Areolino de Almeida Neto
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Areolino de Almeida Neto has a total of 39 co-authors. Here are the top co-authors by number of shared publications:
- Alexandre César Muniz de Oliveira
Federal University of Maranhão (UFMA), São Luís, Brazil
7 shared publications - Paulo Rogério de Almeida Ribeiro
Federal University of Maranhão (UFMA), São Luís, Brazil
5 shared publications - Francisco dos Santos Viana
Federal Institute of Maranhão
4 shared publications - Matheus Menezes
Federal University of Maranhão (UFMA)
3 shared publications - Bianca Valéria Lopes Pereira
Federal University of Maranhão (UFMA), São Luís, Brazil
2 shared publications - Cairo Lucio Nascimento
Division of Electronic Engineering, Instituto Tecnológico de Aeronáutica, São José dos Campos, Brazil
2 shared publications
Latest Publications
- 1 October 2026
- Published by Springer Nature
- p. 475-490
A Unified Framework for Benchmarking Machine Learning Methods in Genetic Circuit Design
- 1 October 2026
- Published by Springer Nature
- p. 475-490
A Unified Framework for Benchmarking Machine Learning Methods in Genetic Circuit Design
- 2 August 2026
- Published by Springer Nature
- p. 725-732
Topology Optimization in Neural Networks for Deep Reinforcement Learning
- 2 August 2026
- Published by Springer Nature
- p. 725-732
Topology Optimization in Neural Networks for Deep Reinforcement Learning
- 1 July 2026
- Published by Uniao Atlantica de Pesquisadores in Peer Review
- Vol. 8 (7) , 144-163
Training activation functions for layer insertion: a polynomial regression approach
Artificial neural networks (ANNs) are widely applied to complex problems; however, the selection of an ideal activation function (AF) remains a challenge, particularly due to the lack of consistent criteria for relating data characteristics to the most suitable function. Recent advances highlight the potential of trainable AFs that dynamically adjust during training to enhance network performance. Additionally, that adjustment can be used for the insertion of a new layer. This work presents a novel approach that utilizes polynomial regression as the foundation for developing a layer insertion method based on trainable AFs. The proposal aims to go beyond conventional weight adjustment by allowing activation functions to evolve in an optimized manner across different contexts, thereby expanding the learning capabilities and performance of deep neural networks. The proposed approach presents a novel method for layer insertion, which can integrate a new layer in a collaborative way, which means that the new layer does not degrade the knowledge already acquired by the previous layers and besides can improve the effectiveness and the impacts of network adaptability and performance, when compared to fixed AFs as additional hidden layers are introduced.
- 1 July 2026
- Published by Uniao Atlantica de Pesquisadores in Peer Review
- Vol. 8 (7) , 144-163
Training activation functions for layer insertion: a polynomial regression approach
Artificial neural networks (ANNs) are widely applied to complex problems; however, the selection of an ideal activation function (AF) remains a challenge, particularly due to the lack of consistent criteria for relating data characteristics to the most suitable function. Recent advances highlight the potential of trainable AFs that dynamically adjust during training to enhance network performance. Additionally, that adjustment can be used for the insertion of a new layer. This work presents a novel approach that utilizes polynomial regression as the foundation for developing a layer insertion method based on trainable AFs. The proposal aims to go beyond conventional weight adjustment by allowing activation functions to evolve in an optimized manner across different contexts, thereby expanding the learning capabilities and performance of deep neural networks. The proposed approach presents a novel method for layer insertion, which can integrate a new layer in a collaborative way, which means that the new layer does not degrade the knowledge already acquired by the previous layers and besides can improve the effectiveness and the impacts of network adaptability and performance, when compared to fixed AFs as additional hidden layers are introduced.
- 15 September 2025
- Published by Springer Nature
- p. 464-476
A New Proposal of Layer Insertion in Stacked Autoencoder Neural Networks
- 15 September 2025
- Published by Springer Nature
- p. 464-476
A New Proposal of Layer Insertion in Stacked Autoencoder Neural Networks
- Open Access
- 3 September 2025
- Published by Brazilian Journals in REVISTA DELOS
- Vol. 18 (71) , e6481
Classificação de sílabas fonéticas via centroides e inserção colaborativa de camadas em redes stacked autoencoders
Este trabalho apresenta uma nova metodologia de inserção colaborativa de camadas em redes neurais do tipo stacked autoencoder. Nessa abordagem, após o treinamento de uma camada existente, uma nova camada é incorporada por meio de um ramo paralelo às camadas oculta e de saída já treinadas. Esse ramo é posteriormente transformado em uma camada da rede, deixando de ser apenas paralelo. A inserção é considerada colaborativa por introduzir novos conhecimentos sem comprometer os já adquiridos. Com essa estratégia, foi possível construir redes neurais com desempenho satisfatório e redução no tempo de projeto, principalmente na definição do número de camadas e de neurônios. Além disso, aplicou-se uma técnica de extração de características de sinais de sílabas fonéticas. Os dados de áudio foram convertidos em frequências pela transformada rápida de Fourier, representadas por barras verticais agrupadas em figuras irregulares formadas pela união de retângulos. A característica principal foi obtida pelo cálculo do centróide dessas figuras, permitindo compactar os dados sem perda de informação relevante. As inserções colaborativas foram avaliadas em termos de grau de crescimento da rede para um problema de classificação multiclasse, com caracterização baseada nos centróides extraídos.
