br/5935862634033333; Duarte, Kedma Batista; http://lattes.cnpq.br/7565342982159217; Rosa, Thierson Couto; http://lattes.cnpq.br/4414718560764818
Embargado
Desenvolvimento e avaliação de aplicações móveis em saúde: uma abordagem baseada em heurísticas de usabilidade e diretrizes de letramento digital em saúde
(Universidade Federal de Goiás, 2025-10-22) Oliveira, Hugo Miranda de; Berretta, Luciana de Oliveira; http://lattes.cnpq.br/0987947348533817; Carvalho, Sérgio Teixeira de; http://lattes.cnpq.br/2721053239592051; Carvalho, Sérgio Teixeira de; http://lattes.cnpq.br/2721053239592051; Viterbo Fillho, José; http://lattes.cnpq.br/8721187139726277; Mata, Luciana Regina Ferreira da; http://lattes.cnpq.br/2530837696657146; Fernandes, Deborah Silva Alves; http://lattes.cnpq.br/0380764911708235; Silva, Nádia Félix Felipe da; http://lattes.cnpq.br/7864834001694765
The advancement of mobile technologies has intensified the development of digital health solutions, yet a gap persists regarding the existence of systematic design processes that integrate usability heuristics, mobile device characteristics, and digital health literacy (DHL) guidelines. This gap can compromise the user experience, hinder the comprehension of clinical information, and reduce the effectiveness of use, especially among populations with low technological familiarity or health literacy limitations. This thesis proposes a structured design process specifically for the development and evaluation of mobile health applications, as well as the creation of the mHealth Application Usability Evaluation Tool (FAU-MH), the central artifact of this thesis. The FAU-MH automates and organizes heuristic evaluation, incorporating Human-Computer Interaction (HCI) and DHL criteria, thereby reducing analysis time and enhancing the rigor of evaluations compared to the traditional manual method. The IUProst application was used as a case study for the application and validation of both the design process and the FAUMH, establishing a real-world scenario that allowed for testing, comparing methods, and refining the proposed artifact. The results demonstrate that the tool and the process are capable of supporting evaluators and developers in constructing mobile solutions that are clearer, more accessible, and suitable for health self-care. The study adopts the Design Science Research (DSR) methodology as the foundation for structuring the conception, implementation, and evaluation stages. Thus, this research offers a robust and innovative methodological basis for designers, specialists, and developers, promoting the creation of more accessible, comprehensible, and efficient interfaces. This, in turn, is capable of increasing adherence, improving the user experience, and maximizing the positive impact of mobile technologies in the context of health care.
Arquitetura modular para navegação autônoma com aceleração de aprendizagem, fusão sensorial e comunicação colaborativa
(Universidade Federal de Goiás, 2025-10-08) Bezerra, Carlos Daniel de Sousa; Vieira, Flávio Henrique Teles; http://lattes.cnpq.br/0920629723928382; Vieira, Flávio Henrique Teles; http://lattes.cnpq.br/0920629723928382; Cardoso, Álisson Assis; http://lattes.cnpq.br/8216536516894987; Pinheiro Junior, Carlos Galvão; http://lattes.cnpq.br/6491947028449971; Fonseca, João Paulo da Silva; http://lattes.cnpq.br/5217261758266411; Brito, Leonardo da Cunha; http://lattes.cnpq.br/6660680440182900
This work proposes a modular architecture for autonomous navigation of mobile robots in dynamic environments, integrating functional modules: (i) intelligent control through deep reinforcement learning, (ii) fusion of visual and non-visual sensors, (iii) a visionbased anti-collision safety module, and (iv) collaborative hybrid communication. The proposed EKF-DQN algorithm accelerates the learning process by integrating state predictions provided by the Extended Kalman Filter. The sensor fusion system combines data from visual and non-visual sensors through strategies such as Late Fusion, allowing the agent to acquire a robust perception of the environment. For operational safety, a person detection module was implemented to identify collision risks. In the communication layer, a hybrid protocol is proposed that combines: (i) LoRa technology, known for its long range and low power consumption, used for communication with remote stations (R2I); and (ii) the DDS middleware, used for short-range robot-to-robot (R2R) communication, ensuring real-time synchronization and sensory data exchange. To enhance the robustness of LoRa connectivity, a new algorithm called “Forced Data Rate” was developed, which adaptively enforces the spreading factor based on link quality, outperforming the traditional Adaptive Data Rate (ADR) algorithm in scenarios with constant mobility. To enable the transmission of high-resolution sensory data, such as those generated by LiDAR sensors, a compression technique based on autoencoders was used, achieving an average payload size reduction of 82% without significant loss of accuracy. Results obtained from both simulations and real-world robot experiments demonstrated significant improvements in navigation performance. The proposed EKF-DQN achieved a 93.33% success rate and outperformed the traditional D3QN. With the integration of the safety module, a 72% success rate was achieved in environments with moving people.
