Selected work

Representative Papers

The Bureaucracy of Speed: Structural Equivalence Between Memory Consistency Models and Multi-Agent Authorization Revocation

Mar 10, 2026

This work addresses the critical issue of excessive unauthorized operations in traditional identity and access management systems under high-concurrency agent environments, which stems from revocation delays and is fundamentally a consistency problem. The paper establishes, for the first time, a structural equivalence between memory consistency models and authorization revocation mechanisms, and proposes a Capability Consistency System (CCS) grounded in release consistency. By introducing a state mapping φ, CCS preserves permission transfer semantics under bounded staleness, and features an RCC revocation strategy whose efficacy is independent of agent velocity. Experimental results demonstrate that RCC reduces unauthorized operations by 120× compared to conventional TTL-based mechanisms in high-throughput scenarios (50 vs. 6,000) and by 184× during anomalous revocations, with zero security boundary violations observed across 120 simulation runs.

0 citationsRead paper

VISP: Volatility Informed Stochastic Projection for Adaptive Regularization

Sep 01, 2025

Traditional stochastic noise injection methods—such as fixed Gaussian noise or uniform dropout—suffer from limited generalization capability and lack gradient awareness. To address this, we propose a gradient-fluctuation-aware adaptive regularization method. Our approach dynamically estimates the variance and coefficient of variation of neuron-wise gradients, then constructs data-dependent random projection matrices; during training, it selectively injects stronger noise into features exhibiting high gradient fluctuation (i.e., instability), thereby enabling fine-grained, adaptive implicit regularization. To the best of our knowledge, this is the first work to explicitly leverage gradient fluctuation to guide the design of stochastic projection noise. Extensive experiments on MNIST, CIFAR-10, and SVHN demonstrate that our method consistently outperforms standard dropout, Gaussian noise injection, and fixed-strength random projection baselines—yielding significant improvements in test accuracy, model robustness, and internal stability.

0 citationsRead paper

TLoRA: Tri-Matrix Low-Rank Adaptation of Large Language Models

Apr 25, 2025

To address the challenge of balancing parameter efficiency and representational capacity in efficient fine-tuning of large language models (LLMs), this paper proposes Tri-Matrix Low-Rank Adaptation (TMLoRA). TMLoRA decomposes weight updates into two fixed Gaussian random matrices, one trainable low-rank matrix, and layer-wise adaptive learnable scaling factors. This design achieves performance on par with LoRA on the GLUE benchmark while using significantly fewer trainable parameters—setting a new state-of-the-art in parameter efficiency. Theoretical and empirical analyses reveal that TMLoRA induces Gaussian-like weight distributions, ensures norm stability, and enables heterogeneous layer-wise scaling—jointly enhancing expressivity and interpretability. Further analyses of feature-space dynamics, spectral distributions, and update directions confirm its strong alignment with LoRA’s behavior. Collectively, TMLoRA establishes a novel paradigm for efficient LLM adaptation, bridging high fidelity and extreme parameter sparsity.

0 citationsRead paper

Extended Histogram-based Outlier Score (EHBOS)

Feb 08, 2025

Traditional Histogram-Based Outlier Score (HBOS) assumes feature independence, limiting its ability to detect anomalies arising from feature dependencies. Method: We propose Extended HBOS (EHBOS), the first HBOS variant incorporating two-dimensional histograms into the framework to explicitly model pairwise joint feature distributions and capture critical dependency structures—thereby relaxing the univariate independence assumption. EHBOS employs combinatorial pairwise feature scanning, bivariate density estimation, and weighted score fusion to enable context-sensitive anomaly detection. Contribution/Results: Extensive evaluation on 17 benchmark datasets demonstrates that EHBOS significantly outperforms HBOS, achieving an average ROC AUC improvement of over 8%. Gains are especially pronounced on datasets exhibiting strong feature interactions. Moreover, EHBOS retains computational efficiency and exhibits robust performance across diverse data characteristics.

