Machine Learning Methods in Software Engineering

We apply anomaly detection algorithms to all publicly avaliable Kotlin code to identify anomalous code fragments that can be of interest to the Kotlin compiler team.

Basically, we downloaded all Kotlin code from GitHub, vectorized it, and fed it into an anomaly detection algorithm. We did the same for bytecode, where we could. We found a bunch of highly unconventional code snippets that are of interest to the Kotlin compiler team, and several examples of suboptimal compiler behaviour.

The paper was accepted at MSR 2020.