Visualização do espaço de solução: uma análise exploratória em problemas de otimização
(Universidade Federal de Goiás, 2025-10-14) Silva, Ennio Willian Lima; Felix, Juliana Paula; http://lattes.cnpq.br/3610115951590691; Nascimento, Hugo Alexandre Dantas do; http://lattes.cnpq.br/2920005922426876; Nascimento, Hugo Alexandre Dantas do; http://lattes.cnpq.br/2920005922426876; Camilo Junior, Celso Gonçalves; http://lattes.cnpq.br/6776569904919279; Ferreira, Joelma de Moura; http://lattes.cnpq.br/3906491664088644; Aloise, Dario José; http://lattes.cnpq.br/7266011798625538; Corso, Gilberto; http://lattes.cnpq.br/0274040885278760
Complex real-world optimization problems often involve numerous variables, multiobjective functions, and conflicting constraints, and may even depend on subjective aspects. In this context, human-computer interaction (HCI) has been an alternative employed for solving such problems. This combination of human capabilities and algorithms is frequently utilized in decision support systems (DSS), where users can interact to improve the resulting solution. A common feature of interactive optimization processes is the use of Information Visualization tools to assist users in decision-making. In the context of optimization problems, visualizing the solution space is crucial for understanding the difficulty of solving a problem, the effectiveness of existing algorithms in exploring the space, and the user’s proximity to an optimal solution. From this perspective, the objective of this work is to systematize knowledge regarding solution space visualization and, based on this analysis, propose an interactive optimization framework that demonstrates the effectiveness of this approach in problem-solving.To achieve this objective, a systematic literature review (SLR) was conducted on the visualization of solution spaces in optimization problems. The articles identified in this review served as input for developing a tool that enables the interactive exploration of scientific literature on the subject. Additionally, a taxonomy was developed to classify interactive actions within the solution space, and a framework for interactive optimization based on its visualization was proposed. As a proof of concept, an interactive system based on an adaptation of this framework was implemented and applied in a case study on the seismic inversion problem, demonstrating the feasibility of the approach.The research provides contributions on multiple fronts, demonstrating that solution space visualization enhances problem understanding and aids in decision-making, thereby guiding the user through the optimization process.
Aprendizado contínuo com Multi-armed Bandits aplicado à predição de séries temporais no contexto de vendas no comércio varejista
(Universidade Federal de Goiás, 2025-10-27) Nogueira, Heber Valdo; Soares, Telma Woerle de Lima; http://lattes.cnpq.br/6296363436468330; Soares, Telma Woerle de Lima; http://lattes.cnpq.br/6296363436468330; Vieira, Flávio Henrique Teles; http://lattes.cnpq.br/0920629723928382; Sousa, Rafael Teixeira; http://lattes.cnpq.br/3819400092348829; Patto, Vinícius Sebba; http://lattes.cnpq.br/3585475958654532; Coelho, Clarimar José; http://lattes.cnpq.br/1350166605717268
Embargado
ARANDU: framework para geração aumentada por recuperação em grafos de conhecimento com fundamentação neuro-simbólica
(Universidade Federal de Goiás, 2025-10-22) Xavier, Otávio Calaça; Soares, Anderson da Silva; http://lattes.cnpq.br/1096941114079527; Soares, Anderson da Silva; http://lattes.cnpq.br/1096941114079527; Bulcão Neto, Renato de Freitas; http://lattes.cnpq.br/5627556088346425; Rosa, Thierson Couto; http://lattes.cnpq.br/4414718560764818; Carvalho, Cedric Luiz de; http://lattes.cnpq.br/4090131106212286; Costa, Gustavo de Assis; http://lattes.cnpq.br/1543798708473666
This work addresses the challenge of Knowledge Graph Question Answering (KGQA), a field transformed by the rise of Large Language Models (LLMs) but still facing limitations such as the generation of factually inconsistent information (``hallucinations'') and difficulty in performing complex reasoning. The central objective of this research was to develop and validate a neuro-symbolic architecture that overcomes the limitations of contemporary Retrieval-Augmented Generation (RAG) systems, aiming to integrally solve the challenges of (1) retrieving evidence with low precision and recall, (2) loss of structural context in communication with the LLM, and (3) the absence of explicit logical orchestration in the reasoning process. To this end, the ARANDU framework was designed, implemented, and made available as open source, materializing the proposed architecture. The methodology is divided into an offline preparation stage, where hybrid indexes (lexical and semantic) are created and logical rules are mined from the graph, and an online execution pipeline with three phases: I) Hybrid Evidence Retrieval, which extracts a cohesive subgraph by combining lexical, semantic, and graph-based structured retrieval; II) Logical Context Orchestration, which enriches the subgraph with logical rules and weights the most relevant inference paths; and III) Neural Representation and Generation, where a Graph Neural Network (GNN) encodes the subgraph into a vector representation (graph token) that, along with the textual context, conditions a compact LLM to generate the final answer. The empirical validation, conducted on the WebQSP and MetaQA datasets and compared with baselines such as NaiveRAG, GraphRAG, and G-Retriever, showed that ARANDU achieved superior performance in most scenarios, especially in multi-hop reasoning tasks, with significant improvements in ranking quality metrics like nDCG@10 and MRR. The results also confirmed that neural representation via GNN is more effective than textual linearization and that the architecture is computationally efficient. The research concludes that the synergy between optimized retrieval, logical orchestration, and neural representation, as implemented in ARANDU, constitutes a robust and effective solution that increases the fidelity and precision of answers in KGQA systems, thus validating the central hypothesis of this work.