- 3 September 2025
- Published by Brazilian Journals in REVISTA DELOS
- Vol. 18 (71) , e6481
Open AccessClassificação de sílabas fonéticas via centroides e inserção colaborativa de camadas em redes stacked autoencoders
Este trabalho apresenta uma nova metodologia de inserção colaborativa de camadas em redes neurais do tipo stacked autoencoder. Nessa abordagem, após o treinamento de uma camada existente, uma nova camada é incorporada por meio de um ramo paralelo às camadas oculta e de saída já treinadas. Esse ramo é posteriormente transformado em uma camada da rede, deixando de ser apenas paralelo. A inserção é considerada colaborativa por introduzir novos conhecimentos sem comprometer os já adquiridos. Com essa estratégia, foi possível construir redes neurais com desempenho satisfatório e redução no tempo de projeto, principalmente na definição do número de camadas e de neurônios. Além disso, aplicou-se uma técnica de extração de características de sinais de sílabas fonéticas. Os dados de áudio foram convertidos em frequências pela transformada rápida de Fourier, representadas por barras verticais agrupadas em figuras irregulares formadas pela união de retângulos. A característica principal foi obtida pelo cálculo do centróide dessas figuras, permitindo compactar os dados sem perda de informação relevante. As inserções colaborativas foram avaliadas em termos de grau de crescimento da rede para um problema de classificação multiclasse, com caracterização baseada nos centróides extraídos.
- Open Access
- 3 July 2025
- Published by Brazilian Journals in REVISTA DELOS
- Vol. 18 (69) , e6077
Reconhecimento de sílabas fonéticas utilizando redes Stacked Autoencoders via compressão de frequências com centroides
Este trabalho apresenta um modelo para o reconhecimento de sílabas fonéticas utilizando uma rede Stacked Autoencoder. O objetivo é identificar e distinguir os diferentes sons da fala, considerando sílabas fonéticas e variações na pronúncia. A rede desenvolvida, denominada CollabNet, é caracterizada pela adição colaborativa e gradual de camadas ocultas, garantindo continuidade no aprendizado e convergência progressiva do erro. Essa estratégia também favorece a integração com camadas anteriores e o alinhamento do processo de treinamento. Para representar as sílabas fonéticas, o método realiza compressão de frequências por meio de centroides. Inicialmente, aplica-se a Transformada Rápida de Fourier (FFT) em cada amostra de áudio. As frequências adjacentes são agrupadas e representadas como retângulos em um gráfico de barras, e o centroide de cada grupo é então calculado. Em seguida, a rede é configurada e treinada para reconhecer sílabas fonéticas. A representação do áudio preserva características específicas, permitindo ao CollabNet identificar com precisão diferentes sons do português brasileiro. O modelo alcançou acurácia de 75,96% e Taxa de Erro Fonético (PER) de 23,73%.
- 3 July 2025
- Published by Brazilian Journals in REVISTA DELOS
- Vol. 18 (69) , e6077
Open AccessReconhecimento de sílabas fonéticas utilizando redes Stacked Autoencoders via compressão de frequências com centroides
Este trabalho apresenta um modelo para o reconhecimento de sílabas fonéticas utilizando uma rede Stacked Autoencoder. O objetivo é identificar e distinguir os diferentes sons da fala, considerando sílabas fonéticas e variações na pronúncia. A rede desenvolvida, denominada CollabNet, é caracterizada pela adição colaborativa e gradual de camadas ocultas, garantindo continuidade no aprendizado e convergência progressiva do erro. Essa estratégia também favorece a integração com camadas anteriores e o alinhamento do processo de treinamento. Para representar as sílabas fonéticas, o método realiza compressão de frequências por meio de centroides. Inicialmente, aplica-se a Transformada Rápida de Fourier (FFT) em cada amostra de áudio. As frequências adjacentes são agrupadas e representadas como retângulos em um gráfico de barras, e o centroide de cada grupo é então calculado. Em seguida, a rede é configurada e treinada para reconhecer sílabas fonéticas. A representação do áudio preserva características específicas, permitindo ao CollabNet identificar com precisão diferentes sons do português brasileiro. O modelo alcançou acurácia de 75,96% e Taxa de Erro Fonético (PER) de 23,73%.
