This website requires JavaScript.

Learning from Heterogeneous Data Based on Social Interactions over Graphs

Virginia BordignonStefan VlaskiVincenzo MattaAli H. Sayed
摘要
This work proposes a decentralized architecture, where individual agents aimat solving a classification problem while observing streaming features ofdifferent dimensions and arising from possibly different distributions. In thecontext of social learning, several useful strategies have been developed,which solve decision making problems through local cooperation acrossdistributed agents and allow them to learn from streaming data. However,traditional social learning strategies rely on the fundamental assumption thateach agent has significant prior knowledge of the underlying distribution ofthe observations. In this work we overcome this issue by introducing a machinelearning framework that exploits social interactions over a graph, leading to afully data-driven solution to the distributed classification problem. In theproposed social machine learning (SML) strategy, two phases are present: in thetraining phase, classifiers are independently trained to generate a belief overa set of hypotheses using a finite number of training samples; in theprediction phase, classifiers evaluate streaming unlabeled observations andshare their instantaneous beliefs with neighboring classifiers. We show thatthe SML strategy enables the agents to learn consistently under thishighly-heterogeneous setting and allows the network to continue learning evenduring the prediction phase when it is deciding on unlabeled samples. Theprediction decisions are used to continually improve performance thereafter ina manner that is markedly different from most existing static classificationschemes where, following training, the decisions on unlabeled data are notre-used to improve future performance.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答