Selected work

Representative Papers

RIS-Kernel: A Model-Agnostic Architecture for Long-Context LLM Inference via Sparse Attention

Jul 23, 2026

This work addresses the challenge of scaling large language models to ultra-long contexts, where the quadratic complexity of full self-attention becomes prohibitive. The authors propose RIS-Kernel, a general-purpose architecture that reduces self-attention complexity from O(N²) to O(N log N) through sparse random geometric sampling, without modifying model weights or requiring GPU acceleration. Notably, this approach pioneers the use of sparse attention as a regularization mechanism, achieving higher inference accuracy even at extremely low densities (e.g., 1%) while circumventing memory bottlenecks. Experiments demonstrate that RIS-Stochastic attains 75.00% accuracy on 32K-context tasks, outperforming dense baselines, and improves over zero-context baselines by 14.06 percentage points on 65K-context tasks—all efficiently executable on standard CPU servers with only 16–128 GB of memory.

0 citationsRead paper

Agentic Knowledge Tracing: A Multi-Agent LLM Architecture for Stealth Assessment of Financial Literacy in Serious Games

Jun 23, 2026

This study proposes a novel, unobtrusive method for assessing players’ financial literacy without disrupting their learning experience in serious games. By integrating structured behavioral logs, domain-specific agents, Bayesian Knowledge Tracing (BKT), and an expert adjudication mechanism into a multi-agent large language model architecture—termed Agentic BKT—the approach uniquely combines domain decomposition with conversational-level reasoning to implicitly and accurately model four key dimensions: risk, investment, consumption, and credit. Evaluated across 193 K–12 students and 264 gameplay sessions, the estimated mastery scores demonstrated significant correlations with learning gains (r = 0.276) and post-test performance (r = 0.333). The method achieves predictive validity three times higher than a single-LLM baseline, substantially enhancing the validity and practicality of implicit assessment in educational gaming contexts.

0 citationsRead paper

From Awareness to Action: How Developers Engage with Accessibility Innovation in LLM-Assisted Development

Jun 08, 2026

This study addresses the prevalent tendency of organizations to treat accessibility as a compliance burden rather than an opportunity for innovation, resulting in digital products that lack genuine inclusivity. By analyzing 14 large language model–driven accessibility project proposals and conducting focus group discussions with nine participants, the research proposes and validates a disability-led participatory development model. Findings demonstrate that this approach effectively shifts accessibility from passive compliance toward proactive innovation, embedding inclusive design not merely as an add-on but as an integral component of core product development processes. Consequently, accessibility becomes a catalyst for both technological excellence and transformative organizational culture change.

0 citationsRead paper

PRISM: Perinuclear Ring-based Image Segmentation Method for Acute Lymphoblastic Leukemia Classification

May 12, 2026

This study addresses the challenges in classifying acute lymphoblastic leukemia (ALL) cells from blood smears, where low cytoplasmic contrast and high morphological variability render conventional membrane-based segmentation methods ineffective and limit the generalization of existing deep learning models. To overcome these limitations, the authors propose PRISM, a novel approach that eschews precise cytoplasmic segmentation and instead constructs adaptive concentric rings around the nucleus. This framework integrates color features with gray-level co-occurrence matrix–derived texture information and employs a calibrated stacked ensemble classifier for robust discrimination. By eliminating reliance on cell boundary delineation, PRISM demonstrates strong robustness and generalizability across diverse staining protocols and imaging conditions, achieving 98.46% accuracy and a PR-AUC of 0.9937 in ALL classification.

0 citationsRead paper

Evaluating the Reliability of Multiple Large Language Models in Risk Assessment: A CIS Controls Based Approach

May 06, 2026

This study addresses the reliability concerns of large language models (LLMs) in cybersecurity risk assessment, where their tendency to generate inaccurate or hallucinated outputs may mislead critical decision-making. Leveraging the CIS Controls framework, the authors develop the first standardized questionnaire comprising realistic risk scenarios and systematically compare the assessments of five leading LLMs against those of 50 human experts. The findings reveal that LLMs consistently underestimate security risks and are generally incapable of performing reliable evaluations independently. The work underscores the necessity of human–AI collaboration, advocating for the use of LLMs strictly as assistive tools under expert supervision to enhance the accuracy and trustworthiness of cybersecurity risk assessments.

