ICSA Colloquium Talk - 07/11/2019

Talk Title: 

Better software through deep learning

Abstract: 

Extracting  all available performance from modern processing systems is a complex task. It involves trial and error and expert knowledge about how hardware and software interact. Our development tools often choose an easier way: not trying too hard to extract performance. With our hardware and our optimisation decisions becoming increasingly complex, this approach will only cause the gap between possible and typically achievable performance to widen. My research aims at bridging this gap with analysis and optimisation methodologies which are fast, easy to use, and require little supervision or expert guidance. While I have worked on a number of different approaches, my current research relies on deep language learning to test, analyse, and improve complex code. This line of work has the potential of radically changing how we create optimisation and analysis heuristics.