Selected work

Representative Papers

Analyzing Visual Aircraft Representations with Sparse Autoencoders

Jun 13, 2026

This work addresses the limited interpretability of internal representations in vision models, which hinders understanding of their decision-making mechanisms. For the first time, sparse autoencoders are applied to decompose intermediate features of ConvNeXt trained on the FGVC-Aircraft dataset. Through analyses of activated image patches, activation magnitudes, and class selectivity, the study reveals that multiple sparse features correspond to identifiable aircraft parts or semantic visual patterns. Ablation studies linking features to input space, together with quantitative class selectivity metrics, systematically demonstrate these features’ influence on classification confidence and decision boundaries. The investigation also highlights inherent limitations, including feature ambiguity and coarse spatial localization, offering critical insights into the trade-offs between sparsity, interpretability, and representational fidelity in deep visual models.

0 citationsRead paper

Measure Many Quantum Finite Automata on Infinite Words

May 08, 2026

This work investigates the extension of quantum finite automata to infinite words by introducing the model of measure-many quantum Büchi automata (MMQBA). The proposed model integrates unitary evolution with repeated projective measurements and employs Büchi acceptance conditions together with cumulative acceptance probabilities to define language semantics. As the first formalization of measure-many quantum automata over infinite words, this study characterizes the asymptotic behavior of their recognized languages, establishing that MMQBAs are closed under union but not under intersection or complementation. Furthermore, it shows that the emptiness problem is semi-decidable, while universality, inclusion, equivalence, and membership problems are all undecidable.

0 citationsRead paper

GShield: Mitigating Poisoning Attacks in Federated Learning

Dec 22, 2025

To address data poisoning attacks launched by malicious clients under non-IID data distributions in federated learning, this paper proposes an unsupervised, adaptive robust defense mechanism. The method first clusters client gradients from initial training rounds and models the resulting benign update distribution via a multivariate Gaussian—establishing a dynamic baseline for identifying and filtering malicious or low-quality model updates. It further introduces a selective aggregation strategy that weights and combines only trustworthy updates, thereby enhancing global model robustness. Extensive experiments on image and tabular datasets demonstrate that the approach improves target-class accuracy by 43–65% compared to state-of-the-art defenses, while requiring no labeled data or prior knowledge of attack patterns. The core contributions lie in (i) the novel integration of gradient clustering with probabilistic modeling to characterize benign update distributions, and (ii) an adaptive, threshold-free filtering mechanism that dynamically excludes outliers without supervision.

0 citationsRead paper

SmartSecChain-SDN: A Blockchain-Integrated Intelligent Framework for Secure and Efficient Software-Defined Networks

Oct 31, 2025International Journal of Electronics and Communication Engineering

SDN networks suffer from weak security guarantees and coarse-grained traffic control. To address these challenges, this paper proposes an intelligent security framework integrating machine learning–based intrusion detection, blockchain-enabled immutable logging, and application-aware traffic regulation. Methodologically, it synergistically combines Hyperledger Fabric—a permissioned blockchain—with multiple IDS models (Random Forest, XGBoost, CatBoost, and CNN-BiLSTM) to achieve high-accuracy attack detection with low false positives, while ensuring tamper-proof, auditable, and traceable log provenance. Furthermore, the framework dynamically optimizes QoS provisioning and bandwidth allocation based on application-layer features. A prototype system is implemented atop Mininet with OpenDaylight/Ryu controllers and evaluated on the InSDN dataset. Results demonstrate real-time detection of diverse attacks—including DDoS and port scanning—and sustained QoS assurance for critical services under resource-constrained conditions.

0 citationsRead paper

The Loupe: A Plug-and-Play Attention Module for Amplifying Discriminative Features in Vision Transformers

Aug 20, 2025

Fine-grained visual categorization (FGVC) confronts dual challenges: difficulty in identifying discriminative local features and poor model interpretability. To address these, we propose Loupe—a lightweight, plug-and-play attention module that implicitly localizes discriminative regions without requiring part-level annotations, while simultaneously generating human-interpretable attention maps. Loupe integrates seamlessly with mainstream vision transformers (ViTs), including Swin Transformer, and is trained end-to-end with a composite loss function designed to jointly optimize classification accuracy and attention fidelity. On the CUB-200-2011 benchmark, integrating Loupe into Swin-Base boosts top-1 accuracy from 85.40% to 88.06% (+2.66%), outperforming strong baselines. Qualitative analysis confirms that Loupe’s attention maps precisely highlight semantically critical regions—e.g., bird beaks, wing patterns—thereby enhancing both predictive performance and decision transparency. This makes Loupe particularly suitable for high-stakes applications demanding reliability and interpretability, such as biodiversity monitoring and medical image diagnosis.

