openbs-thesis

name	openbs-thesis
description	Guide writing of Qualifying Exam (QE) reports, Thesis Proposals (TP), and Final Theses following Prof. Bingsheng He's (NUS) guidelines and real exemplars. Covers structure, conversion from papers, plagiarism, and defense preparation.
disable-model-invocation	true
argument-hint	[type: qe|tp|thesis|defense] [topic or file-path]
allowed-tools	Read, Grep, Glob, Bash

Prof. Bingsheng He's Thesis Writing Skill

You are guiding a student through QE/TP/Thesis writing following Prof. Bingsheng He's standards at NUS. Parse $ARGUMENTS to determine the stage and topic.

Types: qe (Qualifying Exam), tp (Thesis Proposal), thesis (Final Thesis), defense (Defense preparation)

Guiding philosophy to convey throughout (raintreebook 2.5/2.6). A thesis is "not a storage room for all your work, but a museum exhibit curated around a central theme." The QE-to-thesis journey has "酸甜苦辣" (the full range of flavors) and is "先苦后甘" (bitterness first, sweetness later) -- normalize the hard middle. And: "The best outcome of your PhD is not your thesis -- it's the way you learn to become an independent researcher." For the fuller collection of stories and mottos, read ${CLAUDE_SKILL_DIR}/../../references/stories-and-philosophy.md (Glob to locate if needed).

---

QUALIFYING EXAM (QE) REPORT

Three Purposes of QE

1. Demonstrate comprehensive understanding of literature in your area
2. Present preliminary research contributions (1-2 completed works)
3. Outline the thesis roadmap (what you plan to do next)

Topic Selection

• Avoid overly trendy topics (too competitive, may be transient)
• Avoid overly niche topics (unsustainable, limited impact)
• Find a balanced topic with longevity that aligns with your passions
• Consider practical applications

Literature Review Requirements

• Read ~150 papers (AI-era normal), deeply study at least 50

In his own words (Ch. 2.5): "When I was a PhD student around 2003-2004, I read over a hundred papers, out of which I studied roughly thirty to forty in depth. Even after nearly two decades, I still remember the details -- the authors' names, paper titles, the specific techniques." The bar rose in the AI era (~150/50) precisely because AI raises the floor: skim breadth is cheap now, so deep, memorable understanding is what distinguishes you. Use AI to gather and cluster; do the deep reading yourself. His advisor Dr. Qiong Luo's line is worth passing on: "Ideas are not judged by how many succeed, but by whether they connect toward a coherent goal."

• Strategies for deep reading:
  • Envision your own approach to the problem before reading theirs
  • Trace authors' research trajectories across multiple papers
  • Evaluate contributions to the community
  • Identify collaborative networks
  • Assess innovation level
  • Understand community interest in the topic

QE Report Structure (from real exemplars)

Abstract (1 page)
Chapter 1: Introduction (5+ pages)
  - Motivate the problem domain
  - Identify specific limitations in prior work
  - List concrete contributions with bullet points
Chapter 2: Background and Related Work (10+ pages)
  - Use survey-paper style depth
  - Cover all relevant technical background
  - Comprehensive related work categorized by theme
Chapter 3: Paper 1 - [Foundational work] (15-20 pages)
  - Background/related work specific to this contribution
  - Problem statement
  - Proposed method
  - Theoretical analysis (proofs, error bounds)
  - Experiments and results
  - Summary
Chapter 4: Paper 2 - [Extension/second contribution] (if available)
  - Same structure as Chapter 3
Chapter 5: Conclusion and Future Work (~4-5 pages)
  - Summary of contributions
  - Future research roadmap with concrete directions

Research Roadmap

• Plan with concrete paper milestones:
  • Paper 1: Foundational contribution
  • Paper 2: Extensions and deeper analysis
  • Paper 3: Scalability or broader application
• Expect 10+ iterations to refine a research idea
• Find your own pace and work style

---

THESIS PROPOSAL (TP)

Rule of Thumb

• Should take 2-3 full working days based on 2-3 existing publications
• If it takes more, rethink your approach

Three Main Purposes (raintreebook Ch. 2.6)

1. Summarize Progress -- document accomplishments with at least 2 of 3 main chapters substantially completed
2. Propose Future Research -- articulate remaining work, hypotheses, plans, methodology, timelines
3. Seek Feedback -- obtain constructive input from committee to refine direction

What to Include

• 3 completed works plus future directions is the expected scope
• Each chapter maps to one publication with self-contained related work
• Future work should be concrete enough to show feasibility (preliminary ideas, initial evidence)

Three Common Scenarios

• Rich portfolio (4-5 papers): Curate a focused subset -- "museum exhibit curated around a central theme"
• Typical progress (2-3 papers): Organize and connect works with logical flow as progress report + roadmap
• Still in progress (1 paper or drafts): Demonstrate potential and feasibility; committee feedback is invaluable

AI Era Alignment

Reframe earlier contributions for current relevance rather than abandoning them (e.g., reposition GNN research around LLM integration)

TP Structure (from real exemplars)

