检索结果-内蒙古大学图书馆

E2EGI: End-to-End Gradient Inversion in Federated Learning

IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS 2023年第2期27卷 756-767页

作者： Li, Zhaohua Wang, L. Chen, Guangyao Zhang, Zhiqiang Shafiq, Muhammad Gu, Zhaoquan Guangzhou Univ Cyberspace Inst Adv Technol Guangzhou 510006 Peoples R China Dept New Networks Peng Cheng Lab Shenzhen 518066 Peoples R China

A plethora of healthcare data is produced every day due to the proliferation of prominent technologies such as Internet of Medical Things (IoMT). Digital-driven smart devices like wearable watches, wristbands and bracelets are utilized extensively in modern healthcare applications. Mining valuable information from the data distributed at the owners' level is useful, but it is challenging to preserve data privacy. Federated learning (FL) has swiftly surged in popularity due to its efficacy in dealing privacy vulnerabilities. Recent studies have demonstrated that Gradient Inversion Attack (GIA) can reconstruct the input data by leaked gradients, previous work demonstrated the achievement of GIA in very limited scenarios, such as the label repetition rate of the target sample being low and batch sizes being smaller than 48. In this paper, a novel method of End-to-End Gradient Inversion (E2EGI) is proposed. Compared to the state-of-the-art method, E2EGI's Minimum Loss Combinatorial Optimization (MLCO) has the ability to realize reconstructed samples with higher similarity, and the Distributed Gradient Inversion algorithm can implement GIA with batch sizes of 8 to 256 on deep network models (such as ResNet-50) and ImageNet datasets. A new Label reconstruction algorithm is developed that relies only on the gradient information of the target model, which can achieve a label reconstruction accuracy of 81% in one batch sample with a label repetition rate of 96%, a 27% improvement over the state-of-the-art method. This proposed work can underpin data security assessments for healthcare federated learning.

关键词： Image reconstruction Training Collaborative work Data privacy Data models Optimization reconstruction algorithms deep learning federated learning gradient inversion

来源：评论

学校读者我要写书评

暂无评论

Transcending shift-invariance in the paraxial regime via end-to-end inverse design of freeform nanophotonics

引用

OPTICS EXPRESS 2023年第15期31卷 24260-24272页

作者： Li, William F. Arya, Gaurav Roques-Carmes, Charles Lin, Zin Johnson, Steven G. Soljacic, Marin MIT Res Lab Elect Cambridge MA 02139 USA MIT Dept Math Cambridge MA 02139 USA Virginia Tech Bradley Dept Elect & Comp Engn Blacksburg VA 24060 USA MIT Dept Phys Cambridge MA 02139 USA

Traditional optical elements and conventional metasurfaces obey shift-invariance in the paraxial regime. For imaging systems obeying paraxial shift-invariance, a small shift in input angle causes a corresponding shift in the sensor image. Shift-invariance has deep implications for the design and functionality of optical devices, such as the necessity of free space between components (as in compound objectives made of several curved surfaces). We present a method for nanophotonic inverse design of compact imaging systems whose resolution is not constrained by paraxial shift-invariance. Our method is end-to-end, in that it integrates density-based full-Maxwell topology optimization with a fully iterative elastic-net reconstruction algorithm. By the design of nanophotonic structures that scatter light in a non-shift-invariant manner, our optimized nanophotonic imaging system overcomes the limitations of paraxial shift-invariance, achieving accurate, noise-robust image reconstruction beyond shift-invariant resolution.

关键词： Image reconstruction Image sensors Imaging systems Optical devices Optical elements reconstruction algorithms

来源：评论

学校读者我要写书评

暂无评论

Quantum detector tomography applied to the human visual system: a feasibility study

引用

JOURNAL OF THE OPTICAL SOCIETY OF AMERICA A-OPTICS IMAGE SCIENCE AND VISION 2023年第2期40卷 285-293页

作者： Van Der Reep, T. H. A. Molenaar, D. Loffler, W. Pinto, Y. Psychol Res Inst Nieuwe Achtergracht 129b NL-1018 WS Amsterdam Netherlands Leiden Inst Phys Niels Bohrweg 2 NL-2333 CA Leiden Netherlands

