基于Actor模型的众核数据流硬件架构探索

超大规模AI模型的分布式训练对芯片架构的通信能力和可扩展性提出了挑战。晶圆级芯片通过在同一片晶圆上集成大量的计算核心和互联网络,实现了超高的计算密度和通信性能,成为了训练超大规模AI模型的理想选择。AMCoDA是一种基于Actor模型的众核数据流硬件架构,旨在利用Actor并行编程模型的高度并行性、异步消息传递和高扩展性等特点,在晶圆级芯片上实现AI模型的分布式训练。AMCoDA的设计包括计算模型、执行模型和硬件架构3个层面。实验表明,AMCoDA能广泛支持分布式训练中的各种并行模式和集合通信模式,灵活高效地完成复杂分布式训练策略的部署和执行。

Actor-Critic-Based UAV-Assisted Data Collection in the Wireless Sensor Network

Wireless Sensor Network(WSN)is widely utilized in large-scale distributed unmanned detection scenarios due to its low cost and flexible installation.However,WSN data collection encounters challenges in scenarios lacking communication infrastructure.Unmanned aerial vehicle(UAV)offers a novel solution for WSN data collection,leveraging their high mobility.In this paper,we present an efficient UAV-assisted data collection algorithm aimed at minimizing the overall power consumption of the WSN.Firstly,a two-layer UAV-assisted data collection model is introduced,including the ground and aerial layers.The ground layer senses the environmental data by the cluster members(CMs),and the CMs transmit the data to the cluster heads(CHs),which forward the collected data to the UAVs.The aerial network layer consists of multiple UAVs that collect,store,and forward data from the CHs to the data center for analysis.Secondly,an improved clustering algorithm based on K-Means++is proposed to optimize the number and locations of CHs.Moreover,an Actor-Critic based algorithm is introduced to optimize the UAV deployment and the association with CHs.Finally,simulation results verify the effectiveness of the proposed algorithms.

New Rural Community Construction or Retention Development:A Comparative Analysis of Rural Settlement Transition Mechanism in Plain Agriculture Area of China Based on Actor Network Theory

It is an important way to realize rural revitalization and sustainable development to guide rural settlement transition(RST)in an appropriate way.This paper uses actor network theory(ANT)to construct a theoretical framework for the study of RST.Taking two typical villages with different transition paths in rural areas of North China Plain as examples,this paper reveals the mechanism of RST and makes a comparative analysis.The results show that:1)after identifying problems and obligatory passage point,key actors recruit heterogeneous actors into the actor network by entrusting them with common interests,and realize RST under the system operation.2)Rural settlements under different transition paths have similarities in the problems to be solved,collective actions and policy factors,but there are differences in the transition process,mechanism and effect.The actor network and mechanism of RST through the path of new rural community construction are more complex and the transition effect is more thorough.In contrast,the degree of RST of retention development path is limited if there is no resource and location advantage.3)Based on the applicable conditions of different paths,this paper designs a logical framework of‘Situation-Structure-Behavior-Result’to scientifically guide the identification of RST paths under the background of rural revitalization.

融合Dead-ends和离线监督Actor-Critic的动态治疗策略生成模型

强化学习对数学模型依赖性低,利用经验便于架构和优化模型,非常适合用于动态治疗策略学习。但现有研究仍存在以下问题:1)学习策略最优性的同时未考虑风险,导致学到的策略存在一定的风险;2)忽略了分布偏移问题,导致学到的策略与医生策略完全不同;3)忽略患者的历史观测数据和治疗史,从而不能很好地得到患者状态,进而导致不能学到最优策略。基于此,提出了融合Dead-ends和离线监督Actor-Critic的动态治疗策略生成模型DOSAC-DTR。首先,考虑学到的策略所推荐的治疗行动的风险性,在Actor-Critic框架中融入Dead-ends概念;其次,为缓解分布偏移问题,在Actor-Critic框架中融入医生监督,在最大化预期回报的同时,最小化所学策略与医生策略之间的差距;最后,为了得到包含患者关键历史信息的状态表示,使用基于LSTM的编码器解码器模型对患者的历史观测数据和治疗史进行建模。实验结果表明,DOSAC-DTR相比基线方法有更好的性能,可以得到更低的估计死亡率以及更高的Jaccard系数。