0 citationsRead paper

Recent publications

Latest Papers

RIS-Kernel: A Model-Agnostic Architecture for Long-Context LLM Inference via Sparse Attention

Jul 23, 2026

This work addresses the challenge of scaling large language models to ultra-long contexts, where the quadratic complexity of full self-attention becomes prohibitive. The authors propose RIS-Kernel, a general-purpose architecture that reduces self-attention complexity from O(N²) to O(N log N) through sparse random geometric sampling, without modifying model weights or requiring GPU acceleration. Notably, this approach pioneers the use of sparse attention as a regularization mechanism, achieving higher inference accuracy even at extremely low densities (e.g., 1%) while circumventing memory bottlenecks. Experiments demonstrate that RIS-Stochastic attains 75.00% accuracy on 32K-context tasks, outperforming dense baselines, and improves over zero-context baselines by 14.06 percentage points on 65K-context tasks—all efficiently executable on standard CPU servers with only 16–128 GB of memory.

0 citationsRead paper

Agentic Knowledge Tracing: A Multi-Agent LLM Architecture for Stealth Assessment of Financial Literacy in Serious Games

Jun 23, 2026

This study proposes a novel, unobtrusive method for assessing players’ financial literacy without disrupting their learning experience in serious games. By integrating structured behavioral logs, domain-specific agents, Bayesian Knowledge Tracing (BKT), and an expert adjudication mechanism into a multi-agent large language model architecture—termed Agentic BKT—the approach uniquely combines domain decomposition with conversational-level reasoning to implicitly and accurately model four key dimensions: risk, investment, consumption, and credit. Evaluated across 193 K–12 students and 264 gameplay sessions, the estimated mastery scores demonstrated significant correlations with learning gains (r = 0.276) and post-test performance (r = 0.333). The method achieves predictive validity three times higher than a single-LLM baseline, substantially enhancing the validity and practicality of implicit assessment in educational gaming contexts.

0 citationsRead paper

From Awareness to Action: How Developers Engage with Accessibility Innovation in LLM-Assisted Development

Jun 08, 2026

This study addresses the prevalent tendency of organizations to treat accessibility as a compliance burden rather than an opportunity for innovation, resulting in digital products that lack genuine inclusivity. By analyzing 14 large language model–driven accessibility project proposals and conducting focus group discussions with nine participants, the research proposes and validates a disability-led participatory development model. Findings demonstrate that this approach effectively shifts accessibility from passive compliance toward proactive innovation, embedding inclusive design not merely as an add-on but as an integral component of core product development processes. Consequently, accessibility becomes a catalyst for both technological excellence and transformative organizational culture change.

0 citationsRead paper

PRISM: Perinuclear Ring-based Image Segmentation Method for Acute Lymphoblastic Leukemia Classification

May 12, 2026

This study addresses the challenges in classifying acute lymphoblastic leukemia (ALL) cells from blood smears, where low cytoplasmic contrast and high morphological variability render conventional membrane-based segmentation methods ineffective and limit the generalization of existing deep learning models. To overcome these limitations, the authors propose PRISM, a novel approach that eschews precise cytoplasmic segmentation and instead constructs adaptive concentric rings around the nucleus. This framework integrates color features with gray-level co-occurrence matrix–derived texture information and employs a calibrated stacked ensemble classifier for robust discrimination. By eliminating reliance on cell boundary delineation, PRISM demonstrates strong robustness and generalizability across diverse staining protocols and imaging conditions, achieving 98.46% accuracy and a PR-AUC of 0.9937 in ALL classification.

0 citationsRead paper

Evaluating the Reliability of Multiple Large Language Models in Risk Assessment: A CIS Controls Based Approach

May 06, 2026

This study addresses the reliability concerns of large language models (LLMs) in cybersecurity risk assessment, where their tendency to generate inaccurate or hallucinated outputs may mislead critical decision-making. Leveraging the CIS Controls framework, the authors develop the first standardized questionnaire comprising realistic risk scenarios and systematically compare the assessments of five leading LLMs against those of 50 human experts. The findings reveal that LLMs consistently underestimate security risks and are generally incapable of performing reliable evaluations independently. The work underscores the necessity of human–AI collaboration, advocating for the use of LLMs strictly as assistive tools under expert supervision to enhance the accuracy and trustworthiness of cybersecurity risk assessments.

0 citationsRead paper