Estimativa de diâmetro de troncos de eucalipto a partir de nuvens de pontos LiDAR de smartphone e Redes Neurais Profundas
(Universidade Federal de Goiás, 2025-10-21) Rodrigues, Welington Galvão; Vieira, Gabriel da Silva; http://lattes.cnpq.br/9290516928216163; Soares, Fabrizzio Alphonsus Alves de Melo Nunes; http://lattes.cnpq.br/7206645857721831; Soares, Fabrizzio Alphonsus Alves de Melo Nunes; http://lattes.cnpq.br/7206645857721831; Fernandes, Deborah Silva Alves; http://lattes.cnpq.br/0380764911708235; Cabacinha, Christian Dias; http://lattes.cnpq.br/7626216490380053; Pedrini, Helio; http://lattes.cnpq.br/9600140904712115; Costa, Ronaldo Martins da; http://lattes.cnpq.br/7080590204832262
Accurate measurement of dendrometric parameters, particularly the Diameter at Breast Height (DBH), is fundamental for forest inventory and sustainable management. However, traditional field methods are labor-intensive, timeconsuming, and prone to error. The integration of LiDAR sensors into consumer-grade smartphones offers a scalable and cost-effective alternative, yet requires robust computational methods to process the resulting 3D point cloud data. This thesis presents a novel, end-to-end framework for the automated measurement of DBH in eucalyptus trees by leveraging deep learning for 3D reconstruction and semantic segmentation. We develop and validate two distinct processing pipelines: one that reconstructs complete 360- degree tree models from rapid, partial scans using a point cloud completion architecture, and another that processes full-circle scans captured directly in the field. Both workflows converge into a shared segmentation stage where a PointTransformer network performs high-fidelity semantic segmentation to precisely isolate the DBH region from the tree stem. An automated algorithm then calculates the diameter from the segmented point cloud. Our results demonstrate that the proposed framework achieves high precision, with measurements showing no statistically significant difference from manual caliper methods. The direct-scan pipeline proved superior, achieving a RMSE of less than 0.55 cm across all tree classes. Critically, the methodology yields a transformative improvement in operational efficiency, reducing in-field data collection time by three to thirty-fold. By validating a high-precision, low-cost workflow, this research provides a significant step toward the automation of forest inventories, enabling more efficient and data-driven practices in precision forestry. To foster future research, two novel datasets containing 880 annotated tree scans are also made publicly available.
