Naoki Masuyama
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
Research Subjects
Naoki Masuyama has a total of 42 co-authors. Here are the top co-authors by number of shared publications:
- Yusuke Nojima
Department of Core Informatics, Graduate School of Informatics, Osaka Metropolitan University, Sakai, Osaka, Japan
19 shared publications - Hisao Ishibuchi
Southern University of Science and Technology, China
12 shared publications - Chu Kiong Loo
Universiti Malaya
8 shared publications - Naoyuki Kubota
Department of Mechanical Systems Engineering, Graduate School of Systems Design, Tokyo Metropolitan University, 6-6 Asahigaoka, Hino, Tokyo 191-0065, Japan
4 shared publications - Yuichiro Toda
Graduate School of Environmental, Life, Natural Science and Technology, Okayama University, Okayama, Japan
4 shared publications - Zongying Liu
Navigation College, Dalian Maritime University, Dalian 116026, China
4 shared publications
Latest Publications
- 20 November 2026
- Published by Fuji Technology Press Ltd. in Journal of Advanced Computational Intelligence and Intelligent Informatics
- Vol. 30 (6) , 1846-1856
A Continual Self-Supervised Learning Framework Using a Decoder and a Clustering-Based Data Selection Method
Continual learning aims to learn a sequence of tasks without forgetting the previously learned tasks. Promising methods for achieving continual learning include applying a replay buffer that stores previously learned samples in existing learning frameworks. Among these methods, lifelong unsupervised mixup (LUMP) interpolates the current task samples and samples from the replay buffer to prevent forgetting. LUMP randomly selects data for the replay buffer. This random selection is likely to have a negative effect on learning performance. In this study, we propose a clustering-based data selection method for the replay buffer. In addition, we introduce a denoising autoencoder to the self-supervised methods to improve the performance of the learning representations. We compare the clustering algorithms used in our framework and demonstrate the effectiveness of the proposed framework through computational experiments. Experimental results on CIFAR-100 and Tiny-ImageNet demonstrate that the proposed framework improves the average accuracy compared with LUMP while reducing forgetting. In particular, the CA+-based selection approach enhances the learning performance. The source code is available at https://github.com/Masuyama-lab/LUMP_SidAE_Clustering
- 20 November 2026
- Published by Fuji Technology Press Ltd. in Journal of Advanced Computational Intelligence and Intelligent Informatics
- Vol. 30 (6) , 1846-1856
A Continual Self-Supervised Learning Framework Using a Decoder and a Clustering-Based Data Selection Method
Continual learning aims to learn a sequence of tasks without forgetting the previously learned tasks. Promising methods for achieving continual learning include applying a replay buffer that stores previously learned samples in existing learning frameworks. Among these methods, lifelong unsupervised mixup (LUMP) interpolates the current task samples and samples from the replay buffer to prevent forgetting. LUMP randomly selects data for the replay buffer. This random selection is likely to have a negative effect on learning performance. In this study, we propose a clustering-based data selection method for the replay buffer. In addition, we introduce a denoising autoencoder to the self-supervised methods to improve the performance of the learning representations. We compare the clustering algorithms used in our framework and demonstrate the effectiveness of the proposed framework through computational experiments. Experimental results on CIFAR-100 and Tiny-ImageNet demonstrate that the proposed framework improves the average accuracy compared with LUMP while reducing forgetting. In particular, the CA+-based selection approach enhances the learning performance. The source code is available at https://github.com/Masuyama-lab/LUMP_SidAE_Clustering
- 18 July 2026
- Published by Elsevier in Ocean Engineering
- Vol. 364
EAC-based ensemble clustering with four-dimensional features for upbound and downbound route extraction in inland waterways
- 18 July 2026
- Published by Elsevier in Ocean Engineering
- Vol. 364
EAC-based ensemble clustering with four-dimensional features for upbound and downbound route extraction in inland waterways
- 25 March 2026
- Published by Springer Nature in Neural Computing and Applications
- Vol. 38 (7) , 217
A parameter-free adaptive resonance theory-based topological clustering algorithm capable of continual learning
- 25 March 2026
- Published by Springer Nature in Neural Computing and Applications
- Vol. 38 (7) , 217
A parameter-free adaptive resonance theory-based topological clustering algorithm capable of continual learning
- 13 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. PP (9) , 1-15
Class-Imbalanced Learning via Multi-Task Multi-Objective Feature Selection
Class-imbalanced and high-dimensional data pose significant challenges in machine learning, often leading to learning models that underperform on minority classes and struggle with the curse of dimensionality, respectively, and these two issues are inherently interconnected. This paper simultaneously addresses both class imbalance and high dimensionality through a multi-task multi-objective feature selection method. To be specific, the proposed method establishes two related yet distinct feature selection problem formulations for a given task, each focusing on different aspects of overall and minority classification performance. By selecting important features from the original imbalanced data set, the proposed method avoids the need to artificially equalize class distribution during training. Additionally, the cross-task genetic transfer operator enables the reuse of valuable feature subsets evolved from separate searches, thereby improving both minority class accuracy and overall model performance. Experimental results on 17 class-imbalanced data sets show that the proposed method achieves better results in terms of overall classification performance and minority class accuracy than the compared state-of-the-art methods, particularly on the high-dimensional data sets.
