Academic Writing & Publication
學術寫作與論文發表
Bring a question. We bring the method, the data, the compute and a mentor who has published. Curiosity is the only prerequisite.
A complete research project, from a first question to a paper written in English and matched to a real venue. Most students use data and machine learning as their instrument; the same journey works for any field with evidence.
CompetiLab's Academic Writing & Publication program takes a high school student through a real research paper: from the research question, literature, data and method to a paper written in English, and help submitting it to a student research journal, an international journal or a conference workshop. In person in Taipei or online, in English or Mandarin.
Most students meet academic writing as a format: an introduction, a body, a conclusion. We treat it as a way of thinking. A good paper is the visible trace of a question worth asking, evidence gathered with care, and an argument that survives a sceptical reader.
This program is a structured research apprenticeship. In the Standard track, students learn research method and the foundations of their instrument, read real papers, gather and analyze evidence, and write a research report at the level of an IB Extended Essay or AP Research project. The Advanced track continues into reproduction, an original contribution, rigorous evaluation and a manuscript prepared for a student journal, an international journal, or a conference workshop.
Students do not work alone, and they do not start from a blank screen. Each has a mentor with a research background, the studio's GPUs and AI tooling, and a set of designed modules for environments, data, evaluation and figures. The pieces are ready; the building is theirs. The weeks go to thinking, not to installation errors.
What machine learning is, and why it is a good instrument to start with
A machine learning model is a program that learns a rule from examples instead of being told the rule. Show it ten thousand nights of wearable data labeled with how well each person slept, and it learns to predict sleep quality for a night it has never seen.
That makes it the most hands-on kind of science a teenager can do. A biology student may wait weeks for a culture to grow. A student in machine learning changes one line, re-runs the experiment, and has a new result before lunch. The feedback loop is short, so the thinking gets exercised many times in a single project: what did I expect, what did I get, why are they different, what should I try next.
It also teaches intellectual honesty in a way few school subjects do. A model that scores well because it quietly saw the test data is a mistake every beginner makes once. Learning to find that mistake, and to distrust a result that looks too good, is the core of the scientific habit.
- 01
Real data, from day one
Public, properly licensed datasets in health, education, climate, language and vision. No permit, no lab, no waiting.
- 02
Like building with Lego
Designed modules for environments, data pipelines, evaluation and plotting, plus the studio's GPUs. The pieces are ready; what the student builds with them is theirs.
- 03
Thinking is the product
Hypothesis, baseline, control, ablation, interpretation. The paper at the end is the record of a way of thinking that lasts beyond the project.
- 04
A transferable skill
The same method applies in medicine, economics, ecology and engineering, and the writing is in the language of international research.
The Standard track builds research method, the foundations of the instrument and English research writing, at the level of an IB Extended Essay, AP Research or independent study. The Advanced track pursues a verifiable original contribution and a submission to a real venue.
| Standard · Research Foundation | Advanced · Research & Publication | |
|---|---|---|
| For | Beginners; international secondary students | Students with some Python, or a strong research motivation |
| Length | 4 to 8 weeks | 8 to 12 weeks |
| Reading | 3 to 5 core papers | 8 to 15 related papers |
| Skills | Git, Python, basic ML, paper structure | Reproduction, deep learning, ablation, statistical validation |
| Output | Research report | Manuscript in the format of a journal or conference |
| Submission | Student research journals or school research outputs | International journals, conferences or workshops |
Each session is a 90-minute meeting with the mentor, preceded by prepared teaching material and followed by a weekly assignment. Projects outside data science keep the same eight steps and swap the method module.
- 01
Research orientation and Git
A project repository, Python and Colab, version control, research ethics
A project repository and a list of research interests
- 02
Foundations of the instrument
Data cleaning, train/test splits, regression, classification, overfitting
A first baseline model
- 03
Literature deconstruction
Google Scholar and arXiv; abstracts, methods, limitations, citation
Reading notes on 3 to 5 papers and a comparison matrix
- 04
Proposal and research question
Hypothesis, dataset, metric, peer discussion
A two-page research proposal
- 05
Experiment design
scikit-learn, baselines, data leakage, commit discipline
A runnable notebook
- 06
Results and discussion
Model comparison, error analysis, figures, peer critique
A results table
- 07
Academic writing
Abstract, introduction, methods, results, discussion, references
A first full draft in English
- 08
Revision and presentation
Written feedback, language revision, presentation and defence, venue fit
Final paper, repository and research presentation
Session length and the scope of asynchronous feedback are written into the agreement.
After the eight sessions, Advanced students continue into the work that distinguishes a report from a contribution.
