In this study, we propose a novel heuristic two-step algorithm for shared ridehailing in which users can share their rides with only one more user. The algorithm, which is centrally formulated, starts with matching users and creating a set of passenger pairs in step 1 and is followed by solving an assignment problem to assign passenger pairs to the vehicles. To solve the problem of high computational time in dynamic ride-matching problems, we propose a distributed system that is based on vehicle to infrastructure (V2I) and infrastructure to infrastructure (I2I) communication. To evaluate the distributed system's performance, we compare it with the proposed centralized ridehailing algorithm. Both centralized and distributed systems are implemented in a micro-traffic simulator to assess their performance and their impact on traffic congestion. Downtown Toronto road network was chosen as the study area. Based on our obtained results, the service rate of the distributed system was 91.59% which is close to 95.80% in the centralized system. However, the distributed system yielded much lower computational time compared to centralized. Furthermore, the scalability of the distributed system was shown by testing it on a small network and comparing with the entire network.
Towards More Suitable Personalization in Federated Learning via Decentralized Partial Model Training
Tsinghua University（清华大学）；Nanjing University of Science and Technology（南京理工大学）；The University of Sydney（悉尼大学）；Virginia Tech（弗吉尼亚理工大学）；JD Explore Academy（京东探索研究院）
Decentralized learning with budgeted network load using Gaussian copulas and classifier ensembles
Univ. Lille, CNRS, Centrale Lille, UMR 9189 - CRIStAL - Centre de Recherche en Informatique Signal et Automatique de Lille（里尔大学、法国国家科学研究中心、里尔中央理工学院、UMR 9189 - CRIStAL - 里尔信息学信号与自动化研究中心）；Inria（法国国家信息与自动化研究所）；University College London（伦敦大学学院）
我们研究了一个学习者网络，这些学习者处理相同的分类任务，但必须从不同的数据集学习。学习者不能共享数据，而是共享他们的模型。模型只共享一次，以保持网络负载。我们介绍DELCO（Decentralized Ensemble Learning with COpulas），这是一种新方法，可以聚合每个学习者训练的分类器的预测。所提出的方法聚合的基础分类器使用的概率模型依赖于高斯copula。逻辑回归集成的实验证明竞争的准确性和增加的鲁棒性的情况下，依赖分类器。可以从https://github.com/john-klein/DELCO下载配套的python实现
