Murilo Coelho Naldi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Murilo Coelho Naldi has a total of 30 co-authors. Here are the top co-authors by number of shared publications:
- Alan Demétrius Baria Valejo
Universidade Federal de São Carlos
5 shared publications - Guilherme Henrique Messias
Universidade Federal de São Carlos
3 shared publications - Natanael Fabrício Dacioli Batista
Universidade Federal de São Carlos
3 shared publications - Tiago Gonçalves Botelho
Universidade Federal de São Carlos
3 shared publications - Bruno Leonel Nunes
Federal University of São Carlos (UFSCar) Rod
2 shared publications - Elaine Ribeiro Faria
Federal University of Uberlândia
2 shared publications
Latest Publications
- 30 September 2026
- Published by Springer Nature
- p. 262-276
DensGNN: A Density Based Graph Neural Network for Node Clustering in Applications with Noise
- 30 September 2026
- Published by Springer Nature
- p. 262-276
DensGNN: A Density Based Graph Neural Network for Node Clustering in Applications with Noise
- 30 September 2026
- Published by Springer Nature
- p. 18-33
Contrastive Deep Graph Clustering with Dynamic Cluster Estimation
- 30 September 2026
- Published by Springer Nature
- p. 18-33
Contrastive Deep Graph Clustering with Dynamic Cluster Estimation
- 30 January 2026
- Published by Springer Nature
- p. 270-284
Combining Large Language Model Embeddings and Graph Neural Networks for Accurate Job Recommendations
- 30 January 2026
- Published by Springer Nature
- p. 270-284
Combining Large Language Model Embeddings and Graph Neural Networks for Accurate Job Recommendations
- Open Access
- 19 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 215265-215285
A Systematic Review of Graph Neural Network for Node Clustering in Homogeneous Attributed Networks
The objective of node clustering is to identify distinct groups of vertices in a graph, ensuring that similar nodes are grouped together while dissimilar ones remain apart. Building on the success of graph neural networks (GNNs) in tasks such as node classification and link prediction, several clustering strategies have been developed. Although these methods are highly effective, many share similar strengths and weaknesses, which are often difficult to discern in the absence of a clear taxonomy. Previous studies have attempted to categorize node clustering algorithms; however, most are not restricted to GNN-based methods and instead encompass a wide variety of graph learning techniques. Furthermore, despite significant progress in community detection, valuable insights from this area remain underexplored in the context of GNN-based clustering. To bridge this gap and provide a structured overview of the field, we conducted a Systematic Literature Review (SLR) on node clustering in homogeneous networks using GNNs. A total of 84 articles were reviewed based on 20 inclusion and exclusion criteria. From this analysis, we propose a comprehensive taxonomy that integrates both deep clustering and node embedding models. The findings reveal key research gaps, including the absence of transformer-based models, the high computational cost of some approaches, and dataset bias arising from the over-reliance on widely used benchmarks. The proposed taxonomy offers a clearer framework for understanding existing methodologies and guiding future research directions in GNN-based node clustering.
- 19 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 215265-215285
Open AccessA Systematic Review of Graph Neural Network for Node Clustering in Homogeneous Attributed Networks
The objective of node clustering is to identify distinct groups of vertices in a graph, ensuring that similar nodes are grouped together while dissimilar ones remain apart. Building on the success of graph neural networks (GNNs) in tasks such as node classification and link prediction, several clustering strategies have been developed. Although these methods are highly effective, many share similar strengths and weaknesses, which are often difficult to discern in the absence of a clear taxonomy. Previous studies have attempted to categorize node clustering algorithms; however, most are not restricted to GNN-based methods and instead encompass a wide variety of graph learning techniques. Furthermore, despite significant progress in community detection, valuable insights from this area remain underexplored in the context of GNN-based clustering. To bridge this gap and provide a structured overview of the field, we conducted a Systematic Literature Review (SLR) on node clustering in homogeneous networks using GNNs. A total of 84 articles were reviewed based on 20 inclusion and exclusion criteria. From this analysis, we propose a comprehensive taxonomy that integrates both deep clustering and node embedding models. The findings reveal key research gaps, including the absence of transformer-based models, the high computational cost of some approaches, and dataset bias arising from the over-reliance on widely used benchmarks. The proposed taxonomy offers a clearer framework for understanding existing methodologies and guiding future research directions in GNN-based node clustering.