Proposta de arquitetura de AutoML para aprendizado de múltiplos estimadores de séries temporais
(Universidade Federal de Goiás, 2025-10-20) Santos, Danilo Turkievicz dos; Soares, Anderson da Silva; http://lattes.cnpq.br/1096941114079527; Soares, Anderson da Silva; http://lattes.cnpq.br/1096941114079527; Galvão Filho, Arlindo Rodrigues; http://lattes.cnpq.br/7744765287200890; Sousa, Rafael Teixeira; http://lattes.cnpq.br/3819400092348829; Fanucchi, Rodrigo Zempulski; http://lattes.cnpq.br/0937371415675134; Patto, Vinicius Sebba; http://lattes.cnpq.br/3585475958654532
Demand forecasting in the retail sector is a highly complex task, characterized by the vast heterogeneity of patterns across thousands of time series. Traditional approaches, such as custom single models, are costly and poorly scalable, while global foundational models still face challenges in practical applicability. In this context, this thesis proposes and develops an AutoML methodology that enhances the robustness and computational efficiency of predictions, based on the cluster-then-forecast strategy. The cornerstone of this methodology is a novel clustering approach that employs SOM using the LikelihoodDistance metric to identify series with similar underlying generative processes. The architecture was validated in a real-world and challenging scenario, using sales data from the pharmaceutical retail sector. The results demonstrated that the sampling-based approach, derived from the clustering, was particularly effective, successfully identifying the best-performing models from small samples of series. The central finding of the research is that the proposed architecture not only proved to be competitive across all stores and evaluated metrics but also exhibited remarkable consistency and reliability. Furthermore, the success of the clustering provides strong evidence that the similarity between the series' generative models is a determining factor in selecting the most accurate forecasting technique. This work thus contributes an AutoML framework that mitigates the model selection problem in heterogeneous environments, offering a more stable, scalable, and computationally viable forecasting solution for the retail sector.
Alocação dinâmica de recursos em fatias de redes IoT não-3GPP envolvendo VANTs
(Universidade Federal de Goiás, 2025-04-22) Silva, Rogério Sousa e; Oliveira Júnior, Antonio Carlos de; http://lattes.cnpq.br/3148813459575445; Cardoso, Kleber Vieira; http://lattes.cnpq.br/0268732896111424; Cardoso, Kleber Vieira; http://lattes.cnpq.br/0268732896111424; Oliveira Júnior, Antonio Carlos de; http://lattes.cnpq.br/3148813459575445; Abelém, Antônio Jorge Gomes; http://lattes.cnpq.br/5376253015721742; Both, Cristiano Bonato; http://lattes.cnpq.br/2658002010026792; Rocha, Flávio Geraldo Coelho; http://lattes.cnpq.br/5583470206347446
The exponential growth of the Internet of Things (IoT) has introduced increasing challenges to communication infrastructures, particularly in critical scenarios such as natural disasters and densely populated events, where network overload compromises service continuity and quality. In this context, this thesis presents a hybrid approach for dynamic resource allocation in non-3GPP IoT networks, integrating network slicing (NS), heterogeneous access (Multi-RAT), and unmanned aerial vehicles (UAVs) equipped with LoRaWAN gateways. The proposed hybrid approach synergistically combines the precision of exact optimization methods based on Mixed Integer Linear Programming (MILP), employed for determining the optimal initial positioning of UAVs, with the adaptive flexibility of advanced Deep Reinforcement Learning (DRL) algorithms, which enable dynamic and autonomous repositioning in variable environments. The goal of the first stage is to minimize operational and deployment costs while maximizing Quality of Service (QoS), while the second stage is to facilitate the autonomous repositioning of UAVs in response to environmental changes and fluctuations in network demand. We develop and assess four DRL algorithms, e.g., SR-DQN, DA-DDDQN, NSE-A2C, and RG2E-PPO. The proposed solutions were validated through realistic simulations using the ns-3 network simulator, in customized scenarios with non-3GPP connectivity. Results demonstrated significant improvements in QoS, reduced number of deployed UAVs, enhanced decision robustness, and increased spectral efficiency, with notable performance from the NSE-A2C and RG2E-PPO algorithms. The hybrid approach enables the creation of a mobile, scalable, and resilient communication infrastructure capable of autonomously and efficiently addressing the specific requirements of diverse IoT applications, particularly in urban and emergency environments with critical connectivity constraints. This thesis contributes to the state of the art by proposing a replicable, sustainable, and service-oriented hybrid architecture for reliable communication in heterogeneous and dynamic networks based on unlicensed technologies. Potential applications include smart cities, disaster response, and temporary connectivity deployment in degraded or non-existent infrastructure scenarios.
