Selected work

Representative Papers

Reservoir computing and photoelectrochemical sensors: A marriage of convenience

Jul 01, 2023Coordination chemistry reviews

To address the need for real-time, low-power detection of complex chemical components in biological fluids and environmental samples, existing photoelectrochemical (PEC) sensors face critical bottlenecks—including reliance on energy-intensive digital hardware for signal processing and insufficient robustness. This work introduces, for the first time, physical reservoir computing (PRC) into PEC sensing systems, leveraging the intrinsic nonlinear dynamics of the sensor itself as a natural analog computational resource to enable event-driven, in-situ information processing. By eliminating conventional digital signal processing modules, the approach drastically reduces power consumption and latency. In dynamic detection tasks for glucose and dopamine, the system achieves millisecond-scale response times and 98.2% classification accuracy, while reducing power consumption by two orders of magnitude compared to standard approaches. This work establishes a novel paradigm for neuromorphic sensing, brain-inspired molecular perception, and edge-intelligent chemical sensing.

14 citationsRead paper

MIDAS: Mixing Ambiguous Data with Soft Labels for Dynamic Facial Expression Recognition

Jan 03, 2024IEEE Workshop/Winter Conference on Applications of Computer Vision

In dynamic facial expression recognition (DFER) under unconstrained real-world conditions, motion blur and semantic ambiguity of expressions severely hinder accurate classification. To address this, we propose the first video-level soft-label mixup augmentation method, which jointly performs convex interpolation across video frames and their corresponding multi-emotion probability soft labels—explicitly modeling both expression continuity and semantic uncertainty. Our approach comprises three components: (1) soft label construction via emotion distribution estimation, (2) soft-label-guided frame-level mixup augmentation, and (3) an end-to-end trainable framework. Evaluated on the DFEW benchmark, our method achieves significant improvements over existing state-of-the-art methods, demonstrating that soft-label mixing enhances model robustness to ambiguous, dynamically evolving expressions in the wild. This work establishes a novel paradigm for uncertainty-aware learning in DFER, advancing the integration of probabilistic semantics into video-based representation learning.

3 citations1 influentialRead paper

Asymptotic evaluation of the information processing capacity in reservoir computing

Feb 15, 2025arXiv.org

Existing methods for estimating the information processing capacity (IPC) of reservoir computing (RC) systems lack theoretical rigor and suffer from systematic bias when applied to infinitely long time series, as they rely exclusively on finite-length data. Method: This paper proposes a novel IPC estimation algorithm based on asymptotic expansion and least-squares estimation. It establishes, for the first time, a rigorous asymptotic expansion framework for IPC in the infinite-time limit, integrating stochastic dynamical systems modeling, asymptotic analysis, and least-squares regression to analytically characterize and efficiently estimate the leading-order term of IPC. Results: Numerical experiments demonstrate consistently high accuracy and strong generalizability across diverse RC architectures—including echo state networks (ESNs) and liquid state machines (LSMs)—thereby significantly enhancing the reliability and theoretical soundness of long-term performance evaluation.

1 citationsRead paper

Bright-Channel Retinex Enhancement with a Conditional Overdispered-Noise Analysis

Aug 10, 2026

This work addresses the heteroscedastic noise amplification inherent in Retinex-based low-light image enhancement by proposing an unsupervised, non-learning method. The approach integrates bright channel prior–based illumination estimation, Retinex reflectance decomposition, and edge-preserving denoising, and—within a Retinex framework—introduces for the first time a conditional negative binomial pseudo-count model to characterize the over-dispersed noise induced by division operations. It further provides a boundary-constrained maximum likelihood solution for zero-valued observations, eliminating the need for sensor calibration or deep learning. Experimental results demonstrate state-of-the-art performance among traditional methods on the LOL-v1 dataset, achieving 17.74 dB PSNR and 0.739 SSIM, while attaining real-time processing at 43 FPS on an Apple M2 Pro for 600×400 resolution images.

0 citationsRead paper

DeepFreqMark: End-To-End Learnable Frequency-Domain Watermarking with Spherical Attack Simulation for Latent Diffusion Models

Aug 09, 2026

This work addresses the challenges of copyright protection and forgery detection in images generated by latent diffusion models, where existing frequency-domain watermarking methods suffer from limited payload capacity and inflexible pattern design. To overcome these limitations, we propose the first end-to-end learnable frequency-domain watermarking framework, which introduces neural encoders and decoders into the latent space to replace handcrafted watermark designs. By incorporating spherical linear interpolation (Slerp) to simulate realistic attacks, our approach preserves the characteristics of Gaussian perturbations while circumventing the computational bottleneck associated with DDIM inversion. The proposed method supports message payloads of up to 256 bits and achieves significantly lower bit error rates under real-world attacks, outperforming current state-of-the-art baselines.