- 21 October 2025
- Published by Springer Nature in Machine Learning
- Vol. 114 (11) , 1-31
Pdarts: projected differentiable architecture search for seismic inversion
- 21 October 2025
- Published by Springer Nature in Machine Learning
- Vol. 114 (11) , 1-31
Pdarts: projected differentiable architecture search for seismic inversion
- 9 October 2025
- Published by Elsevier in Applied Soft Computing
- Vol. 186
Efficient multiple density-based models over large datasets with data stream applications
- 9 October 2025
- Published by Elsevier in Applied Soft Computing
- Vol. 186
Efficient multiple density-based models over large datasets with data stream applications
- Open Access
Unsupervised Parameter-free Outlier Detection using HDBSCAN* Outlier Profiles
In machine learning and data mining, outliers are data points that significantly differ from the dataset and often introduce irrelevant information that can induce bias in its statistics and models. Therefore, unsupervised methods are crucial to detect outliers if there is limited or no information about them. Global-Local Outlier Scores based on Hierarchies (GLOSH) is an unsupervised outlier detection method within HDBSCAN*, a state-of-the-art hierarchical clustering method. GLOSH estimates outlier scores for each data point by comparing its density to the highest density of the region they reside in the HDBSCAN* hierarchy. GLOSH may be sensitive to HDBSCAN*'s minpts parameter that influences density estimation. With limited knowledge about the data, choosing an appropriate minpts value beforehand is challenging as one or some minpts values may better represent the underlying cluster structure than others. Additionally, in the process of searching for ``potential outliers'', one has to define the number of outliers n a dataset has, which may be impractical and is often unknown. In this paper, we propose an unsupervised strategy to find the ``best'' minpts value, leveraging the range of GLOSH scores across minpts values to identify the value for which GLOSH scores can best identify outliers from the rest of the dataset. Moreover, we propose an unsupervised strategy to estimate a threshold for classifying points into inliers and (potential) outliers without the need to pre-define any value. Our experiments show that our strategies can automatically find the minpts value and threshold that yield the best or near best outlier detection results using GLOSH.
Open AccessUnsupervised Parameter-free Outlier Detection using HDBSCAN* Outlier Profiles
In machine learning and data mining, outliers are data points that significantly differ from the dataset and often introduce irrelevant information that can induce bias in its statistics and models. Therefore, unsupervised methods are crucial to detect outliers if there is limited or no information about them. Global-Local Outlier Scores based on Hierarchies (GLOSH) is an unsupervised outlier detection method within HDBSCAN*, a state-of-the-art hierarchical clustering method. GLOSH estimates outlier scores for each data point by comparing its density to the highest density of the region they reside in the HDBSCAN* hierarchy. GLOSH may be sensitive to HDBSCAN*'s minpts parameter that influences density estimation. With limited knowledge about the data, choosing an appropriate minpts value beforehand is challenging as one or some minpts values may better represent the underlying cluster structure than others. Additionally, in the process of searching for ``potential outliers'', one has to define the number of outliers n a dataset has, which may be impractical and is often unknown. In this paper, we propose an unsupervised strategy to find the ``best'' minpts value, leveraging the range of GLOSH scores across minpts values to identify the value for which GLOSH scores can best identify outliers from the rest of the dataset. Moreover, we propose an unsupervised strategy to estimate a threshold for classifying points into inliers and (potential) outliers without the need to pre-define any value. Our experiments show that our strategies can automatically find the minpts value and threshold that yield the best or near best outlier detection results using GLOSH.
