Selected work

Representative Papers

TAAF: A Trace Abstraction and Analysis Framework Synergizing Knowledge Graphs and LLMs

Jan 06, 2026arXiv.org

This work addresses the challenge of analyzing massive execution traces generated by large-scale systems such as operating system kernels, Chrome, and MySQL, which are difficult to interpret using existing tools that rely on predefined methods or error-prone, labor-intensive domain-specific scripts. The paper proposes TAAF, a novel framework that integrates temporal-indexed knowledge graphs with large language models (LLMs) to enable multi-hop and causal reasoning through a natural language question-answering interface, substantially reducing reliance on manual expertise. Evaluated on the authors’ newly introduced TraceQA-100 benchmark, TAAF achieves up to a 31.2% improvement in answer accuracy over baseline methods, demonstrating particularly strong performance on complex reasoning tasks.

1 citationsRead paper

Arrow Operations in Categories of Lattice-valued Relations

Aug 17, 2026

This study addresses the limitation of fixed truth-value lattices in lattice-valued relations, which hinders adaptation to heterogeneous semantics. Extending the arrow allegory framework through the integration of category theory and Heyting algebras, this work proposes a generalized allegory model that permits relation pairs to employ distinct truth-value lattices, alongside a hierarchical sub-allegory system. Specifically, three novel allegory structures are defined with rigorous categorical specifications. These contributions effectively overcome the expressive constraints imposed by single truth-value lattices and significantly enhance the abstract modeling capabilities of lattice-valued relations. Ultimately, this research establishes a robust theoretical foundation for reasoning within complex heterogeneous relational systems.

0 citationsRead paper

Calibrating WEAT Against Anisotropy: ZCA Whitening as a Geometric Pre-Processing Step for Embedding Association Tests

Aug 07, 2026

This work addresses the pervasive anisotropy in the embedding spaces of mainstream language models, which violates the isotropy assumption underlying the Word Embedding Association Test (WEAT) and thereby distorts bias measurements. To mitigate this issue, the study introduces Zero-phase Component Analysis (ZCA) whitening as a geometric preprocessing step within the WEAT pipeline. ZCA minimally perturbs the original embeddings while transforming their covariance matrix into the identity matrix, thereby restoring isotropy. Experiments across ten standard WEAT test sets and seven language models demonstrate that ZCA substantially reduces anisotropy, altering the statistical significance of over 30% of WEAT results and, in some models, even improving semantic similarity performance. These findings confirm that ZCA enhances the consistency and reliability of bias assessments in word embeddings.

0 citationsRead paper

Multi-Objective Reference-Aligned Machine Unlearning

May 29, 2026

This work addresses the challenges of catastrophic forgetting and degraded model utility in machine unlearning caused by single-objective optimization. To this end, the authors propose RAUL, a novel framework that formulates unlearning as a multi-objective optimization problem for the first time. RAUL simultaneously achieves precise sample removal by aligning with a reference distribution under a bounded KL divergence constraint and preserves pre-trained knowledge. By leveraging the reference distribution to constrain the unlearning objective and integrating multi-objective optimization with Jacobian descent to harmonize conflicting gradients, RAUL effectively mitigates gradient interference. Experimental results demonstrate that RAUL achieves the best trade-off between unlearning efficacy and model utility, yielding performance closest to full retraining.

0 citationsRead paper

Evolutionary Refinement of Generative Graph Topologies: A Hybrid WGAN-GA Approach

May 27, 2026

This work addresses the challenge of generating realistic graph data while preserving critical structural properties such as degree distribution and spectral characteristics, which are often distorted in existing methods. The authors propose a novel hybrid approach that integrates Wasserstein GAN (WGAN) with a genetic algorithm (GA). Initially, WGAN produces a coarse graph, which is subsequently refined through GA-based evolutionary optimization of edge connections in a gradient-free manner. Notably, this is the first study to incorporate GA into the post-processing stage of GAN-based graph generation, using Maximum Mean Discrepancy (MMD) as the optimization objective. The method achieves a significant reduction in aggregate MMD metrics, yielding synthetic graphs that closely match real-world data in key topological features while maintaining diversity, thereby enhancing both the quality and practical utility of generated graphs.

