DTAI

Monday February 13 2012, at 16h30 in Celestijnenlaan 200A (auditorium 00.225)

Machine Learning Approaches for Processor Performance Prediction
By Beau Piccart (PhD student DTAI)

Performance numbers reported by benchmarking consortia and corporations provide little or no insight into the performance of applications that are not part of the benchmark suite. This talk presents an overview of ML methods used to alleviate this problem. The first approach consists of using micro-architecture independent characteristics to estimate program similarity. A second approach uses performance numbers for a limited set of machines. This results in a two-way dataset, a data set consisting of pairs of objects. The proposed two-way setting has some peculiarities that can be exploited, with better predictive performance as a result.