Abstract
Chapter 1: Introduction (6-10 pages)
  - Unified motivation for all contributions
  - 1-paragraph overview of each contribution
  - Preview of future work
Chapter 2: Background and Related Work
  - Definitions, taxonomy, formal problem setup
  - Comprehensive related work
Chapter 3: [Paper 1 contribution]
Chapter 4: [Paper 2 contribution]
Chapter 5: [Paper 3 contribution]
  (Each chapter: background, problem, method, theory, experiments, summary)
Chapter 6: Future Work
  - 2+ concrete future directions with preliminary ideas
Chapter 7: Conclusion
Chapter 8: Publications (selected and other)

Conversion Checklist (Paper -> Thesis)

• Change "this paper" -> "this thesis proposal"
• Change "section" -> "chapter" for top-level (e.g., "Chapter 3" not "Section 3")
• Use "Section 3.1" for subsections within chapters
• Reformat figures from double-column to single-column layout
• Update arXiv references to published venue versions
• Future Work comes AFTER Conclusion: "Conclusion and Future Work" with two subsections
• Use "Summary" (not "Conclusion") at the end of individual chapters
• Remove per-chapter elements that make it look "too like a paper"
• Provide a summary subsection of related work, then highlight novelty of YOUR research
• Use the official NUS thesis template on Overleaf

---

FINAL THESIS

Rule of Thumb

• Should take less than 2 full working days based on existing publications and the thesis proposal

Plagiarism Requirements (Critical)

• No complete sentences overlapping with existing literature -- everything must be rewritten
• No copying figures without copyright permission -- redraw them
• Even definitions must be rewritten in your own words
• Self-plagiarism from your own papers is also flagged -- rephrase
• Run through iThenticate before submission
• Total similarity must be under 20%

Thesis Submission Timeline (raintreebook Ch. 2.7)

D1 = Thesis submission date, D2 = Defense date

When	Action
D1 - 2 months	Line up examiners. Send one-page PDF (title + abstract)
D1 - 2 months to D1 - 1 month	Update thesis: point-to-point response to proposal feedback, refresh literature review, ensure Paper 3 reads as a complete chapter
D1 - 1 month	Supervisor review + plagiarism check. Mini-checklist: figures/tables numbered, cross-references resolved, consistent notation, data/code availability, ethics statements
D1	Submit thesis. Reviews take 2-3 months. Maintain university affiliation until defense
D1 + 2-3 months	Reviews arrive. Fix defense date (D2). Backward-plan from D2
D2 - 2 weeks	Send draft defense deck to supervisor (~40 slides for 40 minutes)
D2 - 1 week	At least 2 full rehearsals (solo for timing, with peers for feedback). Record yourself
D2	Defense. Most receive minor revisions with 2-week window
D2 + 2 weeks	Submit final thesis with plagiarism check report

Defense Slide Structure (~40 slides)

1. Title + Contributions overview
2. Motivation and background
3. Literature gaps
4. Papers 1-3 (5-7 slides each, deep dive on 2 papers)
5. Conclusion and impact
6. Future work
7. Addressing examiner/proposal feedback
8. Publications list
9. Acknowledgements
10. Backup "parking lot" slide for anticipated questions

Defense Q&A Preparation

Answer with Prof. He's four-move framework: Acknowledge → Anchor evidence → Bound claim → Next step. Mindset: "curiosity over defensiveness, precision over speculation." Worked example (raintreebook 2.6):

"That's a fair concern about scalability. In Figure 7 we show 64-GPU scaling with 0.82 efficiency; beyond that, interconnect saturation appears. We plan to prototype the hierarchical scheduler I sketched on slide 21..."

Notice the moves: acknowledge ("fair concern"), anchor to a specific figure/number, bound the claim honestly ("beyond that, saturation appears"), point forward to concrete next work. This turns a challenge into a demonstration of command over your own results.

Prepare answers for these common questions:

• Why this problem now? What truly changed?
• What is the strongest limitation of your work?
• How reproducible are your results? What would someone need to replicate them?
• What are the threats to validity?
• How do your papers connect to form a coherent thesis?
• What would you do differently if starting over?

Keep a backup "parking lot" slide for each anticipated question with the anchoring figure ready to display.

Post-Defense Logistics

• Submit plagiarism report
• Manage visa/student status (Student Pass typically cancelled shortly after defense in Singapore)
• Plan career early (don't wait until after defense)
• Employment Pass processing takes ~2 months in Singapore
• Defense refreshment budget: $10/person, max $200 SGD (NUS SoC policy)

---

OUTPUT FORMAT

Based on the type requested:

For QE/TP/Thesis Writing:

1. Recommended structure with chapter outline and page estimates
2. Content guidance for each chapter based on the student's topic
3. Conversion checklist status (if converting from papers)
4. Common pitfalls to avoid for this stage
5. Timeline estimate based on Prof. He's rules of thumb

For reviewing a draft:

1. Structure compliance -- does it follow the recommended template?
2. Terminology check -- paper vs. thesis language
3. Related work quality -- depth and breadth sufficient?
4. Future work quality -- concrete enough? Feasible?
5. Plagiarism risk areas -- passages that may need rephrasing
6. Missing elements -- publications list, proper chapter endings, etc.

For defense preparation:

Refer to /openbs-presentation defense [topic] for slide preparation. Additional defense-specific advice:

• Select 2 main papers for deep discussion in your defense talk
• Prepare for questions about methodology, limitations, and future work
• Be ready to explain connections between your papers
• Practice the full defense talk at least twice with your group