检索结果-南通市图书馆

Optimization-inspired manual architecture design and neural architecture search

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2023年第11期66卷 100-112页

作者： Yibo YANG Zhengyang SHEN Huan LI zhouchen lin JD Explore Academy School of Mathematical Sciences Peking University Institute of Robotics and Automatic Information Systems College of Artificial Intelligence Nankai University Key Lab of General Artificial Intelligence School of Intelligence Science and Technology Peking University Institute for Artificial Intelligence Peking University Pazhou Lab

Neural architecture has been a research focus in recent years due to its importance in deciding the performance of deep networks. Representative ones include a residual network(Res Net) with skip connections and a dense network(Dense Net) with dense connections. However, a theoretical guidance for manual architecture design and neural architecture search(NAS) is still lacking. In this paper, we propose a manual architecture design framework, which is inspired by optimization algorithms. It is based on the conjecture that an optimization algorithm with a good convergence rate may imply a neural architecture with good performance. Concretely, we prove under certain conditions that forward propagation in a deep neural network is equivalent to the iterative optimization procedure of the gradient descent algorithm minimizing a cost function. Inspired by this correspondence, we derive neural architectures from fast optimization algorithms,including the heavy ball algorithm and Nesterov’s accelerated gradient descent algorithm. Surprisingly, we find that we can deem the Res Net and Dense Net as special cases of the optimization-inspired *** architectures offer not only theoretical guidance, but also good performances in image recognition on multiple datasets, including CIFAR-10, CIFAR-100, and Image Net. Moreover, we show that our method is also useful for NAS by offering a good initial search point or guiding the search space.

关键词： deep neural network manual architecture design neural architecture search image recognition optimization algorithms learning-based optimization

Embodied computational imaging: a new paradigm for observing and analyzing spatiotemporally ultrasensitive phenomena at multiple scales

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2024年第11期67卷 354-356页

作者： Baoquan CHEN zhouchen lin Peng XI Yebin LIU Xiaodian CHEN School of Intelligence Science and Technology Peking University College of Future Technology Peking University Department of Automation Tsinghua University National Astronomical Observatories Chinese Academy of Sciences

Imaging is the foundational tool for scientists to observe and understand the world. It has served as the basis for scientific discoveries over human history. In recent years, with the rapid advancement of artificial intelligence and computing technologies, a new interdisciplinary field, computational imaging, has emerged and developed. By jointly optimizing imaging hardware and processing algorithms, computational imaging continuously breaks through the physical limits of traditional imaging systems in aspects such as resolution,dimension, contrast, and throughput.

关键词：

Newton design:designing CNNs with the family of Newton's methods

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2023年第6期66卷 122-137页

作者： Zhengyang SHEN Yibo YANG Qi SHE Changhu WANG Jinwen MA zhouchen lin School of Mathematical Sciences Peking University Bytedance AI Lab JD Explore Academy Key Laboratory of Machine Perception School of Intelligence Science and TechnologyPeking University Pazhou Lab

Nowadays, convolutional neural networks(CNNs)have led the developments of machine ***, most CNN architectures are obtained by manual design, which is empirical, time-consuming, and non-transparent. In this paper, we aim at offering better insight into CNN models from the perspective of optimization theory. We propose a unified framework for understanding and designing CNN architectures with the family of Newton’s methods, which is referred to as Newton design. Specifically, we observe that the standard feedforward CNN model(PlainNet)solves an optimization problem via a kind of quasi-Newton method. Interestingly, residual network(ResNet)can also be derived if we use a more general quasi-Newton method to solve this problem. Based on the above observations, we solve this problem via a better method,the Newton-conjugate-gradient(Newton-CG)method, which inspires Newton-CGNet. In the network design,we translate binary-value terms in the optimization schemes to dropout layers, so dropout modules naturally appear in the derived CNN structures with specific locations, rather than being an empirical training *** experiments on image classification and text categorization tasks verify that Newton-CGNets perform very competitively. Particularly, Newton-CGNets surpass their counterparts ResNets by over 4% on CIFAR-10 and over 10% on CIFAR-100, respectively.

