Skip to content

Latest commit

 

History

History
69 lines (65 loc) · 2.93 KB

README.md

File metadata and controls

69 lines (65 loc) · 2.93 KB

ML4SE Seminar Papers

List of papers to read

  • Andrew Scott, Johannes Bader, Satish Chandra "Getafix Learning to fix bugs automatically" arxiv 2019.
  • Sifei Luan, Di Yang, Koushik Sen, Satish Chandra "Aroma: Code Recommendation via Structural Code Search." arXiv 2018.
  • Saksham Sachdev, Hongyu Li, Sifei Luan, Seohyun Kim, Koushik Sen, and Satish Chandra. "Retrieval on source code: a neural code search." International Workshop on Machine Learning and Programming Languages (MAPL 2018).
  • Wang, Ke, Rishabh Singh, and Zhendong Su. "Dynamic Neural Program Embeddings for Program Repair." ICLR 2018.
  • Ray, Baishakhi, Vincent Hellendoorn, Saheel Godhane, Zhaopeng Tu, Alberto Bacchelli, and Premkumar Devanbu. "On the" naturalness" of buggy code." ICSE 2016
  • M. Brockschmidt, M. Allamanis, A. L. Gaunt, O. Polozov. "Generative Code Modeling with Graphs." ICLR 2019
  • Xinyun Chen, Chang Liu, Dawn Song "Tree-to-tree Neural Networks for Program Translation." NeurIPS 2018
  • Anshul Gupta and Neel Sundaresan. "Intelligent code reviews using deep learning" KDD 2018
  • Riley Simmons-Edler, Anders Miltner, Sebastian Seung: "Program Synthesis Through Reinforcement Learning Guided Tree Search." arxiv (2018)
  • Alexander L. Gaunt, Marc Brockschmidt, Rishabh Singh, Nate Kushman, Pushmeet Kohli, Jonathan Taylor, Daniel Tarlow "TerpreT: A Probabilistic Programming Language for Program Induction" arxiv (2016)

List of read papers

[Feb 15th, 2019] DeepCoder: Learning to Write Programs

Authors: Matej Balog, Alexander L. Gaunt, Marc Brockschmidt, Sebastian Nowozin, Daniel Tarlow. Venue: ICLR Year: 2017 Labels: Program Synthesis, Programming By Example, Deep Learning Summary:

  • DeepCoder is a technique to augment Inductive Program Synthesis (IPS), by guiding the synthesis with a deep neural network that predicts a set of functions that should be present in the program. A model is trained based on synthetically generated programs and coresponding input-output examples. To constrain the search space they define a DSL in which they synthesize programs.
    • Good Evaluation
    • Nice illustration of embeddings in 2d
    • Discussion of limitations Limitations (Open Questions):
  • Works for very small programs.
  • Data Generation. They use synthetic examples for training and evaluation; how does this generalize to realistic programs.
  • As number of problems solved increases, Deepcoder speedup decreases compared to a baseline. Why is this the case, does the order of programs being solved matter. Possible related work to read:
  • Learning Syntactic Program Transformations from Examples
  • TerpreT: A Probabilistic Programming Language for Program Induction

[Feb 22nd, 2019] Structured generative models of natural source code

Authors: Maddison, Chris, and Daniel Tarlow. Venue: ICML Year: 2014 Summary:

  • They build a generative model of source code using improvement of PCFG, as well as a PCFG for a baseline.