The problem on stabilization for the system with distributed delays is researched. The distributed time-delay under consideration is assumed to be a constant time-delay, but not known exactly. A design method is propo...The problem on stabilization for the system with distributed delays is researched. The distributed time-delay under consideration is assumed to be a constant time-delay, but not known exactly. A design method is proposed for a memory proportional and integral （PI） feedback controller with adaptation to distributed time-delay. The feedback controller with memory simultaneously contains the current state and the past distributed information of the addressed systems. The design for adaptation law to distributed delay is very concise. The controller can be derived by solving a set of linear matrix inequalities （LMIs）. Two numerical examples are given to illustrate the effectiveness of the design method.展开更多
In this paper,direct adaptive-state feedback control schemes are developed to solve the problem of asymptotic tracking and disturbance rejection for a class of distributed large-scale systems with faulty and perturbed...In this paper,direct adaptive-state feedback control schemes are developed to solve the problem of asymptotic tracking and disturbance rejection for a class of distributed large-scale systems with faulty and perturbed interconnection links.In terms of the special distributed architectures,the adaptation laws are proposed to update controller parameters on-line when all interconnected fault factors,the upper bounds of perturbations in interconnection links,and external disturbances on subsystems axe unknown.Then,a class of distributed state feedback controllers is constructed to automatically compensate the fault and perturbation effects,and reject the disturbances simultaneously based on the information from adaptive schemes.The proposed adaptive robust tracking controllers can guarantee that the resulting adaptive closed-loop distributed system is stable and each subsystem can asymptotic-output track the corresponding reference signal in the presence of faults and perturbations in interconnection links,and external disturbances.The proposed design technique is finally evaluated in the light of a simulation example.展开更多
A new adaptive neural network（NN） output-feedback stabilization controller is investigated for a class of uncertain stochastic nonlinear strict-feedback systems with discrete and distributed time-varying delays and ...A new adaptive neural network（NN） output-feedback stabilization controller is investigated for a class of uncertain stochastic nonlinear strict-feedback systems with discrete and distributed time-varying delays and unknown nonlinear functions in both drift and diffusion terms.First,an extensional stability notion and the related criterion are introduced.Then,a nonlinear observer to estimate the unmeasurable states is designed,and a systematic backstepping procedure to design an adaptive NN output-feedback controller is proposed such that the closed-loop system is stable in probability.The effectiveness of the proposed control scheme is demonstrated via a numerical example.展开更多
光栅结构的设计和制作直接决定了分布反馈（DFB）半导体激光器光电特性的优劣。采用传输矩阵法模拟了不同光栅耦合因子下随机相位对均匀光栅DFB芯片特性的影响,获得了芯片的光电参数分布。通过分析耦合因子对芯片光电参数分布的影响,提...光栅结构的设计和制作直接决定了分布反馈（DFB）半导体激光器光电特性的优劣。采用传输矩阵法模拟了不同光栅耦合因子下随机相位对均匀光栅DFB芯片特性的影响,获得了芯片的光电参数分布。通过分析耦合因子对芯片光电参数分布的影响,提高了DFB芯片的成品率。设计并制备了基于Al Ga In As材料体系的脊波导DFB激光器,最终使芯片双峰比例仅为7.7%、成品率达到60%。对合格品在-40~105℃下的P-I特性和在-40~85℃下的光谱进行了测试,结果表明芯片性能优良,芯片远场发散角为25°和21°。芯片的小信号频带响应和眼图测试结果表明芯片完全满足2.5 Gbit/s的应用要求。展开更多
基于人类反馈的强化学习(reinforcement learning with human feedback,RLHF)作为当前大语言模型(large language models,LLMs)对齐的主流方法,其核心优化算法——近端策略优化(proximal policy optimization,PPO)却面临着显著的效率问...基于人类反馈的强化学习(reinforcement learning with human feedback,RLHF)作为当前大语言模型(large language models,LLMs)对齐的主流方法,其核心优化算法——近端策略优化(proximal policy optimization,PPO)却面临着显著的效率问题.PPO由生成、推理、训练3个相互关联的阶段组成,各个阶段有着不同的计算特性.然而,现有的RLHF并行框架采用相同并行策略顺序执行PPO的所有阶段,这导致以下2个问题:其一,生成阶段不能充分利用计算资源,进而影响整体效率;其二,阶段间严格串行执行,未能充分利用潜在并行性.针对上述问题,提出了一个新型RLHF并行框架——Pipe-RLHF.该框架能够自适应地根据各阶段的计算特征确定最优并行策略,突破现有阶段串行范式,采用异步PPO算法发掘阶段间的并行性.具体而言,创新性地提出了适用于PPO生成阶段的延迟批间流水线并行方法,显著提升了该阶段的计算资源利用率;再次,使用异步PPO解放阶段间的依赖关系,将阶段间并行应用到PPO的加速上;最后,针对PPO算法的整体优化,构建了分层并行策略空间,并提出了一套优化算法以实现该空间中的最优解搜索.通过在多个大语言模型上的性能评估实验表明,相较于现有方法,Pipe-RLHF最高可实现3.7倍的加速比,充分验证了该框架的有效性和优越性.展开更多
成功制备出室温激射波长为2μm的Ga Sb基侧向耦合分布反馈量子阱激光器.采用全息曝光及电感耦合等离子体刻蚀技术制备二阶布拉格光栅.优化了光栅制备的刻蚀条件,并获得室温2μm单纵模激射.激光器输出光功率超过5 m W,最大边模抑制比达到...成功制备出室温激射波长为2μm的Ga Sb基侧向耦合分布反馈量子阱激光器.采用全息曝光及电感耦合等离子体刻蚀技术制备二阶布拉格光栅.优化了光栅制备的刻蚀条件,并获得室温2μm单纵模激射.激光器输出光功率超过5 m W,最大边模抑制比达到24 d B.展开更多
基金supported by the National Natural Science Foundation of China (60804017 60835001+3 种基金 60904020 60974120)the Foundation of Doctor (20070286039 20070286001)
文摘The problem on stabilization for the system with distributed delays is researched. The distributed time-delay under consideration is assumed to be a constant time-delay, but not known exactly. A design method is proposed for a memory proportional and integral （PI） feedback controller with adaptation to distributed time-delay. The feedback controller with memory simultaneously contains the current state and the past distributed information of the addressed systems. The design for adaptation law to distributed delay is very concise. The controller can be derived by solving a set of linear matrix inequalities （LMIs）. Two numerical examples are given to illustrate the effectiveness of the design method.
