Thiago José Lucas
Research Subjects
Thiago José Lucas has a total of 17 co-authors. Here are the top co-authors by number of shared publications:
- Kelton Augusto Pontara da Costa
UNESP
6 shared publications - João Paulo Papa
São Paulo State University (UNESP)
3 shared publications - Celso de Oliveira Lisboa2 shared publications
- Rafał Scherer
Department of Artificial Intelligence, Czȩstochowa University of Technology, al. Armii Krajowej 36, 42-200 Czȩstochowa, Poland
2 shared publications Ronaldo Moreno
2 shared publicationsAlex Marino G. de Almeida
Department of Computing, So Paulo State University, Bauru, Brazil
1 shared publication
Latest Publications
- Open Access
- 2 November 2026
- Published by Elsevier in Computer Networks
- Vol. 292
2D-AEF: Low-latency sparse expert routing for network intrusion detection
- 2 November 2026
- Published by Elsevier in Computer Networks
- Vol. 292
Open Access2D-AEF: Low-latency sparse expert routing for network intrusion detection
- 31 October 2026
- Published by Springer Nature
- p. 17-31
Multi-dimensional Privacy Auditing of Large Language Models via Membership Inference Attacks
- 31 October 2026
- Published by Springer Nature
- p. 17-31
Multi-dimensional Privacy Auditing of Large Language Models via Membership Inference Attacks
- Open Access
- 12 June 2026
- Published by Elsevier in SSRN Electronic Journal
2D-AEF: Low-Latency Sparse Expert Routing for Network Intrusion Detection
Network intrusion detection systems must operate under strict latency constraints while maintaining reliable detection performance across heterogeneous traffic patterns. Traditional ensemble methods often improve predictive performance but require multiple model evaluations during inference, increasing computational costs. This paper presents 2D-AEF, a two-stage adaptive framework for network intrusion detection based on sparse conditional computation. A shallow decision-tree gatekeeper routes each traffic sample to a class-specialized expert selected through an offline search over model–feature subset pairs. Unlike conventional ensembles, only one expert is activated during inference, reducing computational overhead. The proposed framework was evaluated on UNSW-NB15 binary, UNSW-NB15 multi-class, and leakage-controlled CIC-IDS2018 robust scenarios. Experimental comparisons included strong standalone baselines and ensemble methods based on hard voting, soft voting, and stacking. Across ten random seeds, 2D-AEF achieved macro-F1 scores of 0.890, 0.461, and 0.871 on the three scenarios, respectively. While not consistently outperforming the strongest classifiers in predictive performance, the proposed framework reduced median inference latency by factors ranging from 6.5x to 12.5x compared with stacking. The results demonstrate that class-conditioned sparse routing provides an effective strategy for balancing detection performance and inference efficiency in network intrusion detection systems.
- 12 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open Access2D-AEF: Low-Latency Sparse Expert Routing for Network Intrusion Detection
Network intrusion detection systems must operate under strict latency constraints while maintaining reliable detection performance across heterogeneous traffic patterns. Traditional ensemble methods often improve predictive performance but require multiple model evaluations during inference, increasing computational costs. This paper presents 2D-AEF, a two-stage adaptive framework for network intrusion detection based on sparse conditional computation. A shallow decision-tree gatekeeper routes each traffic sample to a class-specialized expert selected through an offline search over model–feature subset pairs. Unlike conventional ensembles, only one expert is activated during inference, reducing computational overhead. The proposed framework was evaluated on UNSW-NB15 binary, UNSW-NB15 multi-class, and leakage-controlled CIC-IDS2018 robust scenarios. Experimental comparisons included strong standalone baselines and ensemble methods based on hard voting, soft voting, and stacking. Across ten random seeds, 2D-AEF achieved macro-F1 scores of 0.890, 0.461, and 0.871 on the three scenarios, respectively. While not consistently outperforming the strongest classifiers in predictive performance, the proposed framework reduced median inference latency by factors ranging from 6.5x to 12.5x compared with stacking. The results demonstrate that class-conditioned sparse routing provides an effective strategy for balancing detection performance and inference efficiency in network intrusion detection systems.
- 7 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Latin America Transactions
- Vol. 23 (12) , 1249-1260
Detection of Obfuscation Malware: A Federated Transfer Learning-based Approach with Hybrid Neural Networks
The increase in the incidence of cyberattacks, especially through the use of complex mechanisms for exploiting vulnerabilities, such as malware obfuscation, has driven the adoption of Machine Learning (ML) techniques in cybersecurity. This study investigates the application of Federated Learning (FL), a decentralized approach that preserves data privacy and overcomes challenges in transferring large volumes of information. Two labeled datasets were used, CIC-MalMem-2022 and Malware Detection Dataset, along with two FL frameworks, Flower Framework and TensorFlow Federated. A decentralized model based on a Linear Neural Network (LNN) with federated averaging (FedAvg) was compared to a centralized model using a Recurrent Neural Network (RNN) in supervised binary classifications of malware. The results demonstrate high accuracy across all analyzed scenarios, highlighting the outcomes obtained in centralized training for the CIC-Malware dataset, achieving an accuracy of 0.99, precision of 1.0, and recall of 0.99, emphasizing the potential of FL in cybersecurity.