- 19 September 2024
- Published by Springer Nature in Neural Computing and Applications
- Vol. 36 (35) , 22013-22022
Firearm detection using DETR with multiple self-coordinated neural networks
- 19 September 2024
- Published by Springer Nature in Neural Computing and Applications
- Vol. 36 (35) , 22013-22022
Firearm detection using DETR with multiple self-coordinated neural networks
- 15 April 2024
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 80 (11) , 16654-16678
Automatic phoneme recognition by deep neural networks
- 15 April 2024
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 80 (11) , 16654-16678
Automatic phoneme recognition by deep neural networks
- Open Access
- 17 July 2023
- Published by Springer Nature in Journal of Intelligent & Robotic Systems
- Vol. 108 (4) , 1-21
A Multisession SLAM Approach for RatSLAM
To successfully perform autonomous navigation, mobile agents must solve the Simultaneous Localization and Mapping (SLAM) problem. However, acquiring the map in a single SLAM session may not be possible, thus the map may be incrementally built over multiple sessions. Two solutions could be considered to solve the multisession SLAM problem: (i) the robot must localize itself in the previously stored map before the new session starts; (ii) it can start a new map and merge it with the map from the previous sessions. To date, only scenario (i) has been addressed by RatSLAM, an algorithm inspired by the navigation system in rodent brains. Therefore, this work proposes a multisession solution that solves both scenarios. A new mechanism merges the data from the RatSLAM structures of the current mapping session with those previously stored if there are connections between these paths. This approach was tested in four different scenarios, from virtual controlled environments to real-world environments with two, three, and five sessions. The robot started in an unfamiliar location for each mapping session, but it also works if the agent starts in a known place, scenario (ii) and (i), respectively. For all experiments, the entire map was consistently obtained. Furthermore, the proposed approach updates and enhances the previous session’s map in real-world environments. Therefore, the proposed approach may be a multiple SLAM session solution for the RatSLAM algorithm.
- 17 July 2023
- Published by Springer Nature in Journal of Intelligent & Robotic Systems
- Vol. 108 (4) , 1-21
Open AccessA Multisession SLAM Approach for RatSLAM
To successfully perform autonomous navigation, mobile agents must solve the Simultaneous Localization and Mapping (SLAM) problem. However, acquiring the map in a single SLAM session may not be possible, thus the map may be incrementally built over multiple sessions. Two solutions could be considered to solve the multisession SLAM problem: (i) the robot must localize itself in the previously stored map before the new session starts; (ii) it can start a new map and merge it with the map from the previous sessions. To date, only scenario (i) has been addressed by RatSLAM, an algorithm inspired by the navigation system in rodent brains. Therefore, this work proposes a multisession solution that solves both scenarios. A new mechanism merges the data from the RatSLAM structures of the current mapping session with those previously stored if there are connections between these paths. This approach was tested in four different scenarios, from virtual controlled environments to real-world environments with two, three, and five sessions. The robot started in an unfamiliar location for each mapping session, but it also works if the agent starts in a known place, scenario (ii) and (i), respectively. For all experiments, the entire map was consistently obtained. Furthermore, the proposed approach updates and enhances the previous session’s map in real-world environments. Therefore, the proposed approach may be a multiple SLAM session solution for the RatSLAM algorithm.
- Open Access
xRatSLAM: An Extensible RatSLAM Computational Framework
Simultaneous localization and mapping (SLAM) refers to techniques for autonomously constructing a map of an unknown environment while, at the same time, locating the robot in this map. RatSLAM, a prevalent method, is based on the navigation system found in rodent brains. It has served as a base algorithm for other bioinspired approaches, and its implementation has been extended to incorporate new features. This work proposes xRatSLAM: an extensible, parallel, open-source framework applicable for developing and testing new RatSLAM variations. Tests were carried out to evaluate and validate the proposed framework, allowing the comparison of xRatSLAM with OpenRatSLAM and assessing the impact of replacing framework components. The results provide evidence that the maps produced by xRatSLAM are similar to those produced by OpenRatSLAM when they are fed with the same input parameters, which is a positive result, and that implemented modules can be easily changed without impacting other parts of the framework.
Open AccessxRatSLAM: An Extensible RatSLAM Computational Framework
Simultaneous localization and mapping (SLAM) refers to techniques for autonomously constructing a map of an unknown environment while, at the same time, locating the robot in this map. RatSLAM, a prevalent method, is based on the navigation system found in rodent brains. It has served as a base algorithm for other bioinspired approaches, and its implementation has been extended to incorporate new features. This work proposes xRatSLAM: an extensible, parallel, open-source framework applicable for developing and testing new RatSLAM variations. Tests were carried out to evaluate and validate the proposed framework, allowing the comparison of xRatSLAM with OpenRatSLAM and assessing the impact of replacing framework components. The results provide evidence that the maps produced by xRatSLAM are similar to those produced by OpenRatSLAM when they are fed with the same input parameters, which is a positive result, and that implemented modules can be easily changed without impacting other parts of the framework.