We show that quantum detector tomography can be applied to the human visual system to explore human per-ception of photon number states. In detector tomography, instead of using very hard-to-produce photon number states, the response of a detector to light pulses with known photon statistics of varying intensity is recorded, and a model is fitted to the experimental outcomes, thereby inferring the detector's photon number state response. Generally, light pulses containing a Poisson-distributed number of photons are utilized, which are very easy to produce in the lab. This technique has not been explored to study the human visual system before because it usually requires a very large number of repetitions not suitable for experiments on humans. Yet, in the present study we show that detector tomography is feasible for human experiments. Assuming a simple model for this accuracy, the results of our simulations show that detector tomography is able to reconstruct the model using Bayesian inference with as few as 5000 trials. We then optimize the experimental parameters in order to maximize the probability of showing that the single-photon accuracy is above chance. As such, our study opens the road to study human perception on the quantum level. (c) 2023 Optica Publishing Group

关键词： Avalanche photodiodes Light detection Photon statistics Quantum detectors reconstruction algorithms Visual system

来源：评论

学校读者我要写书评

暂无评论

Plug-and-Play Priors for Multi-Shot Compressive Hyperspectral Imaging

引用

IEEE TRANSACTIONS ON IMAGE PROCESSING 2023年 32卷 5326-5339页

作者： Xie, Ting Liu, Licheng Zhuang, Lina Hunan Normal Univ Coll Engn & Design Changsha 410081 Peoples R China Hunan Univ Coll Elect & Informat Engn Changsha 410082 Peoples R China Chinese Acad Sci Aerosp Informat Res Inst Key Lab Computat Opt Imaging Technol Beijing 100094 Peoples R China

Multi-shot coded aperture snapshot spectral imaging (CASSI) uses multiple measurement snapshots to encode the three-dimensional hyperspectral image (HSI). Increasing the number of snapshots will multiply the number of measurements, making CASSI system more appropriate for detailed spatial or spectrally rich scenes. However, the reconstruction algorithms still face the challenge of being ineffective or inflexible. In this paper, we propose a plug-and-play (PnP) method that uses denoiser as priors for multi-shot CASSI. Specifically, the proposed PnP method is based on the primal-dual algorithm with linesearch (PDAL), which makes it flexible and can be used for any multi-shot CASSI mechanisms. Furthermore, a new subspaced-based nonlocal reweighted low-rank (SNRL) denoiser is presented to utilize the global spectral correlation and nonlocal self-similarity priors of HSI. By integrating the SNRL denoiser into PnP-PDAL, we show the balloons ( $512\times 512\times31$ ) in CAVE dataset recovered from two snapshots compressive measurements with MPSNR above 50 dB. Experimental results demonstrate that our proposed method leads to significant improvements compared to the current state-of-the-art methods.

关键词： Image reconstruction Hyperspectral imaging Sensors Apertures reconstruction algorithms Image coding Optimization Coded aperture snapshot spectral imaging (CASSI) multi-shot CASSI primal-dual algorithm with linesearch plug-and-play (PnP) subspaced-based nonlocal reweighted low-rank (SNRL) denoiser

来源：评论

学校读者我要写书评

暂无评论

Foveated panoramic ghost imaging

引用

OPTICS EXPRESS 2023年第8期31卷 12986-13002页

作者： Cui, Huan Cao, Jie Hao, Qun Zhou, Dong Zhang, Haoyu Zhang, Yingqiang Minist Educ Beijing Inst Technol Sch Opt & Photon Key Lab Biomimet Robots & Syst Beijing 100081 Peoples R China Yangtze Delta Reg Acad Beijing Inst Technol Jiaxing 314003 Peoples R China Changchun Univ Sci & Technol Changchun 130022 Peoples R China

