Untitled

Most hardware predictors are table based (e.g. two-level branch predictors) and have exponential size growth in the number of input bits or features (e.g. previous branch outcomes). This growth severely limits the amount of predictive information that such predictors can use. To avoid exponential growth we introduce the idea of "dynamic feature selection" for building hardware predictors that can use a large amount of predictive information. Based on this idea, we design the dynamic decision tree (DDT) predictor, which exhibits only linear size growth in the number of features. Our initial evaluation, in branch prediction, shows that the general-purpose DDT, using only branch-history features, is comparable on average to conventional branch predictors, opening the door to practically using large numbers of additional features.

show abstract

Untitled

Utgoff

1997

View full text Add to dashboard Cite

Abstract. The ability to restructure a decision tree efficiently enables a variety of approaches to decision tree induction that would otherwise be prohibitively expensive. Two such approaches are described here, one being incremental tree induction (ITI), and the other being non-incremental tree induction using a measure of tree quality instead of test quality (DMTI). These approaches and several variants offer new computational and classifier characteristics that lend themselves to particular applications.

show abstract

Untitled

Cited by 19 publications

References 20 publications

Learning Task-Specific Object Recognition and Scene Understanding

Learning Task-Specific Object Recognition and Scene Understanding

Dynamic feature selection for hardware prediction

Untitled

Contact Info

Product

Resources

About