Day 1 — Introduction and Reading
[to be updated]
Day 1 is mostly introductory: what research is and why it is worth doing; how to read the research brief and research papers; and, choosing the problem to work on for the rest of the week. The examples below walk through an illustrative brief on health misinformation, so you can see each step done once before trying it on the EROP brief.
Schedule
All sessions are at NUS Innovation 4.0.
| Slot | Activity | Location |
|---|---|---|
| 15 min | Introduction | Lecture hall |
| 30 min | Research brief | Lecture hall |
| 15 min | Break | — |
| 45 min | Reading advice | Lecture hall |
| 15 min | Break | — |
| 2 hr | Research project with mentors | Tutorial room |
Deliverable: Outlined a problem to work on with general directions for a proposal.
Pre-Survey
Before the program begins, both mentees and mentors fill in a short self-assessment so we have a baseline to look back on at the end. The link will be shared by the program coordinators on the day.
What is this for?: We use these self-assessments as a signal for how much the program actually helped you. Thank you for taking the time to fill it up 🙇🏻
Introduction
Research is the process of producing new insights, and showing it carefully enough that other people can check it and build on it. In AI, this might be a new method, a new way to evaluate models, or new evidence for or against a popular assumption.
Why do research? I like to consider each research project as an adventure where you have the freedom to explore problems that you are interested in under fewer constraints than a traditional group-based project work. Equally important, it also trains transferable skills: reading critically and efficiently, asking precise questions, and helping you to frame (potentially unbounded) problems to solve them.
Research is a viable career. In Singapore and overseas, people do research at universities, industry research labs, or at other (public) research institutes with very attractive compensations (yes, even more than a software engineer). The catch is that a number of these jobs require a PhD degree. Part of the motivation of EROP was to provide an early exposure to AI research for you to assess your interest. So, suppose you might be interested in these jobs, how do you work towards a PhD program? Get more research experience.
Building research experience: As an undergraduate, you can participate in UROP programmes, final-year projects, or simply emailing a professor (for a research assistant position) whose work you have specific interests in. We designed EROP to specifically prepare you for these opportunities.
For this week, we want you to experience the initial start of research: coming up with an idea, turning it into a proposal, and finally presenting it to a group of judges.
What is a Research Paper?
A research paper is the written record of a piece of research: the problem, what was done, the evidence, and what it means. Papers are published at conferences or in journals. In both cases, other researchers review the paper before it is accepted (peer review), and in AI the top conferences are where most new work first appears. Papers are how academics validate, archive and share discoveries with the scientific community. The Starting Points table in the reading advice lists some of the main AI venues.
Research Brief
EROP gives every team the same general research brief. The brief describes an area and its constraints, but it does not fix the problem. Each team reads the brief and comes up with its own problem statement from it. The brief may come from a supporting industry partner (to be confirmed), and it is on the Research Brief page. Read it before working through the examples below.
Breaking down the brief
When reading the research brief, it can contain two kinds of constraints. Explicit constraints (stated in the brief) and Implicit constraints that are not directly written down. In this week, we would try to identify the implicit constraints to make your proposed project tractable while fulfilling the explicit ones!
The worked example below uses a separate illustrative brief on health misinformation. It is in a different domain from the EROP brief, so you can practice the skill before turning to your own brief.
Example: Identifying Constraints
Brief: AI-Assisted Detection of Health Misinformation in Online Communities
Background: Health misinformation spreads rapidly on social media platforms and online forums. Singaporean health authorities are interested in scalable tools that can assist human moderators in identifying potentially misleading health claims in English and Singlish user-generated content.
Task: Develop or evaluate an AI-based approach to detect health misinformation in short-form text (e.g., forum posts and social media comments up to 280 characters). The solution must be explainable — moderators must understand why a piece of content was flagged.
Resources: The team has access to publicly available English health misinformation datasets. Any approach using local Singapore data must comply with PDPA. The solution must be evaluable within a 5-week research sprint.
Besides the details the brief states, check whether the resulting proposal is sensible. For instance, a brief that gives no budget does not imply that we can propose a project asking for $1M to finetune a model. This would not be realistic, and the proposal itself would not be strong.
Step 1 — Extract explicit constraints
Go through the brief sentence by sentence and pull out anything that directly limits what the project can do.
| Examples | Source in the brief |
|---|---|
| Short-form text only (≤ 280 chars) | “short-form text … up to 280 characters” |
| Explainability is required | “The solution must be explainable” |
| English and Singlish text | “English and Singlish user-generated content” |
| 5-week timeline | “5-week research sprint” |
| PDPA compliance for any local data | “must comply with PDPA” |
| Only publicly available datasets (unless PDPA-compliant) | “access to publicly available English … datasets” |
Step 2 — Infer implicit constraints
For each explicit constraint, ask what it means in practice.
Singlish adds NLP complexity. Most pre-trained language models are trained on standard English. Singlish mixes English with Malay, Hokkien and Tamil, and it is underrepresented in standard NLP benchmarks. An off-the-shelf model may not transfer well to it.
The health domain probably needs high precision. The brief does not say this, but flagging a correct health claim as misinformation (a false positive) could erode public trust. The system should be conservative about what it flags.
“Evaluable in 5 weeks” rules out training from scratch. There is not enough time to collect new data and pre-train a model, so fine-tuning or prompting an existing model is more realistic.