Decomposição de tarefas para problemas de linguagem natural: segmentação de hashtags e anotação de texto argumentativo
(Universidade Federal de Goiás, 2025-04-24) Inuzuka, Marcelo Akira; Silva, Nádia Félix Felipe da; http://lattes.cnpq.br/7864834001694765; Nascimento, Hugo Alexandre Dantas do; http://lattes.cnpq.br/2920005922426876; Nascimento, Hugo Alexandre Dantas do; http://lattes.cnpq.br/2920005922426876; Martins, Wellington Santos; http://lattes.cnpq.br/3041686206689904; Dias, Márcio de Souza; http://lattes.cnpq.br/0095510023252013; Alencar, Wanderley de Souza; http://lattes.cnpq.br/5491185436975801; Rosa, Thierson Couto; http://lattes.cnpq.br/4414718560764818
Corpus annotation is essential for training Natural Language Processing (NLP) models, yet it faces challenges such as high cognitive complexity, annotator inconsistency, and elevated costs. This thesis proposes task decomposition as a methodological strategy to modularize complex NLP processes, promoting greater conceptual clarity, scalability, and reproducibility. Initially focused on Argument Mapping, the research redirected its scope due to the infeasibility of the original task, concentrating on the identification of reusable patterns applicable to annotation and automation stages. Guidelines, a hierarchical decomposition algorithm, and artifacts such as annotated datasets and the Argmap platform — which supports collaborative annotation with quality control — were developed. The approach was validated through three empirical case studies: hashtag segmentation, keyphrase curation, and annotation of argumentative structures. Results demonstrate that decomposition improves consistency among agents (human or automatic), guideline clarity, and automation feasibility. The thesis also introduces the Recruiter–Selector architectural pattern, which structures tasks into two independent modules — candidate generation and final selection — applicable to both annotation workflows and algorithms based on Large Language Models (LLMs). It concludes that decomposition driven by reusable patterns enhances efficiency and reliability in corpus construction and the development of robust NLP systems, contributing to the systematization of annotation processes and their integration with automatic solutions
Preenchendo a Lacuna da Supervisão Humana na Robótica por Meio da Linguagem: Uma IA Agêntica Multimodal para Escalar a Supervisão Humana em Sistemas Autônomos
(Universidade Federal de Goiás, 2025-04-22) Santos, Lara Fernanda Portilho dos; Soares, Anderson da Silva; Soares, Anderson Da Silva; Laureano, Gustavo Teodoro; Salazar, Aldo Andre Diaz; Colombini, Esther Luna; Vieira, Flavio Henrique Teles
While many autonomous systems can navigate and avoid obstacles under predictable conditions, they often rely on a human supervisor (Human-in-the-Loop - HITL) to adapt to adverse obstacles, sudden layout modifications, or partial hardware failures. However, existing HITL strategies frequently leave operators struggling with large volumes of data that demand real-time interpretation. To mitigate these challenges, we propose an agentic AI approach that integrates long-term memory with adaptive reasoning techniques, thereby reducing operator workload and minimizing disruptions in dynamic autonomous robotics operations. The proposed system incorporates hierarchical subagents to systematically integrate historical data, sensor logs, and iterative problem-solving techniques to address frequent challenges in multi-robot deployments, including localization failures, hardware malfunctions, and crowd-induced obstacles. Experimental evaluations comparing memory-augmented and baseline (no-memory) conditions reveal that its usage consistently yields higher solution accuracy and operator satisfaction. In particular, memory retrieval accelerates the resolution of recurring failure modes, while adaptive reasoning enhances real-time decision-making in novel or crowded scenarios. Text-based similarity metrics (Token Overlap and Semantic Alignment) further demonstrate that reusing verified domain language and strategies improves the clarity and maintainability of the recommended actions. The results underscore the viability of a modular, language-based system that combines data-driven diagnostics, robust memory mechanisms, and self-reflective planning for large-scale robot supervision. By uniting flexible LLM capabilities with HITL workflows, our proposal holds considerable promise for improving both efficiency and transparency in real-world autonomous robotics operations.
MDMWare: model-driven domain-specific middleware for smart cities
(Universidade Federal de Goiás, 2025-03-10) Melo, Paulo César Ferreira; Costa, Fábio Moreira; http://lattes.cnpq.br/0925150626762308; Costa, Fábio Moreira; Sampaio Junior, Adalberto Ribeiro; Lucrédio, Daniel; Carvalho, Sérgio Teixeira de; Graciano Neto, Valdemar Vicente
Embargado
Técnicas de reamostragem e super-resolução em imagens de culturas agrícolas
(Universidade Federal de Goiás, 2025-02-28) Nogueira, Emília Alves; Soares, Fabrízzio Alphonsus Alves de Melo Nunes; http://lattes.cnpq.br/7206645857721831; Soares, Fabrizzio Alphonsus Alves de Melo Nunes; Pedrini, Helio; Cabacinha, Christian Dias; Costa, Ronaldo Martins da; Fernandes, Deborah Silva Alves
The increasing demand for food, coupled with climate change, has driven the development of agricultural monitoring technologies to increase the efficiency and sustainability of crop production such as sugarcane and corn. However, the low resolution of images captured by Unmanned Aerial Vehicle (UAV) and satellites limits the detailed analysis of essential agronomic features. This thesis investigates methods to improve the resolution of agricultural images, comparing Traditional Resampling Techniques (TRT) with Super-Resolution with Deep Networks (SRDN) algorithms, such as Real Enhanced Super-Resolution Generative Adversarial Network (Real-ESRGAN), Multi-Level upscaling Transform (MuLUT) and Learning Resampling Function (LeRF). The aim of this study is to investigate the application of deep learning techniques to improve the resolution of agricultural images. For this purpose, existing methods were reviewed and an agricultural dataset was prepared. The research adopted an experimental approach, evaluating the methods quantitatively using metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), and qualitatively by visual analysis. The experiments demonstrate significant improvements in image resolution using the SRDN algorithms compared to TRT, with gains of 484.34% in sugarcane images, 234.4% in corn, and 58.57% in satellite images. Although the SRDN techniques were developed for other purposes, such as improving the resolution of images of people and anime, their performance can be observed in agricultural images. The results obtained are significant for precision agriculture, since the increase in image resolution can aid in monitoring plant growth and health, providing faster and more effective interventions. In future investigations, we hope to expand the comparisons with other SRDN algorithms.
