Overview

Check our homepage for assignments

Introduction to software engineering research

• three easy pieces: reading, writing, and hacking
• grading policies

How to read

• why, what, and how to read a research paper

Why We Created This Course?

Raise the bar.

“Research” is serious and hard

• NOT like your undergrad assignments

Software Engineering Research

Tries to answer the fundamental question of how to build a piece of software

Make 码农 an easier life

• faster (productivity)
• better (quality)

Ultimate goal: take over human's role in software development

To Be More Specific...

See ICSE's call for papers

• AI and software engineering
• Testing and analysis
• Software analytics
• Software evolution
• Social aspects of software engineering
• Requirements, modeling, and design
• Dependability

Examples

“Empirical and human study”

• What happens to a programmer's brain (ICSE'20)

“Technologies”

• FlashFill in Excel (POPL'11)
• AddressSanitizer (ATC'12)

For fun and profits

• Solve hard problems; build useful tools!
• A quick test for evaluating your work: do real engineers (e.g., someone in Google/Facebook) will feel your work interesting?

Reading

Read all papers in the list

• Good papers (novel and significant) of good taste
• Cover many SE areas
  • Empirical software engineering (EMSE)
  • Software engineering process (SEP)
  • Formal methods (FM)
  • Software testing and analysis (STA)
  • Software maintenance and evolution (SME)

(This is today's topic)

Writing

Be concise and precise.

• Bad writing practices
  • broken logic
  • imprecise wording
  • verbose writing

SUGGESTIONS

• Start writing as early as possible
• Rush into your supervisor's office and let him face-to-face revise your manuscripts

Hacking

RTFM, STFW, RTFSC.

• Drop course if you don't know how to use Git.

Coding should mean nothing to you (just some implementation work)

• Intercept file system calls; capture block-device traces; create and compare file system snapshots
• Modify Android Runtime (ART) to expose a debugging interface
• Modify Java regex engine to obtain matching feedbacks
• Parse source code via Clang/LLVM

Grading Policies

Reading (presentation sessions, 30%)

• Four papers a session (15-minute talk + 3-minute QA, as if you were the authors); bid paper today
• Bring a laptop. Projector is 4:3 (VGA)

Why Read a Paper?

Personal experience: almost all of my ideas emerged from reading textbooks and papers.

Gonna some big ideas?

• if your thoughts really matter
  • 99.9% have already been explored
  • find a related work

What is a Paper?

A piece of work that teaches the research community a lesson. (Hope everyone remembers this!)

1. Motivation and observations
   • the most important part; why they came up with the problem?
2. Explanations and analyses
   • arguments, evidence, and defenses against potential threats
3. Solution and evaluation
   • this is the least important part
   • if you have sufficient knowledge about the problem (and background), you are ready to propose your own solution

How to Find a Paper?

All papers form a citation graph

• Your supervisor should give you some seed papers
  • if not, quit now
  • if of low-quality, quit now
  • traverse the graph (from the seeds) to find more papers
• Web resources: Google Scholar recommendations, hacker news, blogs, tweets, github repos (lot of paper lists), ...
• Textbooks: further reading (usually awesome classical papers!)
• Magazines: Communications of the ACM, IEEE Spectrum, IEEE Software, ...

How Many Should I Read?

A LOT!

• Papers: 100/year (~1000 pages, very dense)
  • 2 per week
• Textbooks: 20/year (~10000 pages, very comprehensive)
  • better with online courses
• Magazines: 24/year (~2400 pages, less dense)

How to read a paper

S. Keshav. How to read a paper, ACM SIGCOMM Computer Communication Review, 3(37), 2007.

Three-pass approach

1. Quick scan
   • category, context, correctness, contributions, clarity
2. Jot down the key points
   • read with care, but ignore details as proofs
3. Virtual re-implement
   • re-create the work as if you were the author

Reading Paper: My Approach

Just like back-propagation in training neural networks.

New problem

• Ask myself: why didn't I noticed this important problem?

Old problem, new solution

• Ask myself: how to solve it? is there any tricky parts?

Finally

• Watch the author's presentation to check their own understandings

Traps in Paper Reading

Only read papers related to my topic

• be a good problem solver first

This paper is limited in XXXX. It's a piece of junk!

• every paper has its limitation
• try to be constructive: what can I learn from it?

I cannot hold!

• you're not ready for this paper.

Short Summary

1. Read a lot of papers
2. Read really good papers (to raise your bar)
   • 90% “top-conference” papers have nearly zero “real” contributions
     • they may be doing very well job, but not groundbreaking
   • many non-top-conference papers have negative contributions

FlashFill

Sumit Gulwani. Automating string processing in spreadsheets using input-output examples (POPL'11, Most Influential Test-of-Time Paper Award in 2021)

• The most hard-to-read paper for this semester

1st pass: the problem is to find a program $P$ that generalizes existing “examples” (and 90% students stop here)

Ready for the Second Pass?

Ready for Reading This Paper?

I Cannot Hold 😂😂😂

(Unix is user-friendly; it's just choosy about who its friends are.)

Ready for the Second Pass?

What's this?

• (ICPC WF'08 Problem I: Password Suspects) find the count of all $n$-length passwords that “contain” a given set of observations.

Read More.

Trust the authors: papers are self-contained. Reading them only requires textbook knowledge.

• [Corollary] Go read (good enough) textbooks if you find you can't understand a paper
  • compilers, mathematical logics, algorithms, programing language theory (optional)

You can find open courses on Youtube (or Bilibili)

More Comments on Reading...

Feeling hard? See Manuel Blum's advice to a beginning graduate student.

Virtually Re-implement?

You may get a new paper (similar technique)!