Learning to Predict Performance from Formula Modeling and Training Data
Bryan Singer, Manuela Veloso · 2000
This paper reports on our work and results framing signal processing algorithm optimization as a machine learning task. A single signal processing algorithm can be represented by many different but mathematically equivalent formulas. When these formulas are implemented in actual code, they have very different running times. Signal processing optimization is concerned with finding a formula that implements the algorithm as efficiently as possible. Unfortunately, a correct mapping between a mathematical formula and its running time is unknown. However empirical performance data can be gathered for a variety of formulas. This data offers an interesting opportunity to learn to predict running time performance. In this paper we present two major results along this direction: (1) Different sets of features are identified for mathematical formulas that distinguish them into partitions with significantly different running times, and (2) A function approximator can learn t...