HYPOTHESIS GENERATIONSuccess StoriesA recurring issue in data-driven feature extraction systems is the combinatorics of search in hypothesis space. Brute force attempts at model generation, where all possible combinations of models are evaluated, leads to exponential growth in the size of the search space. Efficient search and exploitation of this sample space is at the heart of the AUTOBOX heuristic as it smartly identifies the efficient model. The AUTOBOX heuristic iterates through model identification by developing the initial model through mappings of theoretical covariances to actual covariances. This is followed by a iterative sequence of model augmentation/simplification strategies culminating in an approximation of the true model (but unknown model) which generated the actual observations. The augmentation strategies speak to the issue of sufficiency and are akin to stepwise forward procedures in regression. The simplification issue deals with necessity checking and it is similar if not identical to stepdown regression. Sometimes necessity leads to model simplification and elimination of redundant or superfluous model structure. There is a proven need to test the robustness of the initial model and to revise and learn how to deal with a number of gaussian violations. Initial model identification takes advantage of the relationship between the theoretical covariance function and alternative model forms. Now on to Success Stories to pursue examples where data leads to conclusions and assertions rather than simple percentages vis-a-vis previous periods. |