0 citationsRead paper

Recent publications

Latest Papers

Arrow Operations in Categories of Lattice-valued Relations

Aug 17, 2026

This study addresses the limitation of fixed truth-value lattices in lattice-valued relations, which hinders adaptation to heterogeneous semantics. Extending the arrow allegory framework through the integration of category theory and Heyting algebras, this work proposes a generalized allegory model that permits relation pairs to employ distinct truth-value lattices, alongside a hierarchical sub-allegory system. Specifically, three novel allegory structures are defined with rigorous categorical specifications. These contributions effectively overcome the expressive constraints imposed by single truth-value lattices and significantly enhance the abstract modeling capabilities of lattice-valued relations. Ultimately, this research establishes a robust theoretical foundation for reasoning within complex heterogeneous relational systems.

0 citationsRead paper

Calibrating WEAT Against Anisotropy: ZCA Whitening as a Geometric Pre-Processing Step for Embedding Association Tests

Aug 07, 2026

This work addresses the pervasive anisotropy in the embedding spaces of mainstream language models, which violates the isotropy assumption underlying the Word Embedding Association Test (WEAT) and thereby distorts bias measurements. To mitigate this issue, the study introduces Zero-phase Component Analysis (ZCA) whitening as a geometric preprocessing step within the WEAT pipeline. ZCA minimally perturbs the original embeddings while transforming their covariance matrix into the identity matrix, thereby restoring isotropy. Experiments across ten standard WEAT test sets and seven language models demonstrate that ZCA substantially reduces anisotropy, altering the statistical significance of over 30% of WEAT results and, in some models, even improving semantic similarity performance. These findings confirm that ZCA enhances the consistency and reliability of bias assessments in word embeddings.

0 citationsRead paper

Multi-Objective Reference-Aligned Machine Unlearning

May 29, 2026

This work addresses the challenges of catastrophic forgetting and degraded model utility in machine unlearning caused by single-objective optimization. To this end, the authors propose RAUL, a novel framework that formulates unlearning as a multi-objective optimization problem for the first time. RAUL simultaneously achieves precise sample removal by aligning with a reference distribution under a bounded KL divergence constraint and preserves pre-trained knowledge. By leveraging the reference distribution to constrain the unlearning objective and integrating multi-objective optimization with Jacobian descent to harmonize conflicting gradients, RAUL effectively mitigates gradient interference. Experimental results demonstrate that RAUL achieves the best trade-off between unlearning efficacy and model utility, yielding performance closest to full retraining.

0 citationsRead paper

Evolutionary Refinement of Generative Graph Topologies: A Hybrid WGAN-GA Approach

May 27, 2026

This work addresses the challenge of generating realistic graph data while preserving critical structural properties such as degree distribution and spectral characteristics, which are often distorted in existing methods. The authors propose a novel hybrid approach that integrates Wasserstein GAN (WGAN) with a genetic algorithm (GA). Initially, WGAN produces a coarse graph, which is subsequently refined through GA-based evolutionary optimization of edge connections in a gradient-free manner. Notably, this is the first study to incorporate GA into the post-processing stage of GAN-based graph generation, using Maximum Mean Discrepancy (MMD) as the optimization objective. The method achieves a significant reduction in aggregate MMD metrics, yielding synthetic graphs that closely match real-world data in key topological features while maintaining diversity, thereby enhancing both the quality and practical utility of generated graphs.

0 citationsRead paper

Beyond Morphology: Quantifying the Diagnostic Power of Color Features in Cancer Classification

May 18, 2026

This study investigates whether color features alone—without reliance on morphological information—can effectively support binary cancer diagnosis. By extracting statistical color moments and discretized RGB/HSV histograms, and integrating them with classical machine learning classifiers, the discriminative capacity of global color features is systematically evaluated across ten experimental configurations. The results demonstrate that the raw color distribution of images alone captures non-random diagnostic signals associated with malignancy, achieving a classification accuracy of 89%, which significantly outperforms random baselines. This work provides the first clear validation that color features inherently possess reliable diagnostic potential, offering both theoretical grounding and a practical pathway for developing lightweight models for initial cancer screening.

0 citationsRead paper