Panoramic ghost imaging (PGI) is a novel method by only using a curved mirror to enlarge the field of view (FOV) of ghost imaging (GI) to 360 degrees, making GI a breakthrough in the applications with a wide FOV. However, high-resolution PGI with high efficiency is a serious challenge because of the large amount of data. Therefore, inspired by the variant-resolution retina structure of human eye, a foveated panoramic ghost imaging (FPGI) is proposed to achieve the coexistence of a wide FOV, high resolution and high efficiency on GI by reducing the resolution redundancy, and further to promote the practical applications of GI with a wide FOV. In FPGI system, a flexible variant-resolution annular pattern structure via log-rectilinear transformation and log-polar mapping is proposed to be used for projection, which can allocate the resolution of the region of interest (ROI) and the other region of non-interest (NROI) by setting related parameters in the radial and poloidal directions independently to meet different imaging requirements. In addition, in order to reasonably reduce the resolution redundancy and avoid the loss of the necessary resolution on NROI, the variant-resolution annular pattern structure with a real fovea is further optimized to keep the ROI at any position in the center of 360 degrees FOV by flexibly changing the initial position of the start-stop boundary on the annular pattern structure. The experimental results of the FPGI with one fovea and multiple foveae demonstrate that, compared to the traditional PGI, the proposed FPGI not only can improve the imaging quality on the ROIs with a high resolution and flexibly remain a lower-resolution imaging on the NROI with different required resolution reduction;but also reduce the reconstruction time to improve the imaging efficiency due to the reduction of the resolution redundancy.(c) 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement

关键词： Ghost imaging Imaging techniques Light intensity reconstruction algorithms Terahertz imaging X ray imaging

来源：评论

学校读者我要写书评

暂无评论

Multiple-scattering simulator-trained neural network for intensity diffraction tomography

引用

OPTICS EXPRESS 2023年第3期31卷 4094-4107页

作者： Matlock, Alex Zhu, Jiabei Tian, Lei Boston Univ Dept Elect & Comp Engn Boston MA 02215 USA MIT Laser Biomed Res Ctr G R Harrison Spect Lab Cambridge MA 02139 USA

Recovering 3D phase features of complex biological samples traditionally sacrifices computational efficiency and processing time for physical model accuracy and reconstruction quality. Here, we overcome this challenge using an approximant-guided deep learning framework in a high-speed intensity diffraction tomography system. Applying a physics model simulator -based learning strategy trained entirely on natural image datasets, we show our network can robustly reconstruct complex 3D biological samples. To achieve highly efficient training and prediction, we implement a lightweight 2D network structure that utilizes a multi-channel input for encoding the axial information. We demonstrate this framework on experimental measurements of weakly scattering epithelial buccal cells and strongly scattering C. elegans worms. We benchmark the network's performance against a state-of-the-art multiple-scatter ing model-based iterative reconstruction algorithm. We highlight the network's robustness by reconstructing dynamic samples from a living worm video. We further emphasize the network's generalization capabilities by recovering algae samples imaged from different experimental setups. To assess the prediction quality, we develop a quantitative evaluation metric to show that our predictions are consistent with both multiple-scatter ing physics and experimental measurements.

关键词： Image quality Optical imaging Phase imaging reconstruction algorithms Scattering measurement Three dimensional imaging

来源：评论

学校读者我要写书评

暂无评论

Helical Artifact Reduction Method Using Image Segmentation With CNN Denoising Technique

引用

IEEE ACCESS 2023年 11卷 49261-49272页

作者： Choi, Seungwon Kim, Byeongjoon Park, Chulkyu Park, Jueon Kim, Yousuk Choi, Sungil Baek, Jongduk Yonsei Univ Yonsei Inst Convergence Technol Sch Integrated Technol Incheon 21983 South Korea VATECH Res & Dev Ctr Dept Artificial Intelligence Hwaseong Si 18449 Gyeonggi Do South Korea Yonsei Univ Coll Comp Dept Artificial Intelligence Seoul 03722 South Korea

Helical computed tomography (CT) scans are often performed to obtain three-dimensional images of an object that is longer than the detector. However, the existing quasi-exact and exact reconstruction methods, such as re-binning and Katsevich algorithm, generate interpolation errors or require high computational power. In this work, we propose a method to reconstruct helical CT projections by iteratively reducing helical artifacts. In each iteration, a convolutional neural network (CNN)-based denoising technique is used to accurately segment the prior image (bone and soft tissue image). The results indicate that the proposed algorithm reduces helical artifacts to a significantly greater extent than the existing single slice re-binning (SSR) and weighted filtered backprojection (W-FBP) methods.