Aplicação de CNN e LLM na Localização de Defeitos de Software
(Universidade Federal de Goiás, 2024-10-16) Basílio Neto, Altino Dantas; Camilo Júnior, Celso Gonçalves; http://lattes.cnpq.br/6776569904919279; Camilo Junior, Celso Gonçalves; Leitão Júnior , Plínio de Sá; Oliveira, Sávio Salvarino Teles de; Vincenzi, Auri Marcelo Rizzo; Souza, Jerffeson Teixeira de
The increase in the quantity or complexity of computational systems has led to a growth in the occurrence of software defects. The industry invests significant amounts in code debugging, and a considerable portion of the cost is associated with the task of locating the element responsible for the defect. Automated techniques for fault localization have been widely explored, with recent advances driven by the use of deep learning models that combine different types of information about defective source code. However, the accuracy of these techniques still has room for improvement, suggesting open challenges in the field. This work aims to formalize and investigate the most impactful aspects of fault localization techniques, proposing a framework for characterizing approaches to the problem and two solution methodologies: a) based on convolutional neural networks (CNNs) and b) based on large language models (LLMs). From experimentation involving public datasets in Java and Python, it was demonstrated that CNNs are comparable to traditional methods but were found to be inferior to other methods in the literature. The LLM-based approach, on the other hand, greatly outperformed heuristics like Ochiai and Tarantula and proved competitive with more recent literature. An experiment in a scenario free from the data leakage problem showed that LLM-based approaches can be improved by combining them with the Ochiai heuristic.
Deep Learning aplicado à classificação em nível de pixel de variedades de culturas por imagens multiespectrais
(Universidade Federal de Goiás, 2024-11-06) Kai, Priscila Marques; Oliveira, Bruna Mendes de; Costa, Ronaldo Martins da; http://lattes.cnpq.br/7080590204832262; Costa, Ronaldo Martins da; Soares, Fabrízzio Alphonsus Alves de Melo Nunes; Leitão Júnior, Plínio de Sá; Arraut, Eduardo Moraes; Costa, Kelton Augusto Pontara da
The classification of different crop varieties still faces significant challenges due to their similar spectral characteristics. To address this issue, the integration of remote sensing techniques with deep learning methods offers a promising solution by analyzing pixel-level data based on spectral bands, band combinations, and vegetation indices. In this study, we developed a cross-deep neural network methodology, referred to as DCN-S, with a case study focused on the classification of sugarcane varieties. The methodology was applied to remote sensing data from cultivation areas in the state of Goiás, Brazil, collected between 2019 and 2021. The DCN-S model was compared with traditional classifiers, such as k-Nearest Neighbors (kNN), Support Vector Machines (SVM), and Random Forest, as well as other neural network configurations. The results indicated that the DCN-S model achieved competitive accuracy in validation scenarios, including temporal variety considerations when compared to other studies in the literature. Moreover, the model excelled in classifying varieties without requiring the separation of developmental stages, surpassing traditional methods. Performance improvements were further observed after applying a voting process. Finally, this work’s main contributions include developing an approach for classifying agricultural varieties by combining deep learning with remote sensing data and validating this methodology in a practical scenario. The results highlight the potential of the DCN-S model to outperform traditional techniques, offering a tool for automated agricultural monitoring
Métodos para Análise de Dano Foliar e Reconhecimento de Pragas na Agricultura Usando Técnicas Computacionais
(Universidade Federal de Goiás, 2024-08-02) Vieira, Gabriel da Silva; Soares, Fabrizzio Alphonsus Alves De Melo Nunes; http://lattes.cnpq.br/7206645857721831; Soares, Fabrizzio Alphonsus Alves De Melo Nunes; Cabacinha, Christian Dias; Pedrini, Helio; Laureano, Gustavo Teodoro; Costa, Ronaldo Martins Da
The application of computer techniques in agriculture has significantly improved rural activities, particularly crop monitoring, plant protection, and overall yield. This thesis emphasizes leaf analysis as a valuable tool for inspecting and continually improving plantations, as well as supporting decision-making and agricultural management interventions. Changes in leaves can lead to irreparable losses in productivity, the delivery of low-quality products, and significant economic impacts. To prevent production failures, it is crucial to efficiently monitor and identify whether pests are affecting productivity or remaining within acceptable levels. However, damage to the leaf silhouette can limit automated analysis, and the diversity in leaf shape and damage levels makes it challenging to delineate the compromised edge regions. This study introduces original computer-based methods for defoliation estimate, damage detection, leaf surface reconstruction, and pest classification that are prepared to address damage to the leaf boundaries. Notable aspects of this study include template matching for pattern recognition and pest classification using only traces of leaf damage. The methodological design of the study consists of a literature review, investigation of digital image processing techniques, computer vision and machine learning, software development, and formulation of experimental tests. The results indicate high accuracy in estimating leaf area loss with a linear correlation of 0.98, damage detection and pest classification with assertiveness above 90%, and visual restoration of regions affected by herbivory with SSIM scores between 0.68 and 0.94.