- Open Access
- 17 October 2023
- Published by Sociedade Brasileira de Computação
- p. 97-108
Comparison of scalable distributed algorithms for assessing the kNNG in multi-GPU
Many applications, require finding a dataset’s k-Nearest Neighbors Graph (kNNG), crucial for many Machine Learning tasks like clustering and anomaly detection. However, its computation can be costly due to the complexity of finding all kNN for every data point. To address this issue, scalable approximated algorithms have been proposed to speed up the kNNG and maintain its quality. This paper presents an adaption of NNDescent using multi-GPU and an experimental comparison of distributed and parallel approximate kNNG algorithms in GPUs, assessing their scalability, computational cost, and solution quality. Our goal is to identify the most efficient method without significant accuracy loss, enabling faster techniques and handling large datasets.
- 17 October 2023
- Published by Sociedade Brasileira de Computação
- p. 97-108
Open AccessComparison of scalable distributed algorithms for assessing the kNNG in multi-GPU
Many applications, require finding a dataset’s k-Nearest Neighbors Graph (kNNG), crucial for many Machine Learning tasks like clustering and anomaly detection. However, its computation can be costly due to the complexity of finding all kNN for every data point. To address this issue, scalable approximated algorithms have been proposed to speed up the kNNG and maintain its quality. This paper presents an adaption of NNDescent using multi-GPU and an experimental comparison of distributed and parallel approximate kNNG algorithms in GPUs, assessing their scalability, computational cost, and solution quality. Our goal is to identify the most efficient method without significant accuracy loss, enabling faster techniques and handling large datasets.
- 12 October 2023
- Published by Springer Nature
- p. 48-62
Efficient Density-Based Models for Multiple Machine Learning Solutions over Large Datasets
- 12 October 2023
- Published by Springer Nature
- p. 48-62
Efficient Density-Based Models for Multiple Machine Learning Solutions over Large Datasets
- Open Access
- 25 June 2022
- Published by MDPI AG in Applied Sciences
- Vol. 12 (13) , 6464
Optimization Algorithms for Scalable Stream Batch Clustering with k Estimation
The increasing volume and velocity of the continuously generated data (data stream) challenge machine learning algorithms, which must evolve to fit real-world problems. The data stream clustering algorithms face issues such as the rapidly increasing volume of the data, the variety of the number of clusters, and their shapes. The present work aims to improve the accuracy of sequential clustering batches of data streams for scenarios in which clusters evolve dynamically and continuously, automatically estimating their number. In order to achieve this goal, three evolutionary algorithms are presented, along with three novel algorithms designed to deal with clusters of normal distribution based on goodness-of-fit tests in the context of scalable batch stream clustering with automatic estimation of the number of clusters. All of them are developed on top of MapReduce, Discretized-Stream models, and the most recent MPC frameworks to provide scalability, reliability, resilience, and flexibility. The proposed algorithms are experimentally compared with state-of-the-art methods and present the best results for accuracy for normally distributed data sets, reaching their goal.
- 25 June 2022
- Published by MDPI AG in Applied Sciences
- Vol. 12 (13) , 6464
Open AccessOptimization Algorithms for Scalable Stream Batch Clustering with k Estimation
The increasing volume and velocity of the continuously generated data (data stream) challenge machine learning algorithms, which must evolve to fit real-world problems. The data stream clustering algorithms face issues such as the rapidly increasing volume of the data, the variety of the number of clusters, and their shapes. The present work aims to improve the accuracy of sequential clustering batches of data streams for scenarios in which clusters evolve dynamically and continuously, automatically estimating their number. In order to achieve this goal, three evolutionary algorithms are presented, along with three novel algorithms designed to deal with clusters of normal distribution based on goodness-of-fit tests in the context of scalable batch stream clustering with automatic estimation of the number of clusters. All of them are developed on top of MapReduce, Discretized-Stream models, and the most recent MPC frameworks to provide scalability, reliability, resilience, and flexibility. The proposed algorithms are experimentally compared with state-of-the-art methods and present the best results for accuracy for normally distributed data sets, reaching their goal.