- 13 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. PP (9) , 1-15
Class-Imbalanced Learning via Multi-Task Multi-Objective Feature Selection
Class-imbalanced and high-dimensional data pose significant challenges in machine learning, often leading to learning models that underperform on minority classes and struggle with the curse of dimensionality, respectively, and these two issues are inherently interconnected. This paper simultaneously addresses both class imbalance and high dimensionality through a multi-task multi-objective feature selection method. To be specific, the proposed method establishes two related yet distinct feature selection problem formulations for a given task, each focusing on different aspects of overall and minority classification performance. By selecting important features from the original imbalanced data set, the proposed method avoids the need to artificially equalize class distribution during training. Additionally, the cross-task genetic transfer operator enables the reuse of valuable feature subsets evolved from separate searches, thereby improving both minority class accuracy and overall model performance. Experimental results on 17 class-imbalanced data sets show that the proposed method achieves better results in terms of overall classification performance and minority class accuracy than the compared state-of-the-art methods, particularly on the high-dimensional data sets.
- Open Access
- 22 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Fuzzy Systems
- Vol. 34 (2) , 650-664
Fairness via Fuzzy Systems: Analysis of Accuracy–Fairness Tradeoff by Multiobjective Fuzzy Genetics-Based Machine Learning
To address the ethical and social risks of artificial intelligence, there is a growing interest in transparency and fairness, and various approaches that consider fairness in highly transparent artificial intelligence have attracted significant attention. In particular, inherently interpretable models can be powerful tools in scenarios where transparency and fairness are important, as they enable fairness to be addressed on the basis of an understanding of their internal mechanisms. A fuzzy system is a representative inherently interpretable model that can make flexible decisions considering real-world uncertainties. Multi-objective fuzzy genetics-based machine learning generates a number of fuzzy classifiers considering trade-offs among multiple objectives by using an evolutionary multi-objective optimization algorithm. In this study, we analyze in detail a set of fuzzy classifiers obtained by multi-objective fuzzy genetics-based machine learning, which simultaneously optimizes accuracy and fairness. We compare this set with other sets of inherently interpretable models and investigate the internal mechanisms of the fuzzy classifiers. Furthermore, we investigate the effects of combining fairness-aware optimization with fairness through unawareness on the fairness of fuzzy classifiers. The experimental results show that fuzzy systems are useful as inherently interpretable and fair models, and provide several insights that offer valuable guidance for fair artificial intelligence design.
- 22 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Fuzzy Systems
- Vol. 34 (2) , 650-664
Open AccessFairness via Fuzzy Systems: Analysis of Accuracy–Fairness Tradeoff by Multiobjective Fuzzy Genetics-Based Machine Learning
To address the ethical and social risks of artificial intelligence, there is a growing interest in transparency and fairness, and various approaches that consider fairness in highly transparent artificial intelligence have attracted significant attention. In particular, inherently interpretable models can be powerful tools in scenarios where transparency and fairness are important, as they enable fairness to be addressed on the basis of an understanding of their internal mechanisms. A fuzzy system is a representative inherently interpretable model that can make flexible decisions considering real-world uncertainties. Multi-objective fuzzy genetics-based machine learning generates a number of fuzzy classifiers considering trade-offs among multiple objectives by using an evolutionary multi-objective optimization algorithm. In this study, we analyze in detail a set of fuzzy classifiers obtained by multi-objective fuzzy genetics-based machine learning, which simultaneously optimizes accuracy and fairness. We compare this set with other sets of inherently interpretable models and investigate the internal mechanisms of the fuzzy classifiers. Furthermore, we investigate the effects of combining fairness-aware optimization with fairness through unawareness on the fairness of fuzzy classifiers. The experimental results show that fuzzy systems are useful as inherently interpretable and fair models, and provide several insights that offer valuable guidance for fair artificial intelligence design.
