Selected work

Representative Papers

Bio-Inspired Fine-Tuning for Selective Transfer Learning in Image Classification

Jan 16, 2026IEEE Access

This work addresses the performance limitations of transfer learning in label-scarce image classification tasks caused by domain discrepancies between source and target datasets. To mitigate this issue, the authors propose BioTune, a novel method that, for the first time, integrates biologically inspired evolutionary optimization into the fine-tuning process. BioTune jointly and adaptively selects which layers to freeze while dynamically adjusting the learning rates of unfrozen layers, operating without manual intervention and remaining compatible with various CNN architectures. Extensive experiments across nine natural and medical image datasets demonstrate that BioTune consistently outperforms state-of-the-art approaches such as AutoRGN and LoRA, achieving superior performance across four mainstream CNN backbones and thereby validating its effectiveness and generalization capability.

3 citationsRead paper

On the Effectiveness of ASR Representations in Real-world Noisy Speech Emotion Recognition

Nov 13, 2023arXiv.org

To address the insufficient robustness of speech emotion recognition (SER) under realistic non-stationary noise, this paper proposes an end-to-end approach that neither requires denoising nor text transcription. Specifically, it leverages intermediate-layer representations from a pre-trained automatic speech recognition (ASR) model as a noise-robust acoustic feature extractor and transfers these representations to emotion classification via feature distillation. This work is the first to directly utilize ASR intermediate representations for non-stationary SER, effectively disentangling non-speech components from noisy signals while avoiding distortions introduced by conventional denoising and biases induced by text dependency. Experiments on realistic noisy scenarios demonstrate that the proposed method significantly outperforms traditional denoising techniques and self-supervised models such as Wav2Vec 2.0, and also surpasses baselines relying on ASR-derived transcriptions or ground-truth text. The approach establishes a novel paradigm for emotion recognition under low-resource and high-noise conditions.

3 citationsRead paper

On estimating the trace of quantum state powers

Oct 17, 2024Electron. Colloquium Comput. Complex.

This work addresses the efficient quantum estimation of the Tsallis entropy (S_q( ho)) for an (n)-qubit mixed state ( ho) with (q geq 1), which is computationally equivalent to estimating the trace power (operatorname{tr}( ho^q)). To overcome the long-standing exponential complexity barrier, we propose the first embedded algorithmic framework based on uniform approximation by positive-power functions and Quantum Singular Value Transformation (QSVT). We uncover a sharp computational phase transition between (q = 1) and (q > 1), and establish a BQP/QSZK-completeness dichotomy for the Tsallis Quantum Entropy Difference problem ( ext{TsallisQED}_q). Our algorithm achieves (operatorname{poly}(n)) time complexity for (q geq 1 + Omega(1)), yielding exponential speedup over prior (exp(n))-time methods. Furthermore, we prove that purity estimation is BQP-complete, and demonstrate the inherent hardness of von Neumann entropy approximation under the QSZK assumption.

2 citationsRead paper

Parameterized Spanning Tree Congestion

Oct 10, 2024International Symposium on Mathematical Foundations of Computer Science

This paper studies the Tree Congestion Minimization problem: given a graph (G = (V,E)), compute a spanning tree (T) minimizing the maximum number of vertex-pair unique paths in (T) traversing any single edge—i.e., the edge congestion. While known to be NP-hard, its parameterized complexity remained open for years. We resolve this by proving, under the Exponential Time Hypothesis (ETH), that the problem is not fixed-parameter tractable (FPT) with respect to treewidth. Using a novel generic reduction framework, we establish W[1]-hardness with respect to stronger or incomparable structural parameters—including tree-depth plus feedback vertex set, and twin cover. Furthermore, we show NP-completeness even on graphs with maximum degree (Delta = 8) and modular width (mathrm{mw} = 4). These results comprehensively settle multiple long-standing open questions and significantly advance the theoretical boundaries of structural parameterized algorithms.

2 citationsRead paper

Quantifying Statistical Significance in Diffusion-Based Anomaly Localization via Selective Inference

Feb 19, 2024

Image anomaly localization is critical in medical diagnosis and industrial inspection, yet existing generative-model-based approaches—particularly diffusion models—lack statistical reliability, suffer from model bias and uncertainty, and fail to quantify false-positive risk. This paper introduces selective inference to diffusion-based anomaly localization for the first time, establishing an interpretable statistical inference framework: for each pixel or region in the model-reconstructed image, it performs conditional hypothesis testing and outputs rigorously calibrated p-values to quantify the false-positive probability. Unlike conventional methods lacking theoretical guarantees, our approach enables statistically controlled, significance-aware anomaly localization. Experiments on multiple medical and industrial datasets demonstrate substantial improvements in false-positive rate control, delivering trustworthy, statistically grounded anomaly localization outputs suitable for high-stakes applications.