0 citationsRead paper

Recent publications

Latest Papers

The Bureaucracy of Speed: Structural Equivalence Between Memory Consistency Models and Multi-Agent Authorization Revocation

Mar 10, 2026

This work addresses the critical issue of excessive unauthorized operations in traditional identity and access management systems under high-concurrency agent environments, which stems from revocation delays and is fundamentally a consistency problem. The paper establishes, for the first time, a structural equivalence between memory consistency models and authorization revocation mechanisms, and proposes a Capability Consistency System (CCS) grounded in release consistency. By introducing a state mapping φ, CCS preserves permission transfer semantics under bounded staleness, and features an RCC revocation strategy whose efficacy is independent of agent velocity. Experimental results demonstrate that RCC reduces unauthorized operations by 120× compared to conventional TTL-based mechanisms in high-throughput scenarios (50 vs. 6,000) and by 184× during anomalous revocations, with zero security boundary violations observed across 120 simulation runs.

0 citationsRead paper

VISP: Volatility Informed Stochastic Projection for Adaptive Regularization

Sep 01, 2025

Traditional stochastic noise injection methods—such as fixed Gaussian noise or uniform dropout—suffer from limited generalization capability and lack gradient awareness. To address this, we propose a gradient-fluctuation-aware adaptive regularization method. Our approach dynamically estimates the variance and coefficient of variation of neuron-wise gradients, then constructs data-dependent random projection matrices; during training, it selectively injects stronger noise into features exhibiting high gradient fluctuation (i.e., instability), thereby enabling fine-grained, adaptive implicit regularization. To the best of our knowledge, this is the first work to explicitly leverage gradient fluctuation to guide the design of stochastic projection noise. Extensive experiments on MNIST, CIFAR-10, and SVHN demonstrate that our method consistently outperforms standard dropout, Gaussian noise injection, and fixed-strength random projection baselines—yielding significant improvements in test accuracy, model robustness, and internal stability.

0 citationsRead paper

TLoRA: Tri-Matrix Low-Rank Adaptation of Large Language Models

Apr 25, 2025

To address the challenge of balancing parameter efficiency and representational capacity in efficient fine-tuning of large language models (LLMs), this paper proposes Tri-Matrix Low-Rank Adaptation (TMLoRA). TMLoRA decomposes weight updates into two fixed Gaussian random matrices, one trainable low-rank matrix, and layer-wise adaptive learnable scaling factors. This design achieves performance on par with LoRA on the GLUE benchmark while using significantly fewer trainable parameters—setting a new state-of-the-art in parameter efficiency. Theoretical and empirical analyses reveal that TMLoRA induces Gaussian-like weight distributions, ensures norm stability, and enables heterogeneous layer-wise scaling—jointly enhancing expressivity and interpretability. Further analyses of feature-space dynamics, spectral distributions, and update directions confirm its strong alignment with LoRA’s behavior. Collectively, TMLoRA establishes a novel paradigm for efficient LLM adaptation, bridging high fidelity and extreme parameter sparsity.

0 citationsRead paper

Extended Histogram-based Outlier Score (EHBOS)

Feb 08, 2025

Traditional Histogram-Based Outlier Score (HBOS) assumes feature independence, limiting its ability to detect anomalies arising from feature dependencies. Method: We propose Extended HBOS (EHBOS), the first HBOS variant incorporating two-dimensional histograms into the framework to explicitly model pairwise joint feature distributions and capture critical dependency structures—thereby relaxing the univariate independence assumption. EHBOS employs combinatorial pairwise feature scanning, bivariate density estimation, and weighted score fusion to enable context-sensitive anomaly detection. Contribution/Results: Extensive evaluation on 17 benchmark datasets demonstrates that EHBOS significantly outperforms HBOS, achieving an average ROC AUC improvement of over 8%. Gains are especially pronounced on datasets exhibiting strong feature interactions. Moreover, EHBOS retains computational efficiency and exhibits robust performance across diverse data characteristics.

0 citationsRead paper