Selected work

Representative Papers

EgoGazeLite: On-Device Egocentric Gaze Prediction for Token-Efficient Multimodal LLM Video Input

Aug 16, 2026

This study addresses the low token efficiency in first-person video understanding caused by the absence of eye-tracking hardware in consumer-grade smart glasses. To overcome this limitation, we propose EgoGazeLite, a lightweight dual-stream gaze predictor that substitutes dedicated sensors with on-device real-time gaze estimation and visual token pruning to enable efficient multimodal large model inference. Experimental results demonstrate that the proposed model, comprising only 15.7M parameters, achieves real-time inference at 21.6 ms per frame. Notably, its hardware-free gaze-guided cropping yields performance statistically indistinguishable from ground-truth annotations. Consequently, EgoGazeLite effectively satisfies stringent edge deployment constraints while significantly enhancing video understanding efficacy, offering a viable software-centric solution for resource-constrained wearable computing.

0 citationsRead paper

MimirRAG: A Multi-Agent RAG Framework for Financial Data Retrieval with Metadata Integration

May 24, 2026

Traditional RAG systems struggle to extract reliable, verifiable insights from financial documents containing mixed formats and integrate them into analysts’ workflows. This work proposes a multi-agent RAG framework that combines structure-preserving PDF parsing, table-aware chunking, metadata enrichment, agent-driven query planning, and hybrid retrieval, augmented with a context generation mechanism capable of numerical reasoning. Evaluated on FinanceBench, the approach achieves an accuracy of 89.3%, substantially outperforming baseline methods. Furthermore, assessments by four professional financial analysts confirm its high accuracy, usability, and practical deployment value in real-world analytical tasks.

0 citationsRead paper

QML-PipeGuard: Drift-Aware Behavioral Fingerprinting for Quantum Machine Learning Pipeline Integrity

May 24, 2026

This work addresses two practical threats in quantum machine learning (QML) deployment—hardware calibration drift and adversarial channel replacement—by proposing the QML-PipeGuard framework. It presents the first unified threat model that jointly characterizes natural drift and malicious substitution, specifically tailored for QML systems. The framework introduces a runtime behavioral fingerprint based on an informationally complete set of Pauli observables. Leveraging statistical verification under limited sampling, a tight frame bound (C = √3), and a tolerance decomposition mechanism, QML-PipeGuard enables efficient end-to-end monitoring of QML pipelines. Evaluated on IBM’s Heron r2 processor, the approach accurately detects adversarial channels while tolerating benign calibration drift within a single batch using only approximately 14,000 measurement samples.

0 citationsRead paper

QCIVET: A Quantum--Classical Pipeline Integrity Framework with Contract-Based Subtype Verification and Hash-Chained Audit Traces

May 13, 2026

Existing integrity verification methods struggle to ensure both semantic and syntactic consistency in the quantum stages of hybrid quantum-classical computing pipelines. This work proposes the first contract-based integrity verification framework, modeling pipelines as sequences of stages annotated with explicit specifications. The approach integrates hash-chain audit trails with observable deviation testing grounded in Liskov–Wing behavioral subtyping to simultaneously verify syntactic correctness and semantic fidelity. A novel multi-Pauli contract mechanism is introduced to defend against Z-only-sneaky coverage attacks, with formal guarantees of soundness, conditional completeness, and compositionality established under the diamond norm. End-to-end validation is demonstrated on the IBM Quantum Heron r2 processor, supporting real-world applications such as VQE-based drug discovery, and the open-source engine achieves per-stage latency below 1 millisecond.

0 citationsRead paper

Nonparametric regression with dependent censoring or competing risks

Mar 24, 2026

This study addresses the non-identifiability of conventional single-index time-to-event models under unknown dependent censoring or competing risks. By introducing an exclusion restriction, the authors establish—within a nonparametric framework—the first identification result for the ratio of marginal covariate effects, thereby circumventing reliance on strong and untestable assumptions. The proposed nonparametric estimation approach is compatible with widely used semiparametric models, including Cox proportional hazards, accelerated failure time, and proportional odds models, and yields a suite of estimators applicable to general settings. Numerical experiments demonstrate that the method provides robust and efficient estimation of relative covariate effects even when the censoring mechanism is misspecified.