关键词： CNN dropout optimization method network design Newton's method

同方期刊数据库

石墨烯-富勒烯铵碘盐复合载体负载Pd催化剂的制备及电催化氧化乙醇性能

在线全文

学校读者我要写书评

暂无评论

高等学校化学学报 2019年第5期40卷 1013-1018页

作者：林周晨黄巧茜雷鸣浙江大学化学系杭州310027

采用水合肼水热还原法制备了不同比例还原氧化石墨烯(RGO)与n型自掺杂富勒烯铵碘盐(PCBANI)的复合载体RGO-PCBANI,并在电极上用这些载体负载Pd纳米粒子制备了Pd/RGO-PCBANI电催化剂.利用扫描电子显微镜(SEM)、透射电子显微镜(TEM)、X射... 详细信息

采用水合肼水热还原法制备了不同比例还原氧化石墨烯(RGO)与n型自掺杂富勒烯铵碘盐(PCBANI)的复合载体RGO-PCBANI,并在电极上用这些载体负载Pd纳米粒子制备了Pd/RGO-PCBANI电催化剂.利用扫描电子显微镜(SEM)、透射电子显微镜(TEM)、X射线衍射(XRD)及X射线光电子能谱(XPS)对RGO-PCBANI和Pd/RGO-PCBANI的形貌及结构进行了表征.利用循环伏安和计时电流等电化学方法研究了该催化剂电催化氧化乙醇的性能.结果表明,所制备的RGO-PCBANI(6∶1)载体的分散性较好,用其负载的Pd纳米粒子平均粒径为5. 2 nm,且Pd/RGO-PCBANI(6∶1)催化剂的催化活性最好,质量电流密度达到1288. 8 m A/mg.

关键词：石墨烯-富勒烯复合载体间隔材料钯电催化剂燃料电池乙醇氧化

在线全文

学校读者我要写书评

暂无评论

用二进小波消除磁共振图像中的振铃效应

模式识别与人工智能 1999年第3期12卷 320-324页

作者：林宙辰石青云北京大学视觉与听觉信息处理国家重点实验室北京100871

核磁共振(NMR)成像中为了节省成像时间,往往只采集一部分NMR信号,余下部分用0补足,致使重构的图像产生振铃效应。本文提出了用二进小波消除振铃效应的新方法。该方法利用图像每一列X_i适当层次的二进小波平滑像X_i(或相邻两层二进小波... 详细信息

核磁共振(NMR)成像中为了节省成像时间,往往只采集一部分NMR信号,余下部分用0补足,致使重构的图像产生振铃效应。本文提出了用二进小波消除振铃效应的新方法。该方法利用图像每一列X_i适当层次的二进小波平滑像X_i(或相邻两层二进小波平滑像的线性组合)作为该列的光滑平均,最后结果是X_i与(?)_i的线性内插。内插系数是自适应的,既能保持强边缘,又能平滑平坦部分的灰度波动。

关键词：二进小波振铃效应核磁共振图像医学诊断

在线全文

学校读者我要写书评

暂无评论

NeurIPS 2020观察与分析

中国图象图形学报 2021年第2期26卷 229-244页

作者：林宙辰王奕森北京大学信息科学技术学院机器感知与智能教育部重点实验室北京100871

神经信息处理系统大会(Conference on Neural Information Processing Systems,Neur IPS)是机器学习领域的顶级会议,在中国计算机学会(China Computer Federation,CCF)推荐国际学术会议中被评为人工智能领域的A类会议,一直广受关注。Neur IPS 2020收到了创纪录的9467篇投稿,最终录用1898篇论文。收录的论文涵盖了人工智能的各种主题,包括深度学习及其应用、强化学习与规划、纯理论研究、概率方法、优化及机器学习与社会等。本文回顾了Neur IPS 2020的亮点及论文录用情况,详细解读了特邀报告、最佳论文、口头报告及部分海报论文,希望能帮助读者快速了解Neur IPS 2020的盛况。

关键词：人工智能机器学习深度学习强化学习理论优化学术会议 NeurIPS 2020

在线全文

学校读者我要写书评

暂无评论

基于规则的连机英语字符笔迹标识的修正

图象识别与自动化 2006年第2期 1-15,29页

作者： zhouchen lin Rongrong Wang Heung - Yeung Shum 李宏（译）

近来，许多基于书写的方式使人们自然书写的数字笔迹输入成为可能。通常，书写时的污点或修改不仅使文本受污，而且看上去也不舒服，还影响到手写体的识别。本文首先论述笔迹修正的问题。我们提出了去除污点和修改笔迹的修正系统，使文... 详细信息