无人机辅助物联网中基于Safe Actor-Critic的信息年龄最小化研究

无人机作为一种新的通信设备,有望在物联网数据采集、监控等业务中发挥关键作用。为保证所采集数据的时效性,利用信息年龄来衡量无人机从物联网设备接收到的数据新鲜度。通过联合优化无人机轨迹和无人机与物联网设备的关联策略以最小化信息年龄加权和,并保证无人机累积飞行能量消耗满足预算要求。由于上述问题同时受短期和长期约束条件的限制,将问题建模为受约束的马尔可夫决策过程(CMDP),并利用Safe Actor-Critic来求解。仿真结果表明,所提算法在最小化信息年龄的同时,能有效保证能量预算。

Felski’s Literary Sociological Thought From the Perspective of Actor-Network Theory

The relationship between literature and society has been a subject of continuous exploration since the inception of literature itself.On the one hand,from Plato’s theory of mimesis onward,literature has consistently been viewed as a representation of social reality,positioning literature as subordinate to society.On the other hand,with the rise of structuralism and the New Criticism,certain schools of thought have focused exclusively on literature itself,deliberately overlooking the complex connections between literature and society.The growing tension between these two perspectives has increasingly placed contemporary literary studies in a polarized state,leading to a crisis in the legitimacy of literary scholarship.In response to this,Rita Felski’s exploration of the uses of literature embodies a new literary sociology that offers a way out of the current impasse in literary studies.

基于不确定性估计的离线确定型Actor-Critic

Actor-Critic是一种强化学习方法,通过与环境在线试错交互收集样本来学习策略,是求解序贯感知决策问题的有效手段.但是,这种在线交互的主动学习范式在一些复杂真实环境中收集样本时会带来成本和安全问题离线强化学习作为一种基于数据驱动的强化学习范式,强调从静态样本数据集中学习策略,与环境无探索交互,为机器人、自动驾驶、健康护理等真实世界部署应用提供了可行的解决方案,是近年来的研究热点.目前,离线强化学习方法存在学习策略和行为策略之间的分布偏移挑战,针对这个挑战,通常采用策略约束或值函数正则化来限制访问数据集分布之外(Out-Of-Distribution,OOD)的动作,从而导致学习性能过于保守,阻碍了值函数网络的泛化和学习策略的性能提升.为此,本文利用不确定性估计和OOD采样来平衡值函数学习的泛化性和保守性,提出一种基于不确定性估计的离线确定型Actor-Critic方法(Offline Deterministic Actor-Critic based on UncertaintyEstimation,ODACUE).首先,针对确定型策略,给出一种Q值函数的不确定性估计算子定义,理论证明了该算子学到的Q值函数是最优Q值函数的一种悲观估计.然后,将不确定性估计算子应用于确定型Actor-Critic框架中,通过对不确定性估计算子进行凸组合构造Critic学习的目标函数.最后,D4RL基准数据集任务上的实验结果表明:相较于对比算法,ODACUE在11个不同质量等级数据集任务中的总体性能提升最低达9.56%,最高达64.92%.此外,参数分析和消融实验进一步验证了ODACUE的稳定性和泛化能力.

Actor-critic框架下的二次指派问题求解方法

二次指派问题(QAP)属于NP-hard组合优化问题,在现实生活中有着广泛应用。目前相对成熟的启发式算法通常以问题为导向来设计定制化算法,缺乏迁移泛化能力。为提供一个统一的QAP求解策略,将QAP问题的流量矩阵及距离矩阵抽象成两个无向完全图并构造相应的关联图,从而将设施和地点的指派任务转化为关联图上的节点选择任务,基于actor-critic框架,提出一种全新的求解算法ACQAP。首先,利用多头注意力机制构造策略网络,处理来自图卷积神经网络的节点表征向量;然后,通过actor-critic算法预测每个节点被作为最优节点输出的概率;最后,依据该概率在可行时间内输出满足目标奖励函数的动作决策序列。该算法摆脱人工设计,且适用于不同规模的输入,更加灵活可靠。实验结果表明,在QAPLIB实例上,本算法在精度媲美传统启发式算法的前提下,迁移泛化能力更强;同时相对于NGM等基于学习的算法,求解的指派费用与最优解之间的偏差最小,且在大部分实例中,偏差均小于20%。

