The K.U. Leuven competition data a challenge for advanced neural network techniques.pdf

下载文档

6
0
约1.12万字
约 6页
2017-04-06 发布于江苏
举报
版权申诉
保障服务

The K.U. Leuven competition data a challenge for advanced neural network techniques.pdf

1、本文档共6页，可阅读全部内容。
2、原创力文档（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。
3、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。

The K.U. Leuven competition data a challenge for advanced neural network techniques

The K.U. Leuven CompetitionData: a Challenge for AdvancedNeural Network TechniquesJ.A.K. Suykens and J. VandewalleKatholieke Universiteit Leuven, Dept. Electr. Eng. ESAT/SISTAKardinaal Mercierlaan 94, B-3001 Heverlee, BelgiumEmail: johan.suykens@esat.kuleuven.ac.beAbstract. In this paper we shortly discuss the K.U. Leuven time-seriesprediction competition, which has been held in the framework of the In-ternational Workshop on Advanced Black-Box Techniques for NonlinearModeling, K.U.Leuven Belgium July 8-10 1998. The data are related toa 5-scroll attractor, generated from a generalized Chuas circuit. Thetime-series consists of a given set of 2000 data points, where the next200 points are to be predicted. In total, 17 entries have been submitted.The winning contribution by McNames succeeds in making an accurateprediction over a time horizon of about 300 points using a nearest tra-jectory method, which incorporates local modeling and cross-validationtechniques. The competition data can serve as a challenging test-casefor advanced nonlinear modelling techniques, including neural networks.The data are able to reveal shortcomings of many methods.Keywords. Time-series prediction, generalized Chuas circuit, chaos.1. IntroductionThe prediction of time-series is a challenging area for nonlinear modeling tech-niques. This has been demonstrated earlier by the Santa Fe time series pre-diction competition [17], where several data sets such as laser data, currencyexchange rates and astrophysical data were to be predicted. Despite the factthat chaotic systems have a limited predictability, related to the rst Lyapunovexponent of the system, it has been shown that accurate short term forecastsare feasible. In this context the choice of the model structure and embedding,This research work was carried out at the ESAT laboratory and the InterdisciplinaryCenter of Neural Networks ICNN of the Katholieke Universiteit Leuven, in the framework ofthe FWO project G.0262.97 Learning and Optimi