- 09
Reproduction
Reproduce a published paper and verify its settings
Reproduction report
- 10
Novel contribution
Form a testable improvement or a new question
Original hypothesis and research plan
- 11
Rigorous evaluation
Ablation, statistical comparison, cross-dataset and robustness tests
Complete experiments with error analysis
- 12
Publication preparation
LaTeX, formatting, research ethics, cover letter, reviewer response
Submission-ready manuscript
An Advanced project does not have to invent a new architecture. Rigorous new evidence on public data, missing data, model fairness, cross-domain generalization or compute cost can be a real contribution. Whether a project is novel enough is judged after the literature and experiment review.
Robust sleep prediction under missing wearable sensor data: how does the performance of lightweight ML models change when part of the wearable data is missing?
Using a public dataset with a valid research license, the student first compares logistic regression, random forest and XGBoost, then simulates different proportions of missing data and evaluates accuracy, robustness and compute cost.
This is a teaching example. Before any submission, novelty, data licensing and research ethics are verified.
Venues are matched to the work after it exists. Below are candidates with the limits that matter; our journal article on publishing as a secondary student has the fuller list. Acceptance is decided by each venue's reviewers.
Standard track candidates
International Journal of High School Research (IJHSR)
Original research by secondary students in natural sciences, engineering and computer science, or business and social sciences
No submission fee; a publication fee after acceptance; typically 1 to 3 months to a decision
Journal of Emerging Investigators (JEI)
Original research with an emphasis on STEM, with editorial mentoring through review
Check current fees and timing on the journal's guidelines page
National High School Journal of Science (NHSJS)
All scientific disciplines, including computing, data and AI
Peer reviewed; check the submission guidelines for fees
IB Extended Essay / AP Research
Course-based research outputs
Not an open journal; only students enrolled in those curricula qualify
Advanced track candidates: international journals
Sleep, circadian rhythm, sleep signals and ML
Must fit the journal's scope; an article processing charge applies and is a third-party cost
AI, computer vision, NLP, applied AI
A general peer-reviewed journal, not a student journal
Machine Learning and Knowledge Extraction (MDPI)
ML methods and applications
Peer reviewed; scope and charges to be confirmed
Interdisciplinary natural and applied sciences
An article processing charge applies and is a third-party cost
Advanced track candidates: conferences and workshops
ML methods and cross-domain applications
A formal international conference; highly challenging
AI, ML, vision, language, AI for science
Each workshop runs its own call; a workshop paper is not a main-conference acceptance
ML methods and theory
Extremely challenging; only for mature research
How topics scale from Standard to Advanced
| Direction | Standard topic | Advanced topic | Candidate venues |
|---|---|---|---|
| Sleep | Sleep quality prediction | Robust prediction with missing sensors | Clocks & Sleep |
| Education | Student performance prediction | Cross-school generalization and fairness | AI / J |
| Environment | Air quality prediction | Uncertainty-aware forecasting | J / ICMLA |
| Vision | Image classification comparison | Efficient models under limited data | MAKE / ICMLA |
| Language | Sentiment classification | Multilingual robustness under domain shift | AI / NeurIPS workshop |
A curious student should not lose a month to a driver that will not install. The studio provides the hardware and the tooling, with quotas defined in the agreement.
01
NVIDIA RTX 5090 workstations
For day-to-day training, ablations and quick iteration.
02
NVIDIA DGX Spark
A desktop AI system for larger models and longer runs.
03
H100 access
Dedicated GPU hours for Advanced projects that need them.
04
Claude API and Codex
AI assistants for reading, debugging and code review. Students disclose their use and remain responsible for every line and every reference.
05
Designed modules
Environment, data pipeline, evaluation and plotting templates, so the first experiment runs in the first week.
GPU model, hours or credits, validity period and overage rates are specified in the agreement.
Mentors are researchers whose own work is in a related field, including scholars with publications at leading machine learning venues. Advanced students are matched to a mentor in their project's area.
- NeurIPS
- ICML
We publish outcome figures only with a verifiable source.
Figures will appear here as they are documented.
- Secondary school students with a subject they cannot stop thinking about, in the sciences, social sciences or humanities. No coding or research experience needed.
- Students who want to try data and machine learning as a research instrument, with the mechanics handled for them.
- Students preparing an Extended Essay, AP Research project, science fair entry, or a first publication attempt.
- 01Turn an interest into a research question that is specific, answerable and worth answering.
- 02Find, read and evaluate scholarly sources, and summarize a field in a comparison matrix.
- 03Choose a method that fits the question, and when it is machine learning, understand what a model is and how to tell whether it actually works.
- 04Run controlled experiments or analyses with version control, and recognize leakage, overfitting and overclaiming.
- 05Write in the structure of a research publication, cite with integrity, and revise after critique.
- 06In the Advanced track: reproduce a published result, propose and test an improvement, and prepare a submission-ready manuscript.
