Introduction

Software tools are extensively used in various phases of software development to make software reliable and efficient. Such tools include code generators and synthesizers, optimizing compilers, automated test generators, debuggers, and smart Integrated Development Environments (IDEs) supporting code completion and static bug detection (e.g. https://www.jetbrains.com/idea/). Software tools are often based on traditional program analysis techniques and formal methods. In the recent years, we have realized that we have access to a large body of open-source software systems. Such open-source systems have massive amount of data and meta-data on programs. This availability of massive amount of data has encouraged researchers to devise data-driven approaches to development of software tools. Such tools have managed to effectively

1. synthesize programs from input output examples
2. discover parts of programs that are likely buggy
3. automatically fix bugs and syntax errors
4. perform code search and code clone detection
5. perform effective fuzz testing
6. de-obfuscate minimized JavaScript programs
7. come up with meaningful names for functions

In this course, we will read several recent papers where machine learning has been used to build effective software development and programming tools. We will read two papers per week. In each lecture a student will present a paper for 40 minutes with all necessary background information. The student will also create three questions on the contents of the paper. The questions should be such that it could be answered only after having a deep understanding of the paper. The rest of the class will then review the paper and post their reviews and answers to the questions on Piazza 14.5 hours before the next lecture. In the next lecture, in addition to the discussion of a new paper, we do a deeper discussion of the paper presented in the previous lecture. If you have registered for 3 units, you need to implement the technique presented in one of the papers and try to reproduce the results. The implementation and report will be due on March 23. If you have registered for 4 units, you need to extend the implementation with new ideas and produce results that are potentially publishable. The deadline for this extended project will be May 4. The relative weight of the components of your grade will be approximately:

1. Class presentations, leading discussions, and commenting on the reviews: 30%
2. Reviews, answering questions, and class participation: 30%
3. Programming assignment (implementing an existing technique and possibly extending it): 10 % (assignment) + 30 % (project)

Office Hours:

• Wednesdays 11 am - 11:59 am at 735 Soda Hall

Tentative Schedule

1. 1/17: Introduction
2. 1/22: Brief introduction to deep learning. Reading material and references:
   1. Karpathy's blog
   2. Karpathy's blog on RNN
   3. Blog on LSTM
   4. Building Auto Encoders in Keras [Quick Reference]
   5. Deep learning with Keras [Quick Reference]
   6. PyTorch [Quick Reference]
   7. Hands-On Machine Learning with Scikit-Learn and TensorFlow [Quick Reference]
3. 1/24: #jsnice: Speaker: Rohan Bavishi
4. 1/29: #deepfix: Speaker: Andrew Head
5. 1/31: #deepcoder: Speaker: Gil Lederman
6. 2/5: #clone: Speaker: Sameer Reddy
7. 2/7: No class (Koushik traveling)
8. 2/12: Summary of MLPS workshop
9. 2/14: #qfuzzer, #mlfuzzer: Speaker: Kevin, Caroline
10. 2/19: Holiday
11. 2/21: Discuss #qfuzzer, #mlfuzzer, #ggnn: Speaker: Edward
12. 2/26: #r3nn: Speaker: Xinyun
13. 2/28: #apilearning: Speaker: Andrew
14. 3/5: #nl2program: Speaker: Katia
15. 3/7: Satish Chandra (Guest speaker) Project proposals (3 minutes presentations).
16. 3/12: #codenn: Speaker: Rohit Ramesh
17. 3/14: #deepxplore: Speaker: Liang
18. 3/19: Rishabh Singh (Guest speaker)
19. 3/21: #bayou: Speaker: Rafael
20. 3/26: Spring Break
21. 3/28: Spring Break
22. 4/2: #opencl: Speaker: Caroline
23. 4/4: #hoarefeedback: Speaker: Rohit Ramesh
24. 4/9: #seq2sql: Speaker: Xinyun
25. 4/11: #dynrepair: Speaker: Sameer
26. 4/16: #characterlevel: Speaker: Rohan
27. 4/18: #code2vec: Speaker: Edward
28. 4/23: Discussion of code2vec and Final project presentation
29. 4/25: Final project presentation

Papers and Tentative Presentation Schedule