0 citationsRead paper

Recent publications

Latest Papers

Bright-Channel Retinex Enhancement with a Conditional Overdispered-Noise Analysis

Aug 10, 2026

This work addresses the heteroscedastic noise amplification inherent in Retinex-based low-light image enhancement by proposing an unsupervised, non-learning method. The approach integrates bright channel prior–based illumination estimation, Retinex reflectance decomposition, and edge-preserving denoising, and—within a Retinex framework—introduces for the first time a conditional negative binomial pseudo-count model to characterize the over-dispersed noise induced by division operations. It further provides a boundary-constrained maximum likelihood solution for zero-valued observations, eliminating the need for sensor calibration or deep learning. Experimental results demonstrate state-of-the-art performance among traditional methods on the LOL-v1 dataset, achieving 17.74 dB PSNR and 0.739 SSIM, while attaining real-time processing at 43 FPS on an Apple M2 Pro for 600×400 resolution images.

0 citationsRead paper

DeepFreqMark: End-To-End Learnable Frequency-Domain Watermarking with Spherical Attack Simulation for Latent Diffusion Models

Aug 09, 2026

This work addresses the challenges of copyright protection and forgery detection in images generated by latent diffusion models, where existing frequency-domain watermarking methods suffer from limited payload capacity and inflexible pattern design. To overcome these limitations, we propose the first end-to-end learnable frequency-domain watermarking framework, which introduces neural encoders and decoders into the latent space to replace handcrafted watermark designs. By incorporating spherical linear interpolation (Slerp) to simulate realistic attacks, our approach preserves the characteristics of Gaussian perturbations while circumventing the computational bottleneck associated with DDIM inversion. The proposed method supports message payloads of up to 256 bits and achieves significantly lower bit error rates under real-world attacks, outperforming current state-of-the-art baselines.

0 citationsRead paper

A Low-Cost Hybrid Reservoir Computing Model for Isolated Sign Language Video Recognition

Aug 04, 2026

This work addresses the high computational cost and limited deployability of existing deep learning approaches for isolated sign language recognition on edge devices by proposing a lightweight hybrid reservoir computing model. The method leverages MediaPipe to extract hand and body keypoints, then efficiently captures spatiotemporal dynamics by integrating Deep Reservoir Computing (DRC) with Bidirectional Reservoir Computing (BRC), followed by fast classification via ridge regression. Evaluated on the WLASL100 dataset, the model achieves 61.12% Top-1, 86.05% Top-5, and 92.56% Top-10 accuracy with only a few seconds of training time. This approach substantially reduces computational resource requirements while maintaining competitive recognition performance, making it well-suited for deployment on resource-constrained edge devices.

0 citationsRead paper

Diffusion-Based Body Schema Learning Enabling Abnormal-State Adaptation in Musculoskeletal Robots

Aug 02, 2026

This work addresses the challenge of maintaining a consistent body schema in musculoskeletal robots under anomalous conditions such as muscle tears or actuator jams. To this end, the authors propose a diffusion model–based framework for body schema learning that eschews conventional low-dimensional latent space modeling and instead directly operates in the high-dimensional sensor–actuator space. By leveraging gradient-guided denoising, the method estimates physically consistent state variables without requiring retraining, even when faced with out-of-distribution scenarios or partial observations. Physical constraints are explicitly incorporated into the estimation process, ensuring plausibility under perturbations. In simulated musculoskeletal environments, the framework accurately recovers muscle lengths and tensions despite severe disruptions like muscle rupture or actuator lock-up, thereby substantially enhancing the system’s robustness and adaptability.

0 citationsRead paper

Reconstructing Backpropagation from Forward Fluctuations in Noise-modulated Neural Networks

Jul 29, 2026

This work addresses the challenge that conventional backpropagation—relying on weight transposes and backward signal propagation—is difficult to implement on biological or neuromorphic hardware. The authors propose a gradient estimation method that operates solely through forward pathways. By injecting uniform noise into neural activities, the approach leverages local covariance statistics to unbiasedly reconstruct gradients, eliminating the need for weight mirroring or reverse data flow. Integrated with local differential error propagation, per-weight Adam optimization, and a noise-based polynomial comparator circuit, the method achieves accuracy comparable to standard backpropagation on simple regression tasks. Critically, its gradient estimates are nearly unbiased, substantially enhancing deployability on digital neuromorphic systems.

0 citationsRead paper