基金Supported by National Basic Research Program of China(973 Program)(2009CB320604)the Key Program of National Natural Science Foundation of China(60534010)+4 种基金National Natural Science Foundation of China(60674021),Program for New Century Excellent Talents in Universities(NCET-04-0283)the Funds for Cre-ative Research Groups of China(60821063)Program for Changjiang Scholars and Innovative Research Team in University(IRT0421)the Funds of Doctoral Program of Ministry of Education,China(20060145019)the 111 Project(B08015)
文摘In this paper,direct adaptive-state feedback control schemes are developed to solve the problem of asymptotic tracking and disturbance rejection for a class of distributed large-scale systems with faulty and perturbed interconnection links.In terms of the special distributed architectures,the adaptation laws are proposed to update controller parameters on-line when all interconnected fault factors,the upper bounds of perturbations in interconnection links,and external disturbances on subsystems axe unknown.Then,a class of distributed state feedback controllers is constructed to automatically compensate the fault and perturbation effects,and reject the disturbances simultaneously based on the information from adaptive schemes.The proposed adaptive robust tracking controllers can guarantee that the resulting adaptive closed-loop distributed system is stable and each subsystem can asymptotic-output track the corresponding reference signal in the presence of faults and perturbations in interconnection links,and external disturbances.The proposed design technique is finally evaluated in the light of a simulation example.
基金supported by the National Natural Science Fundation of China (6080402160974139+3 种基金61075117)the Fundamental Research Funds for the Central Universities (JY10000970001K5051070000272103676)
文摘A new adaptive neural network（NN） output-feedback stabilization controller is investigated for a class of uncertain stochastic nonlinear strict-feedback systems with discrete and distributed time-varying delays and unknown nonlinear functions in both drift and diffusion terms.First,an extensional stability notion and the related criterion are introduced.Then,a nonlinear observer to estimate the unmeasurable states is designed,and a systematic backstepping procedure to design an adaptive NN output-feedback controller is proposed such that the closed-loop system is stable in probability.The effectiveness of the proposed control scheme is demonstrated via a numerical example.
文摘光栅结构的设计和制作直接决定了分布反馈（DFB）半导体激光器光电特性的优劣。采用传输矩阵法模拟了不同光栅耦合因子下随机相位对均匀光栅DFB芯片特性的影响,获得了芯片的光电参数分布。通过分析耦合因子对芯片光电参数分布的影响,提高了DFB芯片的成品率。设计并制备了基于Al Ga In As材料体系的脊波导DFB激光器,最终使芯片双峰比例仅为7.7%、成品率达到60%。对合格品在-40~105℃下的P-I特性和在-40~85℃下的光谱进行了测试,结果表明芯片性能优良,芯片远场发散角为25°和21°。芯片的小信号频带响应和眼图测试结果表明芯片完全满足2.5 Gbit/s的应用要求。
文摘基于人类反馈的强化学习(reinforcement learning with human feedback,RLHF)作为当前大语言模型(large language models,LLMs)对齐的主流方法,其核心优化算法——近端策略优化(proximal policy optimization,PPO)却面临着显著的效率问题.PPO由生成、推理、训练3个相互关联的阶段组成,各个阶段有着不同的计算特性.然而,现有的RLHF并行框架采用相同并行策略顺序执行PPO的所有阶段,这导致以下2个问题:其一,生成阶段不能充分利用计算资源,进而影响整体效率;其二,阶段间严格串行执行,未能充分利用潜在并行性.针对上述问题,提出了一个新型RLHF并行框架——Pipe-RLHF.该框架能够自适应地根据各阶段的计算特征确定最优并行策略,突破现有阶段串行范式,采用异步PPO算法发掘阶段间的并行性.具体而言,创新性地提出了适用于PPO生成阶段的延迟批间流水线并行方法,显著提升了该阶段的计算资源利用率;再次,使用异步PPO解放阶段间的依赖关系,将阶段间并行应用到PPO的加速上;最后,针对PPO算法的整体优化,构建了分层并行策略空间,并提出了一套优化算法以实现该空间中的最优解搜索.通过在多个大语言模型上的性能评估实验表明,相较于现有方法,Pipe-RLHF最高可实现3.7倍的加速比,充分验证了该框架的有效性和优越性.