关键词： Image reconstruction Bones Computed tomography Image segmentation Detectors Biological tissues reconstruction algorithms Helical CT helical artifact filtered backprojection reconstruction

来源：评论

学校读者我要写书评

暂无评论

Pixel-wise rational model for a structured light system

引用

OPTICS LETTERS 2023年第10期48卷 2712-2715页

作者： Vargas, Raul Romero, Lenny A. Zhang, Song Marrugo, Andres G. Univ Tecnol Bolivar Fac Ingn Cartagena Colombia Univ Tecnol Bolivar Fac Ciencias Basicas Cartagena Colombia Purdue Univ Sch Mech Engn W Lafayette IN 47907 USA

This Letter presents a novel structured light system model that effectively considers local lens distortion by pixel-wise rational functions. We leverage the stereo method for ini-tial calibration and then estimate the rational model for each pixel. Our proposed model can achieve high measure-ment accuracy within and outside the calibration volume, demonstrating its robustness and accuracy. (c) 2023 Optica Publishing Group

关键词： Calibration CMOS cameras Distortion reconstruction algorithms Structured light Three dimensional reconstruction

来源：评论

学校读者我要写书评

暂无评论

Robust phase unwrapping via non-local regularization

引用

OPTICS LETTERS 2023年第6期48卷 1399-1402页

作者： Bian, Liheng Wang, Xin LI, Daoyu Ren, Qiuling Zheng, Dezhi Beijing Inst Technol Adv Res Inst Multidisciplinary Sci Beijing 100081 Peoples R China Beijing Inst Technol Sch Informat & Elect Beijing 100081 Peoples R China Beijing Inst Technol MIIT Key Lab Complex Field Intelligent Sensing Beijing 100081 Peoples R China

Phase unwrapping is an indispensable step in recovering the true phase from a modulo-2 pi phase. Conventional phase unwrapping methods suffer from error propagation under severe noise. In this Letter, we propose an iterative frame-work for robust phase unwrapping with high fidelity. The proposed method utilizes the transport-of-intensity equation to solve the phase unwrapping problem with high computational efficiency. To further improve reconstruction accuracy, we take advantage of non-local structural similarity using low-rank regularization. Meanwhile, we use an adaptive iteration strategy that dynamically and automatically updates the denoising parameter to avoid over-smoothing and preserve image details. A set of simulation and experimental results validates the proposed method, which can provide satisfying results under severe noise conditions, and outperform existing state-of-the-art phase unwrapping methods with at least 6 dB higher peak SNR (PSNR).(c) 2023 Optica Publishing Group

关键词： Discrete Fourier transforms Phase unwrapping Phased arrays reconstruction algorithms Spatial frequency Synthetic aperture radar

来源：评论

学校读者我要写书评

暂无评论

Photonics-assisted compressed sensing radar receiver for frequency domain non-sparse signal sampling based on dictionary learning

引用

OPTICS LETTERS 2023年第3期48卷 767-770页

作者： LIu, Shiyang Wang, Shi Shi, Taixia Chen, Yang East China Normal Univ Shanghai Key Lab Multidimens Informat Proc Shanghai 200241 Peoples R China East China Normal Univ Engn Ctr SHMEC Space Informat & GNSS Shanghai 200241 Peoples R China

A photonics-assisted radar receiver based on compressed sensing (CS) technology is proposed to receive frequency domain non-sparse radar signals. The radar echo signal is mixed with a pseudo-random binary sequence (PRBS) in a photonic random demodulator (RD) consisting of a laser diode (LD), a dual-drive Mach-Zehnder modulator (DD-MZM), and a photodetector (PD). After the mixed signal from the photonic RD is undersampled by an analog-to -digital converter (ADC), the echo signal is reconstructed in the digital domain using an overcomplete dictionary gen-erated by the dictionary learning algorithm and sparse reconstruction algorithm. The target range can be then obtained by correlating the reconstructed echo signal and the transmitted one. Experimental results show that the pro-posed system can successfully reconstruct different kinds of undersampled non-sparse radar echo signals. When the compression ratio is 20, the ranging errors do not exceed 5 cm. The system provides a promising solution for recov-ering undersampled frequency domain non-sparse radar signals through photonics-assisted CS.(c) 2023 Optica Pub-lishing Group

关键词： Analog to digital converters Diode lasers Microwave photonic filters Radar reconstruction algorithms Spatial light modulators

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：