移动边缘计算辅助智能驾驶中基于高效联邦学习的碰撞预警算法

智能驾驶中的碰撞避免任务存在对时延要求极高和隐私保护等挑战。首先,该文提出一种基于自适应调整参数的半异步联邦学习(SFLAAP)的门控循环单元联合支持向量机(GRU_SVM)碰撞多级预警算法,SFLAAP可根据训练和资源情况动态调整两个训练参数：本地训练次数和参与聚合的局部模型数量。然后,为解决资源受限的移动边缘计算(MEC)下碰撞预警模型协作训练的效率问题,根据上述参数与SFLAAP训练时延的关系,建立训练总时延最小化模型,并将其转化为马尔可夫决策过程(MDP)。最后,在所建立的MDP中采用异步优势演员-评论家(A3C)学习求解,自适应地确定最优训练参数,从而减少碰撞预警模型的训练完成时间。仿真结果表明,所提算法有效地降低训练总时延并保证预测精度。     

有线电视网络的互动式点播教学环境

远程教学是构建在电信、计算机、传播科技3者之间的应用,是兼有视觉与听觉的双向学习方式.目前远程教学的发展大致分成实时群播、虚拟网站、课程点播3个系统,虽然已经有完整的系统出现,但均存在受时间限制、网络带宽不足、影片画质差、互动性不佳等缺点.这里提出交互式点播教学环境的概念,是利用有线电视点播视频系统以及有线电视网络控制器,提出一套交互式点播教学系统,以解决目前远程教学发展所遇到的瓶颈.     

Extending the participatory learning paradigm to include source credibility

We provide an overview of the participatory learning paradigm (PLP) and discuss the importance of the acceptance function in determining which observations are used for learning. We introduce a formal model that uses this (PLP) We then extend this model in two directions. First, we consider situations in which we have incomplete observations, we only have observations about a subset of the variables of interest. Next we extend this model to allow for the inclusion in the learning process of information about the learning agents belief about the credibility of the source of the learning experience. Here we distinguish between the content of a learning experience and the source of the experience. We provide a means to allow the learning agents belief about the credibility of the source to determine the effect of the content. Furthermore we suggest a method to allow the modification of agents belief about the credibility of the source to also be part of the learning process.     

Time inconsistency and learning in bargaining games

This paper studies an alternating-offers bargaining game between possibly time-inconsistent players. The time inconsistency is modeled by quasi-hyperbolic discounting and the “naive backwards induction” solution concept is used in order to obtain the results. Both naive agents who remain naive and those who learn about their own preferences are considered. Offers of the players who are naive or partially naive are never accepted by any type of player in either no learning or gradual learning cases. The game between a naive or partially naive player who never learns and a time-consistent agent ends in an immediate agreement if the time-consistent agent is the proposer. A one period delay occurs if the time-consistent agent is the responder. The more naive the player is, the higher the share received. In addition, two naive agents who never learn disagree perpetually. When naive and partially naive agents play against exponential or sophisticated agents and they are able to learn their types over time, there exists a critical date before which there is no agreement. Therefore, the existence of time-inconsistent players who can learn their types as they play the game can be a new explanation for delays in bargaining. The relationship among the degree of naivete, impatience level and bargaining delay is also characterized. Specifically, for sufficiently high discount factors, agreement is always delayed. On the other hand, if the naive agent has sufficiently firm initial beliefs (slow learning or high degree of naivete), agents agree immediately.     

示例学习的数据预处理

在各种学习方法中,示例式学习被看作是自动建立基于知识的系统的关键,数据预处理是示例学习的一个重要步骤。本文介绍了预处理的两个算法,解决了小内存处理大数据的问题,     

Fast competitive learning with classified learning rates for vector quantization

In this paper, we present a new competitive learning algorithm with classified learning rates, and apply it to vector quantization of images. The basic idea is to assign a distinct learning rate to each reference vector. Each reference vector is updated independently of all the other reference vectors using its own learning rate. Each learning rate is changed only when its corresponding reference vector wins the competition, and the learning rates of the losing reference vectors are not changed. The experimental results obtained with image vector quantization show that the proposed method learns more rapidly and yields better quality of the coded images than conventional competitive learning method with a scalar learning rate.     

一种用于PCA与MCA的神经网络学习算法

主元分析（ＰＣＡ）和次元分析（ＭＣＡ）是用于特征提取、数据压缩、频率估计、曲线拟合等信号处理的基本技术．以神经网络来实现ＰＣＡ和ＭＣＡ是当今研究的一大热点，相关矩阵Ｒ的特征值重数不为１时的主、次元分析则是其中一大难题．本文提出了一种新的学习算法，使得在输入数据的相关矩阵含多重特征值时，网络权重矢量亦收敛于相关矩阵的单位正交特征矢量     

Classroom mathematical practices and gesturing

The purpose of this paper is to illustrate a methodological approach for empirically investigating the function of gesturing in the collective development of knowledge. We extend the earlier work of Stephan and Rasmussen [Stephan, M., & Rasmussen, C. (2002). Classroom mathematical practices in differential equations. Journal of Mathematical Behavior 21, 459-490] who analyzed classroom discourse and symbolizing to document the emergence of six classroom mathematical practices over the course of 22 days of instruction on first-order differential equations. We complement and extend this previous analysis by re-examining the same data for gesturing and coordinate this analysis with the evolution of the classroom mathematical practices as they developed in this particular community of learners. Our illustration of the methodology we developed suggests that (1) gestures and argumentation can function as a unit that supports the establishment of one or more taken-as-shared ideas, and (2) that a gesture/argumentation pair that develops while establishing one practice can change function to support the establishment of ideas embedded in other classroom mathematical practices.     

Environmental noise classifier using a new set of feature parameters based on pitch range

Automatic Noise Recognition was performed in two stages: (1) feature extraction based on the pitch range, found by analyzing the autocorrelation function and (2) classification using a classifier trained on the extracted features. Since most environmental noise types change their acoustical characteristics over time, we focused on the “pitch range” of the sounds in order to extract features. Two different classifiers, Support Vector Machines (SVM) and k-means clustering, were performed and compared using the proposed features. The SVM and k-means clustering classifiers achieve recognition rates up to 95.4% and 92.8%, respectively. Although both classifiers provided high accuracy, the SVM-based classifier outperformed the k-means clustering classifier by approximately 7.4%.     

QoS commitment between vertically integrated autonomous systems

Vertically integrated autonomous systems bargain to provide quality of service guarantees and revenue sharing. Depending on the perceived quality of service and access price, consumers determine whether they subscribe to the access provider’s service. Four types of contracts are compared: (i) best effort, (ii) bilateral bargaining, (iii) cascade negotiations and (iv) grand coalition cooperation; the impact of the consumers’ QoS sensitivity parameter and power relation are tested for each contract. Assuming that the consumers’ quality of service sensitivity parameter is unknown and might evolve dynamically due to error judgement, word-of-mouth effect or competition pressure, a learning algorithm is detailed and implemented by each integrated autonomous systems under asymmetrical information. Its convergence and the influence of bias introduction by the most informed autonomous system is analyzed.