- 7 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Latin America Transactions
- Vol. 23 (12) , 1249-1260
Detection of Obfuscation Malware: A Federated Transfer Learning-based Approach with Hybrid Neural Networks
The increase in the incidence of cyberattacks, especially through the use of complex mechanisms for exploiting vulnerabilities, such as malware obfuscation, has driven the adoption of Machine Learning (ML) techniques in cybersecurity. This study investigates the application of Federated Learning (FL), a decentralized approach that preserves data privacy and overcomes challenges in transferring large volumes of information. Two labeled datasets were used, CIC-MalMem-2022 and Malware Detection Dataset, along with two FL frameworks, Flower Framework and TensorFlow Federated. A decentralized model based on a Linear Neural Network (LNN) with federated averaging (FedAvg) was compared to a centralized model using a Recurrent Neural Network (RNN) in supervised binary classifications of malware. The results demonstrate high accuracy across all analyzed scenarios, highlighting the outcomes obtained in centralized training for the CIC-Malware dataset, achieving an accuracy of 0.99, precision of 1.0, and recall of 0.99, emphasizing the potential of FL in cybersecurity.
- Open Access
- 30 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 122638-122676
A Comprehensive Survey on Ensemble Learning-Based Intrusion Detection Approaches in Computer Networks
Machine learning algorithms present a robust alternative for building Intrusion Detection Systems due to their ability to recognize attacks in computer network traffic by recognizing patterns in large amounts of data. Typically, classifiers are trained for this task. Together, ensemble learning algorithms have increased the performance of these detectors, reducing classification errors and allowing computer networks to be more protected. This research presents a comprehensive Systematic Review of the Literature where works related to intrusion detection with ensemble learning were obtained from the most relevant scientific bases. We offer 188 works, several compilations of datasets, classifiers, and ensemble algorithms, and document the experiments that stood out in their performance. A characteristic of this research is its originality. We found two surveys in the literature specifically focusing on the relationship between ensemble techniques and intrusion detection [1] [2]. We present for the last eight years covered by this survey a timeline-based view of the works studied to highlight evolutions and trends. The results obtained by our survey show a growing area, with excellent results in detecting attacks but with needs for improvement in pruning for choosing classifiers, which makes this work unprecedented for this context.
- 30 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 122638-122676
Open AccessA Comprehensive Survey on Ensemble Learning-Based Intrusion Detection Approaches in Computer Networks
Machine learning algorithms present a robust alternative for building Intrusion Detection Systems due to their ability to recognize attacks in computer network traffic by recognizing patterns in large amounts of data. Typically, classifiers are trained for this task. Together, ensemble learning algorithms have increased the performance of these detectors, reducing classification errors and allowing computer networks to be more protected. This research presents a comprehensive Systematic Review of the Literature where works related to intrusion detection with ensemble learning were obtained from the most relevant scientific bases. We offer 188 works, several compilations of datasets, classifiers, and ensemble algorithms, and document the experiments that stood out in their performance. A characteristic of this research is its originality. We found two surveys in the literature specifically focusing on the relationship between ensemble techniques and intrusion detection [1] [2]. We present for the last eight years covered by this survey a timeline-based view of the works studied to highlight evolutions and trends. The results obtained by our survey show a growing area, with excellent results in detecting attacks but with needs for improvement in pruning for choosing classifiers, which makes this work unprecedented for this context.
- Open Access
Non-Fragmented Network Flow Design Analysis: Comparison IPv4 with IPv6 Using Path MTU Discovery
With the expansion in the number of devices that connect to the Internet, a new area, known as the Internet of Things (IoT), appears. It was necessary to migrate the IPv4 protocol by the IPv6 protocol, due to the scarcity of IPv4 addresses. One of the advances of IPv6 concerning its predecessor was the Path MTU Discovery protocol, in which this work aims to demonstrate its effectiveness in a virtual environment. Employing the VirtualBox virtualization program, a scenario is defined with fifteen machines with the Debian operating system and two network scenarios, one using the IPv4 network configuration and the other using the IPv6 network configuration. In both situations, the MTU values of all machines were changed to perform the performance tests while using UDP transport traffic. The fragmentation of packets demonstrated the effectiveness of the Path MTU Discovery protocol. The results achieved point to a stabilization in bandwidth and jitter when Path MTU Discovery is used and to change when it is no longer applied, proving its effectiveness.
Open AccessNon-Fragmented Network Flow Design Analysis: Comparison IPv4 with IPv6 Using Path MTU Discovery
With the expansion in the number of devices that connect to the Internet, a new area, known as the Internet of Things (IoT), appears. It was necessary to migrate the IPv4 protocol by the IPv6 protocol, due to the scarcity of IPv4 addresses. One of the advances of IPv6 concerning its predecessor was the Path MTU Discovery protocol, in which this work aims to demonstrate its effectiveness in a virtual environment. Employing the VirtualBox virtualization program, a scenario is defined with fifteen machines with the Debian operating system and two network scenarios, one using the IPv4 network configuration and the other using the IPv6 network configuration. In both situations, the MTU values of all machines were changed to perform the performance tests while using UDP transport traffic. The fragmentation of packets demonstrated the effectiveness of the Path MTU Discovery protocol. The results achieved point to a stabilization in bandwidth and jitter when Path MTU Discovery is used and to change when it is no longer applied, proving its effectiveness.