Applying machine learning to coding itself
We'll get back to stochastic programming soon; I wanted to do a quick post about some updates to my earlier series on anti-unification.
As I noted in the final part of that series, I spent a few months in 2018 helping out with a research effort called "GetAFix" to see if it was possible to use AST differencing and anti-unification to deduce patterns in bug fixes, and to then infer possible bug fixes for new bugs from those patterns. I am very pleased to report that yes, it does work.
This was not the only research project in the area of big-code analysis for developer productivity that our group was pursuing; here's a fascinating video where my colleagues Satish, Sonia, Frank and Johannes discuss their work in this area; the bit about GetAFix starts around the 30 minute mark, but it is all good stuff.
If you want a more technical description of the algorithms we used for GetAFix, the current team has written a paper which you can find here.