Tutorial at IEEE World Congress on Computational Intelligence (IEEE WCCI) 2016
Learning in non-stationary environments
Many real-world machine learning applications assume the stationarity hypothesis for the process generating the data. This assumption guarantees that the model learnt during the initial training phase remains valid over time and that its performance is in line with our expectations. Unfortunately, this assumption does not truly hold in the real world, in many cases representing only a simplistic approximation.
Current research in machine learning aims at removing/weakening the stationary assumption so that time variance is detected as soon as possible and suitable actions activated afterwards. In this direction, the literature addressing the learning in nonstationary environments classifies existing approaches as passive or active depending on the learning mechanism adopted to deal with the process evolution. Passive approaches rely on a continuous adaptation of the application without explicitly knowing whether a change has occurred or not, while, in active approaches, triggering mechanisms, e.g., Change Detection Tests (CDTs) or Change Point Methods (CPMs), are considered to detect a change in the process generating the data. Once the change has been detected the application might require (self) adaptation to track the system evolution.
The tutorial will introduce and contrast passive and active approaches by providing those details the scholar and the practitioner need to be able to design machine learning applications working in nonstationary environments.
MANUEL ROVERI received the Dr.Eng. degree in Computer Science Engineering from the Politecnico di Milano (Milano, Italy) in June 2003, the MS in Computer Science from the University of Illinois at Chicago (Chicago, Illinois, U.S.A.) in December 2003 and the Ph.D. degree in Computer Engineering from Politecnico di Milano (Milano, Italy) in May 2007. Currently, he is an associate professor at the Department of Electronics and Information of the Politecnico di Milano.
He has been visiting researcher at Imperial College London (UK). Manuel Roveri is an Associate Editor of the IEEE Transactions on Neural Networks and Learning Systems and served as chair and member in many IEEE subcommittees. He received the 2016 IEEE CIS Outstanding Transaction on Neural networks and Learning Systems award. He is the co-organizer of the IEEE Symposium on Intelligent Embedded Systems in 2014 and organizer and co-organizer of workshops and special sessions at IEEE-sponsored conferences. Current research activity addresses adaptation and learning in non-stationary environments and intelligence for embedded systems and cognitive fault diagnosis.
CESARE ALIPPI received the degree in electronic engineering cum laude in 1990 and the PhD in 1995 from Politecnico di Milano, Italy. Currently, he is a Full Professor of information processing systems with the Politecnico di Milano. He has been a visiting researcher at UCL (UK), MIT (USA), ESPCI (F), CASIA (RC), USI (CH).
Alippi is an IEEE Fellow, Vice-President education of the IEEE Computational Intelligence Society (CIS), Board of Governors member of the International Neural Networks Society, Associate editor (AE) of the IEEE Computational Intelligence Magazine, past AE of the IEEE-Tran. Neural Networks (2005-2012), IEEE-Trans Instrumentation and Measurements (2003-09) and member and chair of other IEEE committees. He was awarded the 2016 IEEE CIS Outstanding Transaction on Neural networks and Learning Systems award, the 2013 IBM Faculty award; the 2004 IEEE Instrumentation and Measurement Society Young Engineer Award; in 2011 has been awarded Knight of the Order of Merit of the Italian Republic. Current research activity addresses adaptation and learning in non-stationary environments and Intelligence for embedded systems.