Selected work

Representative Papers

Machines that know they are aging: a framework for hardware-aware autonomous intelligence

Jul 30, 2026

This work addresses the critical challenge that autonomous systems often fail in practice due to hardware aging—such as battery degradation and sensor drift—which causes their actual capabilities to deviate from AI assumptions. To bridge this gap, the paper introduces the Aging-Aware Autonomous Intelligence (AAAI) framework, which uniquely integrates physics-of-failure–based hardware health estimation directly into the reasoning, planning, and execution loop. Without requiring additional hardware, AAAI enables self-awareness, adaptive inference, and survival-oriented decision-making. By dynamically adjusting task priorities and resource allocation, the approach supports graceful degradation and mission continuity in unreachable or safety-critical environments—such as deep-space exploration and implantable medical devices—thereby significantly enhancing system resilience and operational lifespan.

0 citationsRead paper

What the Waveform Knows: Transparent-first Speech and Audio Intelligence with Caption Studio

Jul 21, 2026

This work addresses the lack of transparency and poor interpretability in conventional speech and audio analysis systems by proposing a “transparency-first” framework and implementing an end-to-end speech intelligence platform, Caption Studio. The platform employs a three-tier architecture that integrates Whisper-like automatic speech recognition (ASR), pyannote-based speaker diarization, and multidimensional signal-level analyses—including fundamental frequency, speech rate, filler words, and emotion—while explicitly labeling each metric as “measured,” “derived,” or “unavailable” to enhance traceability and credibility. Built on FastAPI for high-availability enterprise deployment, Caption Studio supports real-time processing, seamless workflow integration, and scalable downstream applications, all while incorporating comprehensive interpretability mechanisms and uncertainty management.

0 citationsRead paper

FuzzPilot: Plateau-Triggered Recipe Validation for Structured Text Fuzzing

May 25, 2026

This work addresses the efficiency bottleneck in fuzzing caused by coverage stagnation by proposing a coverage-triggered controller that, upon detecting stagnation, snapshots the corpus and generates lightweight, interpretable JSON-based mutation recipes to avoid expensive inference from blocking the main execution path. The approach decouples mutation strategy from execution, integrating AFL++, Ghidra decompilation data, language model agents, and custom native mutators, with recipe efficacy validated through a micro-campaign mechanism. Evaluated on the cJSON benchmark, the system maintains 1.06× throughput while reducing the median plateau duration from 2,532 to 1,384 seconds (not statistically significant); notably, none of the 20 model-proposed recipes passed validation, highlighting the challenges in automated recipe generation.

0 citationsRead paper

Internet of Things Platform Service Supply Innovation: Exploring the Impact of Overconfidence

Nov 03, 2025

This study investigates how manufacturer overconfidence affects platform-enabled co-innovation in the Internet of Things (IoT) ecosystem. Method: We develop a game-theoretic model grounded in behavioral contract theory, comparing hardware/software innovation investments, pricing strategies, and profit allocation under usage-based versus revenue-sharing contracts. Contribution/Results: Moderate overconfidence incentivizes manufacturers to increase hardware innovation investment, enabling supply-chain Pareto improvements under specific contractual arrangements. The proportion of non-privacy-sensitive customers exerts heterogeneous effects on hardware versus software innovation. Platform software investment exhibits a nonlinear response to revenue-sharing rates. These findings provide a novel behavioral-contract-theoretic framework for understanding and governing manufacturer–platform collaboration in IoT ecosystems, offering actionable governance insights for platform governance and innovation policy.

0 citationsRead paper

DeSamba: Decoupled Spectral Adaptive Framework for 3D Multi-Sequence MRI Lesion Classification

Jul 21, 2025

To address cross-modal feature coupling interference and insufficient spatial-frequency information utilization in 3D multi-sequence MRI lesion classification, this paper proposes a Decoupled Spectral-domain Adaptive Fusion (DSAF) framework. DSAF comprises a decoupled representation learning module and a spectral-domain adaptive modulation block, enabling self-reconstruction and cross-reconstruction decoupling of multi-sequence features while dynamically fusing spatial- and frequency-domain information according to lesion characteristics. Integrating the Mamba architecture, 3D CNNs, and a self-supervised reconstruction strategy, DSAF enhances feature discriminability and robustness. On the six-class spinal metastasis classification task, it achieves 62.10% Top-1 and 93.55% Top-3 accuracy. For spinal spondylitis classification, it attains internal and external validation AUCs of 74.75% and 73.88%, respectively—surpassing all existing state-of-the-art methods.