基于Actor-Critic自适应PID的钢筋套丝头跟踪检测控制系统研究

为适应流水线节奏,不影响生产进程,从而更好地实现钢筋套丝头质量检测和尺寸测量的自动化与智能化,设计了基于同步带直线导轨的钢筋套丝头检测跟踪系统,并提出了一种基于Actor-Critic的自适应PID控制方法,用强化学习的方法根据环境反馈自动调节PID控制器的比例、积分、微分参数。对该方法和其他PID控制方法的响应性能指标进行实验和分析,实验结果表明该方法能够实现高精度、快速响应的跟踪拍摄,保证高精度的套丝头质量检测。

Actors：实现并发面向对象程序设计的模型

Ａｃｔｏｒｓ是一些计算单位。这些计算单位为了回答不断增长的通信而进行各自的工作。原始的Ａｃｔｏｒｓ语言语义基于纯消息传递。文中建立了Ａｃｔｏｒｓ模型的基本框架，并且讨论了利用Ａｃｔｏｒｓ语言开发大型并发系统的优点。还讨论了Ａｃｔｏｒｓ语言抽象结构的数学表示。

基于改进Actor-Critic算法的多传感器交叉提示技术

针对在减少战场资源浪费、平衡战场效费比的同时提高目标探测概率,保证目标的可持续跟踪,提出利用改进Actor-Critic算法的多传感器交叉提示技术进行目标探测。首先,综合传感器探测、能耗、时效等因素搭建基于“交叉提示”传感器的动态管理评估模型;其次,重点分析利用Actor-Critic交叉提示算法的传感器管理决策规则,并且提出了Actor-Critic算法,以根据任务自身需求组建中央评价网络,加大传感器与外部环境的交互。仿真结果表明,改进的算法可以加速网络收益,实现对目标的持续性探测,加强传感器之间的交叉提示功能,提升调度的智能化水平,具有较大的应用价值。

Analysis of Hypertension Status and Its Influencing Factors in Some Areas of Hubei Province, China

Objective: To know about the current situation of hypertension in some areas of Hubei Province and analyze the influencing factors. Methods: According to the principle of Stratified sampling, we conducted health examination and questionnaire survey for 1500 residents over the age of 18 at observation point of chronic noninfectious diseases at 10 sub-districts (towns) of Wuhan City, Jingzhou City, Huanggang City, Shiyan City, which used descriptive statistics and logistics to regressively analyze Current situation of hypertension and its influencing factors of residents. Results: Prevalence of hypertension of rural and urban residents over 18 is 27.44% in Hubei province. There are many differences among prevalence of hypertension of male and female, distribution of BMI of rural and urban residents and prevalence of hypertension and so on, and it has statistical significance (P-value < 0.05 averagely);logistic regressive analysis result shows that different age, gender, education level, dieting habits (high salt and high oil), family per capita monthly income, BMI have statistical significance on the prevalence of hypertension in urban and rural residents of Hubei Province. Conclusions: The prevalence of hypertension in the residents of five cities and prefectures in Hubei Province is on the trend of rising in ladder form and at a much earlier age. The health education, monitoring and intervention of chronic diseases need to be widely carried out, with emphasis on the intervention of the residents’ eating habits of high salt and oil, smoking, drinking and other bad lifestyle.

Advocacy Coalition by External Actors and Strategies Used to Influence the Emergence of the National Nutrition Policy in Lao PDR

This article aims to explore the coalition of external actors and the strategies it deployed to influence the emergence of the National Nutrition Policy (NNP) in Lao People’s Democratic Republic (Lao PDR). The Advocacy Coalition Framework and the conceptual model of Effective Advocacy Strategies for Influencing Government Nutrition Policy were used to frame the data collection and their analysis. Sources of information were semi-structured interviews conducted with government and external actors, as well as all available documents on nutrition policy in Laos. The commitment of the government to achieve the Millennium Development Goals (MDGs) and to leave the Least Developed Country status created a favorable condition to support the emergence of the NNP in Laos. This context was a driving force for the building of an effective and convincing coalition of United Nations agencies able to accompany the government in redefining health priorities. Various strategies were used by the coalition to this end, including generating, disseminating, and using scientific evidence, assisting the government with a budget and technical expertise, providing decision-makers with opportunities to learn from other countries, and building relationships with the key actor. External actors can be a major force to support the emergence of a public policy in Laos, but this requires a window of opportunity like what the MDGs have been able to bring.

