ABSTRACT
The problem is arguably hard enough for long noise-free data sets generated on a computer from low-dimensional maps or differential equations. For “real” data, as the speakers at this conference have repeatedly emphasized, the problem is far more difficult. (And as the organizers have repeatedly reminded us, far more valuable.) Real data is contaminated with noise (which is rarely additive, Gaussian, or white), measured with finite precision, and subject to innumerable external influences in the environment and the measurement apparatus. And, of course, there is never enough of it.
