BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20121114T203000Z
DTEND:20121114T204500Z
LOCATION:155-F
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: Selecting suitable optimizations for a particular class of applications is difficult because of the complex interactions between the optimizations themselves and the involved hardware. It has been shown that machine-learning based driven optimizations often outperform bundled optimizations or human-constructed heuristics.=0A	In this dissertation, we propose to use different modeling techniques and characterizations to solve the current issues in machine-learning based selection of compiler optimizations. =0A	In the first part, we evaluate two different state-of-the-art predictive modeling techniques against a new modeling technique we invented, named the tournament predictor.  We show that this novel technique can outperform the other two state-of-the-art techniques. =0A	In the second, we evaluate three different program characterization techniques including performance counters, reactions, and source code features. We also propose a novel technique using control flow graphs (CFG), which we named graph-based characterization. =0A	In the last part, we explored different graph-based IRs other than CFGs to characterize programs.
SUMMARY:Automatic Selection of Compiler Optimizations Using Program Characterization and Machine Learning
PRIORITY:3
END:VEVENT
END:VCALENDAR