0 citationsRead paper

Recent publications

Latest Papers

EgoGazeLite: On-Device Egocentric Gaze Prediction for Token-Efficient Multimodal LLM Video Input

Aug 16, 2026

This study addresses the low token efficiency in first-person video understanding caused by the absence of eye-tracking hardware in consumer-grade smart glasses. To overcome this limitation, we propose EgoGazeLite, a lightweight dual-stream gaze predictor that substitutes dedicated sensors with on-device real-time gaze estimation and visual token pruning to enable efficient multimodal large model inference. Experimental results demonstrate that the proposed model, comprising only 15.7M parameters, achieves real-time inference at 21.6 ms per frame. Notably, its hardware-free gaze-guided cropping yields performance statistically indistinguishable from ground-truth annotations. Consequently, EgoGazeLite effectively satisfies stringent edge deployment constraints while significantly enhancing video understanding efficacy, offering a viable software-centric solution for resource-constrained wearable computing.

0 citationsRead paper

MimirRAG: A Multi-Agent RAG Framework for Financial Data Retrieval with Metadata Integration

May 24, 2026

Traditional RAG systems struggle to extract reliable, verifiable insights from financial documents containing mixed formats and integrate them into analysts’ workflows. This work proposes a multi-agent RAG framework that combines structure-preserving PDF parsing, table-aware chunking, metadata enrichment, agent-driven query planning, and hybrid retrieval, augmented with a context generation mechanism capable of numerical reasoning. Evaluated on FinanceBench, the approach achieves an accuracy of 89.3%, substantially outperforming baseline methods. Furthermore, assessments by four professional financial analysts confirm its high accuracy, usability, and practical deployment value in real-world analytical tasks.

0 citationsRead paper

QML-PipeGuard: Drift-Aware Behavioral Fingerprinting for Quantum Machine Learning Pipeline Integrity

May 24, 2026

This work addresses two practical threats in quantum machine learning (QML) deployment—hardware calibration drift and adversarial channel replacement—by proposing the QML-PipeGuard framework. It presents the first unified threat model that jointly characterizes natural drift and malicious substitution, specifically tailored for QML systems. The framework introduces a runtime behavioral fingerprint based on an informationally complete set of Pauli observables. Leveraging statistical verification under limited sampling, a tight frame bound (C = √3), and a tolerance decomposition mechanism, QML-PipeGuard enables efficient end-to-end monitoring of QML pipelines. Evaluated on IBM’s Heron r2 processor, the approach accurately detects adversarial channels while tolerating benign calibration drift within a single batch using only approximately 14,000 measurement samples.

0 citationsRead paper

QCIVET: A Quantum--Classical Pipeline Integrity Framework with Contract-Based Subtype Verification and Hash-Chained Audit Traces

May 13, 2026

Existing integrity verification methods struggle to ensure both semantic and syntactic consistency in the quantum stages of hybrid quantum-classical computing pipelines. This work proposes the first contract-based integrity verification framework, modeling pipelines as sequences of stages annotated with explicit specifications. The approach integrates hash-chain audit trails with observable deviation testing grounded in Liskov–Wing behavioral subtyping to simultaneously verify syntactic correctness and semantic fidelity. A novel multi-Pauli contract mechanism is introduced to defend against Z-only-sneaky coverage attacks, with formal guarantees of soundness, conditional completeness, and compositionality established under the diamond norm. End-to-end validation is demonstrated on the IBM Quantum Heron r2 processor, supporting real-world applications such as VQE-based drug discovery, and the open-source engine achieves per-stage latency below 1 millisecond.

0 citationsRead paper

Nonparametric regression with dependent censoring or competing risks

Mar 24, 2026

This study addresses the non-identifiability of conventional single-index time-to-event models under unknown dependent censoring or competing risks. By introducing an exclusion restriction, the authors establish—within a nonparametric framework—the first identification result for the ratio of marginal covariate effects, thereby circumventing reliance on strong and untestable assumptions. The proposed nonparametric estimation approach is compatible with widely used semiparametric models, including Cox proportional hazards, accelerated failure time, and proportional odds models, and yields a suite of estimators applicable to general settings. Numerical experiments demonstrate that the method provides robust and efficient estimation of relative covariate effects even when the censoring mechanism is misspecified.

0 citationsRead paper