Redes de telecomunicações convergentes: modelagem e implementação de arquitetura para infraestruturas pós-5G
(Universidade Federal de Goiás, 2024-09-25) Macedo, Ciro José Almeida; Both, Cristiano Bonato; http://lattes.cnpq.br/2658002010026792; Cardoso, Kleber Vieira; http://lattes.cnpq.br/0268732896111424; Cardoso, Kleber Vieira; Martins, Joberto Sérgio Barbosa; Oliveira Júnior, Antonio Carlos de; Klautau Júnior, Aldebaro Barreto da Rocha; Alberti, Antônio Marcos
The evolution of cellular mobile networks has been guided by directives specified by institutions such as 3GPP, aimed at supporting demanding services. Simultaneously, non3GPP wireless communication technologies have also evolved and play a significant role in various contexts. These technologies are essential for connectivity in diverse scenarios where long-range communication and low power consumption are crucial. Recent studies have shown that the integration and harmonious coexistence of 3GPP and non-3GPP technologies are vital in the context of post-5G networks, enhancing ubiquitous and seamless connectivity. In this context, the present thesis investigated the feasibility of converging non-3GPP communication technologies with the 5G core, dividing the investigation into two phases. In the first phase, an architecture was proposed to integrate these technologies. In the second phase, a functional prototype of this architecture was built to conduct experiments demonstrating its viability in different use cases. The thesis conducted a detailed technical analysis, offering a comprehensive view of the benefits of convergence for consumers and infrastructure providers. Significant gaps were identified that still need to be addressed in post-5G/6G networks, such as the current inability to monitor non-3GPP networks by the 5G infrastructure operator. Some of these gaps were explored and investigated in the context of the solution proposed in this thesis
MemoryGraph: uma proposta de memória para agentes conversacionais utilizando grafo de conhecimento
(Universidade Federal de Goiás, 2024-09-25) Oliveira, Vinicius Paulo Lopes de; Soares, Anderson da Silva; http://lattes.cnpq.br/1096941114079527; Soares, Anderson da Silva; Galvão Filho, Arlindo Rodrigues; Silva, Nadia Felix Felipe da; Pereira, Fabíola Souza Fernandes; Fanucchi, Rodrigo Zempullski
With the advancement of massive language models applied to natural language processing, many proposals have become viable, such as the use of conversational agents applied to various everyday tasks. However, these models still have limitations in both the integration of new knowledge and the representation and retrieval of that knowledge, being constrained by costs, execution time, and training. Furthermore, their black-box nature prevents the direct manipulation of knowledge, mainly due to the vector representation that indirectly represents it, making the control and explanation of their results more difficult. In contrast, knowledge graphs allow for a rich and explicit representation of relationships between real-world entities. Despite the challenges in their construction, studies indicate that these can complement each other to produce better results. Therefore, the objective of this research is to propose a memory system for conversational agents based on massive language models through the combination of explicit knowledge (knowledge graphs) and implicit knowledge (language models) to achieve better semantic and lexical representation. This methodology was called MemoryGraph and is composed of three processes: graph construction, graph search, and user representation. Various knowledge graph construction workflows were proposed and compared, considering their costs and influences on the final result. The agent can search for information in this base through various search proposals based on RAG, referred to here as GraphRAG. This search methodology was evaluated by humans in five proposed question scenarios, showing superior average results in all five proposed search approaches (29% in the best approach). In addition, six RAG metrics, evaluated by a massive model, were applied to the proposed application results from two popular datasets and one composed of diabetes guidelines, showing superior results in all datasets. Furthermore, a method for long-term user representation, called user_memory, was proposed, demonstrating 93% retention of user information. To reinforce this result, case studies were conducted, demonstrating the agent's ability to personalize the user experience based on past experiences, increasing the speed of information delivery and user satisfaction. The results demonstrate that the MemoryGraph paradigm represents an advance over vector representation in environments where richer, temporal, and mutable contextualization is necessary. It also indicates that the integration of knowledge graphs with massive language models, especially in the construction of long-term memory and rich contextualization based on past experiences, can represent a significant advance in creating more efficient, personalized conversational agents with enhanced capacity for retaining and utilizing information over time.