- Open Access
- 22 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 215704-215714
Adaptive Topological Mapping With Free Area-Based Node Deletion for Autonomous Mobile Robots
This paper proposes an adaptive topological map building method, called Adaptive Resonance Theory-based Topological Clustering with Different Topologies (ATC-DT), for autonomous mobile robots using 3D point cloud data. ATC-DT framework integrates a novel node deletion mechanism that detects layout changes through free area detection. This allows the robot to update topological maps dynamically, removing outdated nodes caused by environmental changes. Experiments in real environments validate the ability of the method to perform global path planning, free area estimation, and adaptive navigation. The approach significantly improves navigation performance by improving map relevance and reducing redundancy of paths.
- 22 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 215704-215714
Open AccessAdaptive Topological Mapping With Free Area-Based Node Deletion for Autonomous Mobile Robots
This paper proposes an adaptive topological map building method, called Adaptive Resonance Theory-based Topological Clustering with Different Topologies (ATC-DT), for autonomous mobile robots using 3D point cloud data. ATC-DT framework integrates a novel node deletion mechanism that detects layout changes through free area detection. This allows the robot to update topological maps dynamically, removing outdated nodes caused by environmental changes. Experiments in real environments validate the ability of the method to perform global path planning, free area estimation, and adaptive navigation. The approach significantly improves navigation performance by improving map relevance and reducing redundancy of paths.
- Open Access
- 14 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 287-290
LIMES: Learning from Imbalanced Data via Multiform Multi-Objective Feature Selection
- 14 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 287-290
Open AccessLIMES: Learning from Imbalanced Data via Multiform Multi-Objective Feature Selection
- Open Access
- 4 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 91625-91641
Neuro-Mimetic Developmental Architecture for Continual Learning Through Self-Organizing Multimodal Perception Coordination
Developing cognitive capabilities in autonomous agents stands as a challenging yet pivotal task. In order to construct cohesive representations and extract meaningful insights from the environment, the brain utilizes patterns from a variety of sensory inputs, including vision and sound. Furthermore, related sensory inputs can activate each other’s representations, highlighting the brain’s ability to associate and integrate information across different sensory channels – key to cognitive development and adaptive learning. Current learning methods often mirror the developmental process of infants, who enhance their cognition through guidance and exploration. These methods often struggle with issues such as catastrophic forgetting and stability-plasticity trade-offs. This study presents a novel brain-inspired hierarchical autonomous framework, Cognitive Deep Self-Organizing Neural Network (CDSN), designed to enable autonomous agents to acquire object concepts dynamically. The architecture includes dual parallel audio-visual information pathways, incorporating three layers based on a Topological Kernel CIM-based Adaptive Resonance Neural Network (TC-ART). The first layer, referred to as the receptive layer, learns and organizes visual attributes and object names autonomously in an unsupervised manner. Subsequently, the second layer, the concept layer, distills clustered results from the corresponding receptive layer to create succinct symbol representation. In a synchronized manner, visual and auditory concepts are combined concurrently in the third layer, the associative layer, to establish real-time associative connections between the modalities. Furthermore, this layer introduces a top-down response approach, allowing agents to independently retrieve associated modalities and adapt acquired knowledge in a hierarchical manner. Experimental evaluations conducted on object datasets demonstrate the proposed architecture’s efficacy in online learning and the association of object viewpoints and labels. Our approach achieves superior recall accuracy, with rates of 92.25% for visual recall and 92.45% for auditory recall. A real-world simulation using humanoid platform further validate proposed neuromimetic architecture’s capabilities. Notably, the cognitive capability can advance autonomously by drawing on its own experiences, reducing the need for human intervention.
- 4 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 91625-91641
Open AccessNeuro-Mimetic Developmental Architecture for Continual Learning Through Self-Organizing Multimodal Perception Coordination
Developing cognitive capabilities in autonomous agents stands as a challenging yet pivotal task. In order to construct cohesive representations and extract meaningful insights from the environment, the brain utilizes patterns from a variety of sensory inputs, including vision and sound. Furthermore, related sensory inputs can activate each other’s representations, highlighting the brain’s ability to associate and integrate information across different sensory channels – key to cognitive development and adaptive learning. Current learning methods often mirror the developmental process of infants, who enhance their cognition through guidance and exploration. These methods often struggle with issues such as catastrophic forgetting and stability-plasticity trade-offs. This study presents a novel brain-inspired hierarchical autonomous framework, Cognitive Deep Self-Organizing Neural Network (CDSN), designed to enable autonomous agents to acquire object concepts dynamically. The architecture includes dual parallel audio-visual information pathways, incorporating three layers based on a Topological Kernel CIM-based Adaptive Resonance Neural Network (TC-ART). The first layer, referred to as the receptive layer, learns and organizes visual attributes and object names autonomously in an unsupervised manner. Subsequently, the second layer, the concept layer, distills clustered results from the corresponding receptive layer to create succinct symbol representation. In a synchronized manner, visual and auditory concepts are combined concurrently in the third layer, the associative layer, to establish real-time associative connections between the modalities. Furthermore, this layer introduces a top-down response approach, allowing agents to independently retrieve associated modalities and adapt acquired knowledge in a hierarchical manner. Experimental evaluations conducted on object datasets demonstrate the proposed architecture’s efficacy in online learning and the association of object viewpoints and labels. Our approach achieves superior recall accuracy, with rates of 92.25% for visual recall and 92.45% for auditory recall. A real-world simulation using humanoid platform further validate proposed neuromimetic architecture’s capabilities. Notably, the cognitive capability can advance autonomously by drawing on its own experiences, reducing the need for human intervention.