0 citationsRead paper

Recent publications

Latest Papers

Analyzing Visual Aircraft Representations with Sparse Autoencoders

Jun 13, 2026

This work addresses the limited interpretability of internal representations in vision models, which hinders understanding of their decision-making mechanisms. For the first time, sparse autoencoders are applied to decompose intermediate features of ConvNeXt trained on the FGVC-Aircraft dataset. Through analyses of activated image patches, activation magnitudes, and class selectivity, the study reveals that multiple sparse features correspond to identifiable aircraft parts or semantic visual patterns. Ablation studies linking features to input space, together with quantitative class selectivity metrics, systematically demonstrate these features’ influence on classification confidence and decision boundaries. The investigation also highlights inherent limitations, including feature ambiguity and coarse spatial localization, offering critical insights into the trade-offs between sparsity, interpretability, and representational fidelity in deep visual models.

0 citationsRead paper

Measure Many Quantum Finite Automata on Infinite Words

May 08, 2026

This work investigates the extension of quantum finite automata to infinite words by introducing the model of measure-many quantum Büchi automata (MMQBA). The proposed model integrates unitary evolution with repeated projective measurements and employs Büchi acceptance conditions together with cumulative acceptance probabilities to define language semantics. As the first formalization of measure-many quantum automata over infinite words, this study characterizes the asymptotic behavior of their recognized languages, establishing that MMQBAs are closed under union but not under intersection or complementation. Furthermore, it shows that the emptiness problem is semi-decidable, while universality, inclusion, equivalence, and membership problems are all undecidable.

0 citationsRead paper

GShield: Mitigating Poisoning Attacks in Federated Learning

Dec 22, 2025

To address data poisoning attacks launched by malicious clients under non-IID data distributions in federated learning, this paper proposes an unsupervised, adaptive robust defense mechanism. The method first clusters client gradients from initial training rounds and models the resulting benign update distribution via a multivariate Gaussian—establishing a dynamic baseline for identifying and filtering malicious or low-quality model updates. It further introduces a selective aggregation strategy that weights and combines only trustworthy updates, thereby enhancing global model robustness. Extensive experiments on image and tabular datasets demonstrate that the approach improves target-class accuracy by 43–65% compared to state-of-the-art defenses, while requiring no labeled data or prior knowledge of attack patterns. The core contributions lie in (i) the novel integration of gradient clustering with probabilistic modeling to characterize benign update distributions, and (ii) an adaptive, threshold-free filtering mechanism that dynamically excludes outliers without supervision.

0 citationsRead paper

SmartSecChain-SDN: A Blockchain-Integrated Intelligent Framework for Secure and Efficient Software-Defined Networks

Oct 31, 2025International Journal of Electronics and Communication Engineering

SDN networks suffer from weak security guarantees and coarse-grained traffic control. To address these challenges, this paper proposes an intelligent security framework integrating machine learning–based intrusion detection, blockchain-enabled immutable logging, and application-aware traffic regulation. Methodologically, it synergistically combines Hyperledger Fabric—a permissioned blockchain—with multiple IDS models (Random Forest, XGBoost, CatBoost, and CNN-BiLSTM) to achieve high-accuracy attack detection with low false positives, while ensuring tamper-proof, auditable, and traceable log provenance. Furthermore, the framework dynamically optimizes QoS provisioning and bandwidth allocation based on application-layer features. A prototype system is implemented atop Mininet with OpenDaylight/Ryu controllers and evaluated on the InSDN dataset. Results demonstrate real-time detection of diverse attacks—including DDoS and port scanning—and sustained QoS assurance for critical services under resource-constrained conditions.

0 citationsRead paper

The Loupe: A Plug-and-Play Attention Module for Amplifying Discriminative Features in Vision Transformers

Aug 20, 2025

Fine-grained visual categorization (FGVC) confronts dual challenges: difficulty in identifying discriminative local features and poor model interpretability. To address these, we propose Loupe—a lightweight, plug-and-play attention module that implicitly localizes discriminative regions without requiring part-level annotations, while simultaneously generating human-interpretable attention maps. Loupe integrates seamlessly with mainstream vision transformers (ViTs), including Swin Transformer, and is trained end-to-end with a composite loss function designed to jointly optimize classification accuracy and attention fidelity. On the CUB-200-2011 benchmark, integrating Loupe into Swin-Base boosts top-1 accuracy from 85.40% to 88.06% (+2.66%), outperforming strong baselines. Qualitative analysis confirms that Loupe’s attention maps precisely highlight semantically critical regions—e.g., bird beaks, wing patterns—thereby enhancing both predictive performance and decision transparency. This makes Loupe particularly suitable for high-stakes applications demanding reliability and interpretability, such as biodiversity monitoring and medical image diagnosis.

0 citationsRead paper