Finding the question
- Research topic exploration
- Research question development
- Scope and feasibility
Reading the field
- Google Scholar, arXiv and databases
- Critical reading
- Reading notes and comparison matrices
Instruments
- Python, Colab and Git
- Data cleaning, splits, baselines
- Regression, classification, overfitting
Experimenting
- Controlled comparisons and ablations
- Error analysis and figures
- Statistics and uncertainty
Writing and integrity
- Proposal and paper structure
- Citation and academic integrity
- Revision, presentation, defence
Publishing
- Student journals and school venues
- International journals, conferences, workshops
- Cover letters and reviewer responses
The learning journey
Map the territory
Read widely around an interest, keep a question log, and notice where the explanations run out.
Commit to a question
Draft, test and narrow a research question until it is specific enough to investigate and open enough to matter.
Gather evidence
Search the literature, choose an instrument, and collect data or sources with a record of every decision.
Make sense of it
Build a baseline, run comparisons, interpret results, and be explicit about uncertainty.
Draft the paper
Structure the argument for a reader, write section by section, and cite as you go.
Revise with feedback
Respond to critique from mentors and peers, cut what the argument does not need, and check every claim.
Share the work
Prepare an abstract, poster or talk, and match the paper to a realistic venue.
Each student works with a mentor whose own research is in a related field. Our mentors have research backgrounds in molecular biology, modeling and data science, bioinformatics, clinical and public-health research, strategy and design, and include researchers who have published at leading machine learning venues such as NeurIPS and ICML. Advanced students are paired with a mentor whose background matches the project.
Mentors review code and drafts, question designs and give written feedback. Students write every line of their paper and run every experiment, using modules we have built so that the mechanical parts are already in place. Authorship follows actual contribution.
Sleep
Predict sleep quality from wearable data; in the Advanced track, test robustness when sensors go missing.
Education
Predict student performance; in the Advanced track, study generalization across schools and fairness across groups.
Environment
Forecast air quality or urban heat; in the Advanced track, forecasts that report their own uncertainty.
Language and history
Trace how a scientific controversy was framed in a newspaper archive with text mining.
Vision
Compare image classifiers; in the Advanced track, efficient models under limited data.
Replication
Repeat a small published behavioural experiment with a student population and discuss where results diverge.
01
A satellite and a regression
Open land-surface temperature data plus walking routes: a model of how heat shapes a school commute.
02
A century of newspapers
Topic models over a digitized archive to show how a city argued about vaccination, pollution or a dam.
03
A wearable at night
Three small models, one public dataset, and a careful study of what happens when the data has gaps.
04
A classroom's air
Low-cost sensors, a forecasting model, and an honest error bar.
05
A famous experiment, re-run
Reproduce a psychology result from the paper's own numbers, then test whether it holds on new data.
06
A paper's own reviewer
Use an AI assistant to check citations and logic, disclose it, and keep every sentence your own.
- Standard 4 to 8 weeks · Advanced 8 to 12 weeks
- Eight 90-minute mentoring sessions (plus four research stages in Advanced), weekly assignments and written feedback between sessions. In person in Taipei or online, in English or Mandarin.
- Standard: curiosity is the only prerequisite. Advanced: some Python, or a question you cannot put down.
Does my child need to know how to code or to have done research?
No. The Standard track starts from the beginning, with prepared notebooks and modules. Students who already code move faster into experiments; students who do not learn enough in the first two sessions to run their own. Curiosity and the willingness to revise matter more than anything else.
Why do most projects use machine learning?
Because a student can run a real experiment on a real, public dataset on the first day, see the result in minutes, and try again. That short loop exercises hypothesis, test and interpretation many times in one project. Students in the humanities or social sciences follow the same eight sessions with a method module suited to their field.
Will the paper be published?
It depends on the work and the venue. The Standard track targets student research journals or school research outputs. The Advanced track prepares a manuscript for international journals, conferences or workshops, and we help choose a realistic target and respond to reviewers. Acceptance is decided by their reviewers.
Does a paper help with university applications?
Admissions officers at selective universities read independent research seriously, because it shows initiative, depth and the ability to finish. We publish outcome figures only when they come from a verifiable source.
Is the program taught in English or Chinese?
Sessions can be held in either language. Papers are usually written in English because most literature and venues use it, but this is decided with the student.
- How a high school student writes a research paper and gets it published17 Oct 2026 · 5 min read
- Where a secondary school student can publish research15 Oct 2026 · 5 min read
- How to write a research question you can actually answer2 Sept 2026 · 3 min read
- What machine learning is, and why it is a good first research field10 Oct 2026 · 3 min read
Begin with a question.
Tell us what a student cannot stop thinking about. We will suggest where it could go.