Machine Learning on Source Code
summary
Provides concrete tooling and datasets: a public archive (3 TB, repositories with 50+ stars) pulls entire git blobs for training, parsing into a universal AST and language-agnostic roles. Models include RNNs for code completion, biLSTM for identifier splitting, and graph‑aware methods like var-misuse detection and code2vec/code2seq for function naming. Demos show Collab TPU training, a GitHub review bot that flags low‑probability tokens (e.g. i++ vs i--) and copy‑paste close‑file bugs, and a prototype clone detector and style enforcer.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 3:52Machine Learning on Source Code
- 7:37Data retrieval
- 9:08Data analysis
- 10:47Code analysis
- 12:03Babelfish
- 16:17Machine Learning
- 18:35Neural Networks
- 25:26Prediction
- 29:48Demos
- 34:01Life on GitHub
related talks
28mDelivering AI on Code: Live Demo of source{d}
25mMachine Learning for source code analysis
16mMachine Learning on Source Code
39mTraining AI To Code Using The Largest Code Dataset (Project CodeNet)
43mMachine Learning for Developer Productivity
36mIt's Alive! Machine Learning Writes Your Code! 🧠🔮
42mTraining AI To Code Using The Largest Code Dataset (Project CodeNet)
46mCode + AI: Will Robots Take Our Coding Jobs? ML Applied to Programming