- Open Access
- 24 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 139692-139710
Privacy-Preserving Continual Federated Clustering via Adaptive Resonance Theory
With the increasing importance of data privacy protection, various privacy-preserving machine learning methods have been proposed. In the clustering domain, various algorithms with a federated learning framework (i.e., federated clustering) have been actively studied and showed high clustering performance while preserving data privacy. However, most of the base clusterers (i.e., clustering algorithms) used in existing federated clustering algorithms need to specify the number of clusters in advance. These algorithms, therefore, are unable to deal with data whose distributions are unknown or continually changing. To tackle this problem, this paper proposes a privacy-preserving continual federated clustering algorithm. In the proposed algorithm, an adaptive resonance theory-based clustering algorithm capable of continual learning is used as a base clusterer. Therefore, the proposed algorithm inherits the ability of continual learning. Experimental results with synthetic and real-world datasets showthat the proposed algorithm has superior clustering performance to state-of-the-art federated clustering algorithms while realizing data privacy protection and continual learning ability. The source code is available at https://github.com/Masuyama-lab/FCAC.
- 24 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 139692-139710
Open AccessPrivacy-Preserving Continual Federated Clustering via Adaptive Resonance Theory
With the increasing importance of data privacy protection, various privacy-preserving machine learning methods have been proposed. In the clustering domain, various algorithms with a federated learning framework (i.e., federated clustering) have been actively studied and showed high clustering performance while preserving data privacy. However, most of the base clusterers (i.e., clustering algorithms) used in existing federated clustering algorithms need to specify the number of clusters in advance. These algorithms, therefore, are unable to deal with data whose distributions are unknown or continually changing. To tackle this problem, this paper proposes a privacy-preserving continual federated clustering algorithm. In the proposed algorithm, an adaptive resonance theory-based clustering algorithm capable of continual learning is used as a base clusterer. Therefore, the proposed algorithm inherits the ability of continual learning. Experimental results with synthetic and real-world datasets showthat the proposed algorithm has superior clustering performance to state-of-the-art federated clustering algorithms while realizing data privacy protection and continual learning ability. The source code is available at https://github.com/Masuyama-lab/FCAC.
- Open Access
- 12 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 111371-111385
Adaptive Resonance Theory-Based Global Topological Map Building for an Autonomous Mobile Robot
3D space perception is one of the key technologies for autonomous mobile robots that perform tasks in unknown environments. Among these, building global topological maps for autonomous mobile robots is a challenging task. In this study, we propose a method for learning topological structures from unknown data distributions based on competitive learning, a type of unsupervised learning. For this purpose, adaptive resonance theory-based Topological Clustering (ATC), which can avoid catastrophic forgetting of previously measured point clouds, is applied as a learning method. Furthermore, by extending ATC with Different Topologies (ATC-DT) with multiple topological structures for extracting the traversable information of terrain environments, a path planning method is realized that can reach target points set in an unknown environment. Path planning experiments in unknown environments show that, compared to other methods, ATC-DT can build a global topology map with high accuracy and stability using only measured 3D point cloud and robot position information.
- 12 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 111371-111385
Open AccessAdaptive Resonance Theory-Based Global Topological Map Building for an Autonomous Mobile Robot
3D space perception is one of the key technologies for autonomous mobile robots that perform tasks in unknown environments. Among these, building global topological maps for autonomous mobile robots is a challenging task. In this study, we propose a method for learning topological structures from unknown data distributions based on competitive learning, a type of unsupervised learning. For this purpose, adaptive resonance theory-based Topological Clustering (ATC), which can avoid catastrophic forgetting of previously measured point clouds, is applied as a learning method. Furthermore, by extending ATC with Different Topologies (ATC-DT) with multiple topological structures for extracting the traversable information of terrain environments, a path planning method is realized that can reach target points set in an unknown environment. Path planning experiments in unknown environments show that, compared to other methods, ATC-DT can build a global topology map with high accuracy and stability using only measured 3D point cloud and robot position information.