Monotony and Surprise

Abstract

The extraction of recurrent patterns from sequences is a popular and intensive application, ubiquitous to many domains. Often, however, the process exposes more candidates than one can afford to inspect, thereby defying its whole purpose. While part of this problem is endemic, part of it can be attributed to the traditional definitions of what constitutes a pattern, that hinge alternatively on syntactic or statistical properties alone. It has been seen recently that this part of the problem may be mitigated by more prudent paradigms, in which the syntactic description of a pattern and the list of all of its occurrences are tightly intertwined. This approach leads to identify regions of monotonicity for some scores of surprise in use, within which it is enough to consider and weigh only extremal terms and values. This talk reviews concepts, constructs, and application results obtained along this line of research.