近来，许多基于书写的方式使人们自然书写的数字笔迹输入成为可能。通常，书写时的污点或修改不仅使文本受污，而且看上去也不舒服，还影响到手写体的识别。本文首先论述笔迹修正的问题。我们提出了去除污点和修改笔迹的修正系统，使文本变得清晰、可识别，以改善手写体的识别率。基于规则的算法可处理大部分如单一笔划的同一笔划重复涂写、笔划间的中间部分重复涂写、改错、修改、插入以及书写顺序错误等情况。实验结果显示，该系统对笔迹标识的修正是有效的，并有希望改善其识别率。

关键词：文本分析文本和正文编辑手写体分析手写体识别

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

多媒体信号处理的数学理论前沿进展

中国图象图形学报 2020年第1期25卷 1-18页

作者：熊红凯戴文睿林宙辰吴飞于俊清申扬眉徐明星上海交通大学电子信息与电气工程学院上海200240 北京大学信息科学技术学院北京100080 浙江大学计算机科学与技术学院杭州310027 华中科技大学计算机科学与技术学院武汉430074

深度学习模型广泛应用于多媒体信号处理领域,通过引入非线性能够极大地提升性能,但是其黑箱结构无法解析地给出最优点和优化条件。因此如何利用传统信号处理理论,基于变换/基映射模型逼近深度学习模型,解析优化问题,成为当前研究的前沿... 详细信息

深度学习模型广泛应用于多媒体信号处理领域,通过引入非线性能够极大地提升性能,但是其黑箱结构无法解析地给出最优点和优化条件。因此如何利用传统信号处理理论,基于变换/基映射模型逼近深度学习模型,解析优化问题,成为当前研究的前沿问题。本文从信号处理的基础理论出发,分析了当前针对高维非线性非规则结构方法的数学模型和理论边界,主要包括:结构化稀疏表示模型、基于框架理论的深度网络模型、多层卷积稀疏编码模型以及图信号处理理论。详细描述了基于组稀疏性和层次化稀疏性的表示模型和优化方法,分析基于半离散框架和卷积稀疏编码构建深度/多层网络模型,进一步在非欧氏空间上扩展形成图信号处理模型,并对国内外关于记忆网络的研究进展进行了比较。最后,展望了多媒体信号处理的理论模型发展,认为图信号处理通过解析谱图模型的数学性质,解释其中的关联性,为建立广义的大规模非规则多媒体信号处理模型提供理论基础,是未来研究的重要领域之一。

关键词：结构化稀疏表示基于框架理论的深度卷积网络多层卷积稀疏编码图信号处理多媒体信号处理

Emulation of Brain Metabolic Activities Based on a Dynamically Controllable Optical Phantom

在线全文

学校读者我要写书评

暂无评论

Cyborg and Bionic Systems 2023年第1期4卷 84-92页

作者： Yuxiang lin Cheng Chen zhouchen Ma Nabil Sabor Yanyan Wei Tianhong Zhang Mohamad Sawan Guoxing Wang Jian Zhao Department of Micro-Nano Electronics Shanghai Jiao Tong UniversityShanghaiChina Electrical Engineering Department Faculty of EngineeringAssiut UniversityAssiutEgypt Shanghai Mental Health Center Shanghai Jiao Tong University School of MedicineShanghaiChina CenBRAIN Neurotech School of EngineeringWestlake UniversityHangzhouChina

This paper presents a dynamic optical phantom for the simulation of metabolic activities in the brain,and a linear equivalent model is built for control voltage versus substance concentration.A solid-solid dynamic optical phantom is realized by using liquid crystal film as a voltage-controlled light intensity regulator on the surface of basic phantom,which uses epoxy resin as matrix material and nanometer carbon powder and titanium dioxide powder as absorption and scattering dopants,*** dynamic phantom could mimic near-infrared spectrum(NIRS)signals with sampling rate up to 10 Hz,and the maximum simulation errors for oxy-hemoglobin and deoxy-hemoglobin concentrations varying in the range of 1μmol/l are 7.0%and 17.9%,*** with similar solid biomimetic phantoms,the adjustable mimic substance concentration range is extended by an order of magnitude,which meets the simulation requirements of most brain NIRS signals.

关键词： dynamic powder titanium