2 citationsRead paper

Recent publications

Latest Papers

Validation-Centric AI-Assisted GPU Porting of a 250,000+ Line Legacy Weather Simulation Code

Aug 13, 2026

This work addresses the tension between performance gains and scientific validity when porting large legacy scientific codes to GPUs by proposing a verification-centric, AI-assisted migration workflow. The approach integrates a large language model–driven CLI agent, OpenACC-based automated code transformation, and physics-informed kernel benchmark generation, ensuring consistency through both element-wise numerical comparison and application-level meteorological simulations. For the first time, scientific validation is deeply embedded into an AI-assisted porting pipeline, enabling automatic detection of floating-point semantic discrepancies and branch sensitivity, while highlighting the critical roles of conversational context management and runtime state reconstruction. Applied to the 250K-line Fortran weather model CReSS, the method successfully produced verified GPU implementations for 162 core kernels, achieving a 5.1× speedup in real typhoon simulations and uncovering five instances of numerical divergence, substantially reducing migration costs.

0 citationsRead paper

Intuitive Hand Positional Guidance Using McKibben-Based Surface Tactile Sensations to Shoulder and Elbow

Aug 10, 2026

This study addresses the limited intuitiveness of conventional haptic guidance methods, which typically require users to pre-learn mappings between signals and target positions. To overcome this limitation, the authors propose a wearable fabric-based actuator utilizing McKibben-type pneumatic artificial muscles, applied for the first time to the shoulder–elbow region to deliver surface haptic cues that intuitively guide six distinct upper-limb movements. By integrating the equilibrium-point hypothesis with the Weber–Fechner law, the system dynamically modulates stimulation intensity, enabling high-accuracy guidance without prior user training. Experimental results demonstrate that the proposed approach significantly outperforms traditional voice synthesis and vibrotactile schemes in both guidance speed and precision.

0 citationsRead paper

Intuitive Directional Sense Presentation to the Torso Using McKibben-Based Surface Haptic Sensation in Immersive Space

Aug 10, 2026

This study addresses the challenge of delivering intuitive navigational cues in immersive environments, where visual overload often impedes users’ ability to perceive critical guidance. Existing haptic approaches typically lack intuitiveness and require extensive user training. To overcome these limitations, this work proposes a wearable fabric-based actuator leveraging McKibben artificial muscles, which dynamically modulates pneumatic pressure to generate spatially distributed tactile stimuli across large areas of the torso. This enables vision-free, learnable-free directional guidance through intuitive somatosensory feedback. The research presents the first application of McKibben fabric actuators for conveying directional awareness during full-body movement, integrated within a mixed-reality system. User studies and statistical analyses confirm the approach’s efficacy in directional perception, with successful deployment demonstrated in teleoperated micro-manipulation tasks.

0 citationsRead paper

Privileged Solutions or Context-Induced Teacher Behavior? Dissecting On-Policy Self-Distillation

Aug 10, 2026

This work investigates the true source of performance gains in On-Policy Self-Distillation (OPSD): whether they stem from the teacher model’s privileged access to ground-truth solutions or from contextual shifts induced by these solutions that alter teacher behavior. To disentangle these factors, we propose OP²SD, a novel method that substitutes the teacher’s input with solutions from other problems while preserving the student’s trajectory and distillation target. Experiments across three large language models and three mathematical reasoning benchmarks demonstrate that OP²SD not only significantly outperforms baseline approaches but also matches the performance of standard OPSD. These findings reveal that context-induced changes in teacher behavior—not privileged information—are the key driver of improvement, thereby challenging conventional interpretations of the OPSD mechanism.

0 citationsRead paper

Immersive Micromanipulation Integrating Pipette and Injector Operations with McKibben-Based Haptic Sensations for Workload Reduction

Aug 10, 2026

This study addresses the high cognitive load in conventional intracytoplasmic sperm injection (ICSI) procedures, which rely solely on visual feedback and require frequent switching between pipette and injector interfaces while lacking haptic cues. To overcome these limitations, this work proposes an immersive micromanipulation system that, for the first time, integrates McKibben pneumatic artificial muscles to deliver multidimensional haptic feedback. By unifying the human–machine interface and employing a tactile mapping algorithm, the system enables one-handed simultaneous control of aspiration and injection while providing real-time haptic perception of fluid uptake, ejection, and oocyte contact. Experimental results demonstrate that the proposed system significantly enhances operational efficiency, reduces cognitive workload, and improves overall usability.

0 citationsRead paper