IVF/NSGA-III: Uma Metaheurística Evolucionária Many-Objective com Busca Guiada por Balizas e Fertilização In Vitro
(Universidade Federal de Goiás, 2024-04-11) Sampaio, Sávio Menezes; Camilo Junior, Celso Gonçalves; http://lattes.cnpq.br/6776569904919279; Camilo Junior; Camilo Junior, Celso Gonçalves; Lima Neto, Fernando Buarque de; Leite, Karla Tereza Figueiredo; Rodrigues, Vagner José do Sacramento; Oliveira, Sávio Salvarino Teles de
Sampaio, Sávio Menezes. The In Vitro Fertilization Genetic Algorithm (IVF/GA) demonstrates robust applicability to single-objective optimization problems, particularly those that are complex and multimodal. This work proposes the expansion of the IVF method to many-objective optimization, which deals with more than three simultaneous objectives. The study introduces new activation criteria, selection, assisted exploration, and transfer mechanisms, consolidating innovation through the integration of the IVF method with NSGA-III, here referred to as IVF/NSGA-III. This approach incorporates the Beacon-Guided Search strategy in a Steady State configuration, aiming to overcome the inherent challenges of many-objective optimization. It focuses on dynamic convergence to promising regions of the solution space and adopts an adaptive scale factor within the context of Differential Evolution, providing an alternative methodology to conventional intensification methods. Experiments conducted with the many-objective benchmarks DTLZ, MaF, WFG show that IVF/NSGA-III significantly enhances performance compared to the standard NSGA-III algorithm across various tested problems, validating its potential as a valuable contribution to the field of Many-Objective Evolutionary Algorithms (MOEAs). The study suggests new directions for the development of many-objective memetic strategies and offers significant insights for researchers seeking more effective and adaptable optimization methods.. Goiânia-GO, 2024. 220p. PhD. Thesis Relatório de Graduação. Instituto de Informática, Universidade Federal de Goiás.
Framework para sistemas de recomendação baseados em neural contextual Bandits com restrição de justiça
(Universidade Federal de Goiás, 2024-06-03) Santana, Marlesson Rodrigues Oliveira de; Soares, Anderson da Silva; http://lattes.cnpq.br/1096941114079527; Soares, Anderson da Silva; Rosa, Thierson Couto; Carvalho, Cedric Luiz De; Araújo, Aluizio Fausto Ribeiro; Veloso, Adriano
The advent of digital businesses such as marketplaces, in which a company mediates a commercial transaction between different actors, presents challenges to recommendation systems as it is a multi-stakeholder scenario. In this scenario, the recommendation must meet conflicting objectives between the parties, such as relevance versus exposure, for example. State-of-the-art models that address the problem in a supervised way not only assume that the recommendation is a stationary problem, but are also user-centered, which leads to long-term system degradation. This thesis focuses on modeling the recommendation system as a reinforcement learning problem, through a Markovian decision-making process with uncertainty where it is possible to model the different interests of stakeholders in an environment with fairness constraints. The main challenges are the need for real interactions between stakeholders and the recommendation system in a continuous cycle of events that enables the scenario for online learning. For the development of this work, we present a model proposal, based on Neural Contextual Bandits with fairness constrain for multi-stakeholder scenarios. As results, we present the construction of MARS-Gym, a framework for modeling, training and evaluating recommendation systems based on reinforcement learning, and the development of different recommendation policies with fairness control adaptable to Neural models. Contextual Bandits, which led to an increase in fairness metrics for all scenarios presented while controlling the reduction in relevance metrics.

Navegar

Submissões Recentes