0 citationsRead paper

Recent publications

Latest Papers

Machines that know they are aging: a framework for hardware-aware autonomous intelligence

Jul 30, 2026

This work addresses the critical challenge that autonomous systems often fail in practice due to hardware aging—such as battery degradation and sensor drift—which causes their actual capabilities to deviate from AI assumptions. To bridge this gap, the paper introduces the Aging-Aware Autonomous Intelligence (AAAI) framework, which uniquely integrates physics-of-failure–based hardware health estimation directly into the reasoning, planning, and execution loop. Without requiring additional hardware, AAAI enables self-awareness, adaptive inference, and survival-oriented decision-making. By dynamically adjusting task priorities and resource allocation, the approach supports graceful degradation and mission continuity in unreachable or safety-critical environments—such as deep-space exploration and implantable medical devices—thereby significantly enhancing system resilience and operational lifespan.

0 citationsRead paper

What the Waveform Knows: Transparent-first Speech and Audio Intelligence with Caption Studio

Jul 21, 2026

This work addresses the lack of transparency and poor interpretability in conventional speech and audio analysis systems by proposing a “transparency-first” framework and implementing an end-to-end speech intelligence platform, Caption Studio. The platform employs a three-tier architecture that integrates Whisper-like automatic speech recognition (ASR), pyannote-based speaker diarization, and multidimensional signal-level analyses—including fundamental frequency, speech rate, filler words, and emotion—while explicitly labeling each metric as “measured,” “derived,” or “unavailable” to enhance traceability and credibility. Built on FastAPI for high-availability enterprise deployment, Caption Studio supports real-time processing, seamless workflow integration, and scalable downstream applications, all while incorporating comprehensive interpretability mechanisms and uncertainty management.

0 citationsRead paper

FuzzPilot: Plateau-Triggered Recipe Validation for Structured Text Fuzzing

May 25, 2026

This work addresses the efficiency bottleneck in fuzzing caused by coverage stagnation by proposing a coverage-triggered controller that, upon detecting stagnation, snapshots the corpus and generates lightweight, interpretable JSON-based mutation recipes to avoid expensive inference from blocking the main execution path. The approach decouples mutation strategy from execution, integrating AFL++, Ghidra decompilation data, language model agents, and custom native mutators, with recipe efficacy validated through a micro-campaign mechanism. Evaluated on the cJSON benchmark, the system maintains 1.06× throughput while reducing the median plateau duration from 2,532 to 1,384 seconds (not statistically significant); notably, none of the 20 model-proposed recipes passed validation, highlighting the challenges in automated recipe generation.

0 citationsRead paper

Internet of Things Platform Service Supply Innovation: Exploring the Impact of Overconfidence

Nov 03, 2025

This study investigates how manufacturer overconfidence affects platform-enabled co-innovation in the Internet of Things (IoT) ecosystem. Method: We develop a game-theoretic model grounded in behavioral contract theory, comparing hardware/software innovation investments, pricing strategies, and profit allocation under usage-based versus revenue-sharing contracts. Contribution/Results: Moderate overconfidence incentivizes manufacturers to increase hardware innovation investment, enabling supply-chain Pareto improvements under specific contractual arrangements. The proportion of non-privacy-sensitive customers exerts heterogeneous effects on hardware versus software innovation. Platform software investment exhibits a nonlinear response to revenue-sharing rates. These findings provide a novel behavioral-contract-theoretic framework for understanding and governing manufacturer–platform collaboration in IoT ecosystems, offering actionable governance insights for platform governance and innovation policy.

0 citationsRead paper

DeSamba: Decoupled Spectral Adaptive Framework for 3D Multi-Sequence MRI Lesion Classification

Jul 21, 2025

To address cross-modal feature coupling interference and insufficient spatial-frequency information utilization in 3D multi-sequence MRI lesion classification, this paper proposes a Decoupled Spectral-domain Adaptive Fusion (DSAF) framework. DSAF comprises a decoupled representation learning module and a spectral-domain adaptive modulation block, enabling self-reconstruction and cross-reconstruction decoupling of multi-sequence features while dynamically fusing spatial- and frequency-domain information according to lesion characteristics. Integrating the Mamba architecture, 3D CNNs, and a self-supervised reconstruction strategy, DSAF enhances feature discriminability and robustness. On the six-class spinal metastasis classification task, it achieves 62.10% Top-1 and 93.55% Top-3 accuracy. For spinal spondylitis classification, it attains internal and external validation AUCs of 74.75% and 73.88%, respectively—surpassing all existing state-of-the-art methods.

0 citationsRead paper