Global climate governance in the new era:Potential of business actors and technological innovation

The United Nations Framework Convention on Climate Change(UNFCCC)has established a climate governance mechanism with intergovernmental negotiations among sovereign states as the core.After nearly 30 years,progress in combating climate change has remained very modest compared with the numerous challenges raised.The global climate governance has entered a new era,such that incorporating other factors into the governance process is timely.Therefore,the study emphasizes technological innovation and business actors in climate governance after the Paris Agreement.Technological innovation can provide effective solutions for combating climate change and has been a crucial driving force in climate governance's evolution.Business actors are significant because they are actual implementers of technological innovation and can apply different types of power and influence on climate governance processes at various levels.In summary,business actors,as well as technological innovation in line with governments and the UNFCCC governance frameworks,create a new potential for climate governance in the new era.

The Role and Impact of Nonstate Actors from the Perspective of the Russia-Ukraine Crisis

TheRussia-Ukrainecrisiswas araremilitary crisis between European powers after the end of the Cold War,which led to confrontation between Western countries and Russia.Such a crisis also triggered the extensive involvement of nonstate actors,such as international capital giants,small and micro technology enterprises,emerging international platforms,and multinational corporations.Nonstate actors play an important role in international politics.Benefiting from the development of globalization and the accelerated evolution of the digital age,nonstate actors have exerted a significant influence on the international pattern by intervening in the competition between countries,participating in the process of global governance,and influencing the thoughts of science and technology.These actors have become an important variable affecting the world.The rising role of nonstate actors has many implications for China to expand its strength and safeguard national interests and security.

Actors’ Power, Livelihood Assets and Participatory Forestry in Bangladesh: Evidence from the Sal Forests Area

Decentralize in forest management policies towards more people-oriented approaches has become major policy trends in many of the world’s developing countries during the last decade. However, the power of important actors to misuse the participatory forestry (PF) approach for their self-interests has been stated as a main problem to success. So, this study attempted to identify the most powerful actors and the extent at which they influence the outcomes of PF, and also to measure the impact of PF on livelihood assets of participants. The empirical data were collected from the well-established PF programs at Madhupur Sal forests area of Bangladesh. The results showed that the forest department (FD) proved itself as the most powerful and influential actor in every element of power analysis in PF. Regarding to livelihood analysis, the results revealed that the overall value of participants’ livelihood assets was 0.85 and it was significantly differed from the non-participants value of 0.66. However, the development of human, physical and financial assets was not indicating a decent improvement like as natural and social assets of participants. So, it is necessary to pay more attention to boost up participants’ human and financial assets through intensive training and adopting proper tree-crop production techniques, and also ensuring alternative livelihoods approaches to the local people. In addition, the local government will need to pay more emphasis on constructing village roads and infrastructure so as to enhance physical assets of the local people. Finally, the study would recommend promoting PF with apposite government facilities and also empowering local participants in order to balance the power among different actors, and this will facilitate the participants in governing all of their development activities efficiently.

Analysis of Actors Involved in the Bamboo Value Chain and Strategies for the Development of the Bamboo Sector in the Menoua Division, West Region of Cameroon

