SOME OPEN AREAS IN SYSTEMIDENTIFICATIONSystem Identification is quite a mature area that has hadan interesting and productive development. Much has beendone, but many problems remain. I shall in this sectionoutline a few areas that I believe are worthy of morestudies, and I would like to encourage young researchersto have a go at these problems. Some open areas from anindustrial perspective follow in Section 6.4.1 Issues in Identification of Nonlinear ModelsA nonlinear dynamic model is one where ˆy(t|θ) = g(Zt, θ)is nonlinear in ZN (but could be any function, includinglinear, of θ). Identification of nonlinear models is probablythe most active area in System Identification today, Ljungand Vicino (2005). It is clear from Section 3 that there is acorresponding wide activity in neighboring communities,

The term machine learning was coined in Artificial In-telligence, see e.g. the classical book Nilsson (1965). Thearea has housed many approaches, like Kohonen’s self-organizing and self-learning maps, (Kohonen, 1984), toQuinlan’s tree-learning for binary data, (Quinlan, 1986),and the early work on perceptrons, (Rosenblatt, 1962),that later led to neural networks. More recent efforts,include Gaussian Process Regression (kriging), e.g. Ras-mussen and Williams (2006), which in turn can be tracedback to general nonlinear regression. Overall, the fieldson machine learning and statistical learning appear to beconverging

Among developments with relevance to System Identifica-tion are for example the bootstrap, see e.g. Efron and Tib-shirani (1993), and the EM algorithm, (Dempster et al.,1977). Other results of relevance to order selection are newtechniques for regularization (variants of (5b)), such asLars, Lasso, NN-garotte, see e.g. Hastie et al. (2001).