学在北邮

/ Study in BUPT

Enabling Simulation-Based Optimization Through Machine Learning

主讲人 :Michele Zorzi教授(IEEE Fellow) 地点 :教三楼3-305教室 开始时间 : 2019-09-12 10:00 结束时间 : 2019-09-12 12:00

Abstract: Complex phenomena are generally modeled with sophisticated simulators that, depending on their accuracy, can be very demanding in terms of computational resources and simulation time. Their time-consuming nature, together with a typically vast parameter space to be explored, make simulation-based optimization often infeasible. In this work, we present a method that enables the optimization of complex systems through Machine Learning (ML) techniques. We show how well-known learning algorithms are able to reliably emulate a complex simulator with a modest dataset obtained from it. The trained emulator is then able to yield values close to the simulated ones in virtually no time. Therefore, it is possible to perform a global numerical optimization over the vast multi-dimensional parameter space, in a fraction of the time that would be required by a simple brute-force search. As a testbed for the proposed methodology, we used a network simulator for next-generation mmWave cellular systems. After simulating several antenna configurations and collecting the resulting network-level statistics, we feed it into our framework. Results show that, even with few data points, extrapolating a continuous model makes it possible to estimate the global optimum configuration almost instantaneously. The very same tool can then be used to achieve any further optimization goal on the same input parameters in negligible time.

该讲座为前沿讲座,欢迎全校师生踊跃参加。

 

 

校学术委员会

信息与通信工程学院

信息网络体系构建与融合学科创新引智基地

                               2019912

 

分享到

XML 地图 | Sitemap 地图