Bamboo is an important non-timber forest product owing to its multipurpose nature. In Cameroon, bamboo has always been neglected and seen as worthless by many communities. However, in recent years, bamboo has received renewed attention which has made many communities and different stakeholders to gain interest in the resource. This study which was carried out in the Menoua division, West region of Cameroon between February to July 2021 sought to identify and characterise the actors involved in the bamboo sector, and assess the different strategies that could be employed to develop the bamboo sector. Data were collected using quantitative and qualitative methods in 6 sub-divisions in Menoua division (Dschang, Santchou, Fongo Tongo, Fokoué, Penka Michel and Nkong-Ni). For the selection of study sites, the criterion used was the proximity of households to the area where the bamboo resource is found. Using the simple random sampling method, 53 households were surveyed across the Menoua division. Key informant and expert interviews were also conducted with mayors, traditional chiefs, heads of forestry posts, heads of research institutions/structures and craftsmen involved in the bamboo sector. Using SPHINX software version 5, a survey form was designed. The data collected was coded and analysed using EXCEL 2010 and SPSS. Map data was analysed using ARCGIS version 2.18. Results indicated that there were two main groups of actors: direct actors (collectors, collectors/transporters, producers/collectors, producers/collectors/transporters, collectors/traders) and indirect actors (Municipal Councils, Decentralized Services of the Ministry of Forestry and Wildlife—MINFOF, Development partners like INBAR, Research Institutions like the Institute of Agricultural Research for Development—IRAD, and academic institutions like the University of Dschang). The direct actors are directly linked to the bamboo value chain while the indirect actors are those whose decisions influence the sector (e.g. MINFOF) or those who provide financial and technical support (Municipal Councils, INBAR;IRAD, the University of Dschang). The main strategies proposed by the local population for the development of the bamboo sector were regular sensitization campaigns on the benefits of bamboo (92%) and the provision of technical, material and financial support to bamboo producers (41%). Key informants/experts proposed the following in order to ensure the development of the bamboo sector in the Menoua division: raising awareness about bamboo, its different varieties and benefits;creating bamboo plantations (with varieties adapted to the agro-ecological zone i.e. the western highlands) in order to reduce the pressure on other resources;setting up support mechanisms for producers and other actors in the bamboo value chain;allocating land/agricultural areas for bamboo plantations as there is land scarcity in the Menoua division. Based on the strategic framework developed from this study, in order to ensure an adequate and effective development of the bamboo sector in the Menoua division, there should be among others: multiplication of awareness-raising and training programmes for farmers on bamboo production techniques;more support for smallholder farmers by providing them bamboo plants in quality and quantity;production of bamboo stems in quality and quantity;more awareness campaigns for young craftsmen on the advantages of the bamboo craft sector;more training campaigns for craftsmen on modern bamboo processing techniques;and the establishment of a well-developed and sustainable bamboo-based craft sector.

广义行为正则化离线Actor-Critic

行为正则化Actor-Critic(BRAC)是一种离线强化学习算法,通过将当前策略与行为策略之间的Kullback-Leibler(KL)散度作为策略目标函数的正则化项来缓解分布偏移问题.但是,由于KL散度是一种无界的分布差异度量,在策略差异过大时,策略目标函数中的累积期望回报项将仅对策略改进发挥有限的作用,从而导致最终学到的策略性能较差.针对该问题,将当前策略与行为策略之间的斜对称Jensen-Shannon(JS)散度作为策略目标函数的正则化项,提出了一种广义行为正则化离线Actor-Critic(GOACBR)算法.理论分析表明:由于斜对称JS散度有界,将其作为正则化项有助于降低策略性能差异.进一步,针对行为策略未知导致难以直接计算当前策略和行为策略间斜对称JS散度的问题,设计了一个辅助网络来对其进行间接估计.最后,给出了GOACBR的收敛性理论证明.在D4RL基准数据集上的评估结果表明:相较于BRAC,GOACBR在所有测试任务上获得的平均累积回报总和提升了289.8%.相关代码公布在https://github.com/houge1996/GOAC.

打造产教融合新空间:市域产教联合体建设的理论基础、原则与优化路径

创新搭建产教融合平台载体,形成地方政府、产业园区、高校、企业、科研院所等多元行动主体创新互动、人才共培、成果同享的共生空间,完善校政企协同“以教促产、以产助教”机制,是深化现代职业教育体系改革的重要举措,更是创新驱动发展战略和区域协调发展战略对职业教育提出的新要求和新方向。市域产教联合体应以产业园区为基础,以价值共创、多元共生、文化共塑为原则,形成主体共识、实现多元治理、推动联合协同,打造增强职业教育服务区域高质量发展的多主体共同成长、边界突破、场景驱动的异质行动者产教融合新空间。