arXiv subjects

Leilei Gu

Publications and source records attributed to Leilei Gu.

3 recordsLinked to original sources

arXiv2025

MicroAUNet: Boundary-Enhanced Multi-scale Fusion with Knowledge Distillation for Colonoscopy Polyp Image Segmentation

Early and accurate segmentation of colorectal polyps is critical for reducing colorectal cancer mortality, which has been extensively explored by academia and industry. However, current deep learning-based polyp segmentation models either compromise clinical decision-making by providing ambiguous polyp margins in segmentation outputs or rely on heavy architectures with high computational complexity, resulting in insufficient inference speeds for real-time colorectal endoscopic applications. To address this problem, we propose MicroAUNet, a lightweight attention-based segmentation network, which synergistically combines depthwise-separable dilated convolutions with a single-path, parameter-shared channel-spatial attention block to effectively strengthen multi-scale boundary features. On the basis of it, a progressive two-stage knowledge-distillation scheme is introduced to transfer semantic and boundary cues from a high-capacity teacher. Extensive experiments on benchmarks also demonstrate the state-of-the-art accuracy under extremely low model complexity, indicating that MicroAUNet is suitable for real-time clinical polyp segmentation. The code is publicly available at https://github.com/JeremyXSC/MicroAUNet.

cs.CV↗

arXiv2024

Synergistic Development of Perovskite Memristors and Algorithms for Robust Analog Computing

Analog computing using non-volatile memristors has emerged as a promising solution for energy-efficient deep learning. New materials, like perovskites-based memristors are recently attractive due to their cost-effectiveness, energy efficiency and flexibility. Yet, challenges in material diversity and immature fabrications require extensive experimentation for device development. Moreover, significant non-idealities in these memristors often impede them for computing. Here, we propose a synergistic methodology to concurrently optimize perovskite memristor fabrication and develop robust analog DNNs that effectively address the inherent non-idealities of these memristors. Employing Bayesian optimization (BO) with a focus on usability, we efficiently identify optimal materials and fabrication conditions for perovskite memristors. Meanwhile, we developed "BayesMulti", a DNN training strategy utilizing BO-guided noise injection to improve the resistance of analog DNNs to memristor imperfections. Our approach theoretically ensures that within a certain range of parameter perturbations due to memristor non-idealities, the prediction outcomes remain consistent. Our integrated approach enables use of analog computing in much deeper and wider networks, which significantly outperforms existing methods in diverse tasks like image classification, autonomous driving, species identification, and large vision-language models, achieving up to 100-fold improvements. We further validate our methodology on a 10$\times$10 optimized perovskite memristor crossbar, demonstrating high accuracy in a classification task and low energy consumption. This study offers a versatile solution for efficient optimization of various analog computing systems, encompassing both devices and algorithms.

cs.LG↗

arXiv2024

Dynamic Nanodomains Dictate Macroscopic Properties in Lead Halide Perovskites

Empirical A-site cation substitution has advanced the stability and efficiency of hybrid organic-inorganic lead halide perovskites solar cells and the functionality of X-ray detectors. Yet, the fundamental mechanisms underpinning their unique performance remain elusive. This multi-modal study unveils the link between nanoscale structural dynamics and macroscopic optoelectronic properties in these materials by utilising X-ray diffuse scattering, inelastic neutron spectroscopy and optical microscopy complemented by state-of-the-art machine learning-assisted molecular dynamics simulations. Our approach uncovers the presence of dynamic, lower-symmetry local nanodomains embedded within the higher-symmetry average phase in various perovskite compositions. The properties of these nanodomains are tunable via the A-site cation selection: methylammonium induces a high density of anisotropic, planar nanodomains of out-of-phase octahedral tilts, while formamidinium favours sparsely distributed isotropic, spherical nanodomains with in-phase tilting, even when crystallography reveals cubic symmetry on average. The observed variations in the properties of dynamic nanodomains are in agreement with our simulations and are directly linked to the differing macroscopic optoelectronic and ferroelastic behaviours of these compositions. By demonstrating the influence of A-site cation on local nanodomains and consequently, on macroscopic properties, we propose leveraging this relationship to engineer the optoelectronic response of these materials, propelling further advancements in perovskite-based photovoltaics, optoelectronics, and X-ray imaging.

cond-mat.mtrl-sci↗