Reading

Newbie phase (unfamiliar research area)

• Newbies: courses $\approx$ tutorials $>$ textbooks $>$ talks $>$ papers
• Ask GPT when you are out of sync

Professional phase (your topic's related research)

• Build a reference graph
• Virtually re-implement every paper
  • Be constructive (rather than destructive) on every paper

Writing

Writing is program synthesis

• Manuscript is a program (with control- and data-flow)
• Keep a tree-like structure for your manuscript

Readers are neural networks

• Follow the “least surprise principle”: make predictions online
• Take care of all types of readers (teach them lessons!)

Ask your advisor for comments

Writing (cont'd)

We have this year's proceedings!

• Thank you all!

Complains from the Teaching Assistant

Writings are a bit loose...

Although LLVM IR is powerful and widely used, it is too low-level and lacks flexibility...

PGO leverages the dynamic runtime behavior information of a specific program, namely profile data, to seek more optimization opportunities...

Ask GPT to help you!

Hacking (Not Covered in This Course)

Write good code

• Readable - make specifications explicit
• Verifiable - make correctness explicit

Life is short, use

• ChatGPT, IDE (copilot), and command line tools
• Domain-specific languages
  • Example: I use Jinja2 for spamming

Evaluation Criteria

Distinguished Presentation Award

• Logic flow, presentation, slides, paper quality

Distinguished Proposal Award

• Novelty and significance, research plan, presentation

Award Selection Process

First-round

• Two reviewers (all proposals)
• Each proposed 5 candidates

Second-round

• Voted by the advisors
  • There is randomness! See this paper

Prize: USB Stick (64 GB)

Somehow useful...

Instructor's Comments on Awards

有人的地方，就有江湖。

• Not so simple as “random”
• There are “human factors”
  • Scholarships/Fellowship Awards
  • Dissertation Awards
  • Funding Awards
  • ...

Instructor's Comments on Spamming Papers

这里面就有我审的 paper，向来普遍拿高分的都是 incremental novelty 的，比较保守但实验又比较好比的，在一个小框子里和前面的比了有收益，但是这样的工作也注定格局不大，甚至自己都没有意识到这个所谓的框子都是伪需求，没有任何意义，或这个框子的方向本身就是错的，没有未来。

审这样的 paper 最直观的感受就是你找不出它的任何毛病，但是从 merit for acceptance 角度就觉得真的没什么意义，但这个不能成为拒掉的理由，完全是 research taste。

—— An Anonymous Reviewer

Instructor's Course Summary

This year...

• We've cheated
  • We've lightened our teaching loads
  • I still loved the proceedings
• We've decided to eventually close this course
  • I'd like to spare some time for more important stuffs (undergraduate teaching)

Take Away Messages

Three easy pieces: reading, writing, and hacking

• The reading list (read good papers)
• How to speak
• Benchmarking crimes

Always seek for significant contributions

• Stay foolish, stay hungry
  • (This is not “求知若渴，虚心若愚”)