“Assist human moderators” implies a human in the loop. The system is a triage tool, not a fully autonomous one. This relaxes some accuracy requirements and raises a new question: what does the moderator need to make a decision?
Step 3 — Note what to emphasise in a problem statement
From these constraints, three points stand out for a problem statement:
- The Singlish challenge is distinctive and under-studied. It explains why existing work might not apply.
- The explainability requirement creates a research tension, since the most accurate models are often the least interpretable.
- The human-in-the-loop framing changes how “success” is defined. Accuracy alone does not capture it.
Take the brief on the Research Brief page and, as a team, produce:
- A table of explicit constraints with their source in the brief.
- A list of implicit constraints, each with a short justification.
- Two or three points to foreground in a problem statement, and why.
Reading Advice
Once a team has a problem in mind, most of its time goes into reading what others have already done about it. This hour covers how to efficiently read a research paper: deciding if it makes sense to read a paper, how to do so effectively, and what to take away from each one.
Components of a Paper
Most AI papers follow a similar layout:
- Abstract — the whole paper in one paragraph: problem, approach, main result.
- Introduction — the problem, why it matters, and what the paper contributes. The contribution is usually stated at the end of the introduction, often as a short list.
- Related work — how the paper positions itself against earlier work.
- Method — what the authors propose, in enough detail to reproduce it.
- Experiments — the setup, datasets, baselines and results, usually as tables and figures.
- Discussion and limitations — what the results mean and where they fall short. Limitations are usually found here.
- Conclusion — a short restatement of the contribution and future work.
- References and appendix — the works cited, and extra details, proofs and results that did not fit in the main text.
The Three-Pass Method
The three-pass method one efficient way to read a research article. It follows the following three steps (passes):
- Pass 1 — Overview. Consider and read the title, abstract, section headings, and conclusion. From here, we can already decide if we want to read further for this paper.
- Pass 2 — Method . Read the method and experiments carefully, skipping proofs and implementation details, but pay attention to the core ideas and the figures in these sections. If this paper’s method addresses your problem statement or is similar to your current ideas, proceed to Pass 3.
- Pass 3 — Details. If we reach this stage, this paper can be considered as a “baseline” (aka a potential competing method). In this case, read what is their specific contributions and claims — are they tested well? are their experiments sound? can we do better (implementation-wise)? Answering these questions can be the starting point of how you frame your own proposal.
Most papers stop after the first pass. For detailed tips, refer to the Reading Advice page.
Contribution, Evidence and Gap
Each key paper should give you four things:
- Contribution — what the paper adds. What can be done now that could not be done before?
- Main evidence — the table, figure or result that supports the contribution. A claim without evidence you can point to is not yet trustworthy.
- Limitations — where it falls short: assumptions, narrow evaluation, settings it does not cover.
- Gaps — what it leaves unaddressed. A gap that overlaps your problem is an opening for your own contribution.
Reading a paper on explainable misinformation detection:
- Contribution: a token-attribution method that lets a moderator see which words drove a “misinformation” flag. This is often found at the end of the Introduction section.
- Main evidence: a user study showing moderators trust it more than a black-box score. This is usually a table or figure in the Experiments section.
- Limitations: evaluated only on English Twitter data; the user study had 12 participants; no code-mixed text. This is often found at the Discussion section.
- Gaps (the opening): no one has tested whether token attributions stay meaningful for code-mixed Singlish, where the model may attend to tokens from the “wrong” language. A project on the illustrative brief could contribute here.
As a team, pick the single most central paper to your chosen problem (ask your mentor if you are unsure) and read it without AI assistance. Write down its contribution, its main evidence, its limitations, and at least one gap. Then phrase that gap as a one-sentence opening for your own research.
Research Project: Problem Selection
For the last two hours, each team works with its mentor to choose the problem it will take on for the week and to draft a rough outline of a solution. The rest of this section covers what that discussion needs.
Problem Statements
The problem statement justifies why the research is worth doing. In a paragraph or two, it answers three questions:
- What gap exists? — the open problem or underexplored area.
- Why does it matter? — the real-world or scientific motivation.
- What direction are you taking? — a high-level sense of the research approach.
A problem statement is not a solution statement. It doesn’t say “we will build X.” It says “Y is a challenge because of Z, and addressing Z matters because of W.”
Research Questions
Research questions (RQs) specify what the project will experiment on (e.g., modifying objective functions) to address the problem statement. The two RQs below come from the illustrative brief.
RQ 1 — Singlish-aware misinformation detection
Health misinformation classifiers are typically fine-tuned on standard-English corpora, so Singlish — with its code-switching, borrowed lexicon (Malay/Hokkien/Tamil), discourse particles (lah/leh/sia), and non-standard orthography — presents a token-level distribution shift the model never saw in training. How large is the performance gap (macro-F1, and the false-positive rate in particular) between English and Singlish inputs, and can cross-lingual transfer or parameter-efficient fine-tuning (e.g., LoRA/adapters) close it under a limited labelled-data and single-GPU budget?
RQ 2 — Explainability for health moderators
An accurate classifier is unhelpful if a moderator can’t tell why a post was flagged — and post-hoc attribution methods (LIME, SHAP, attention weights) are not guaranteed to be faithful to the model’s actual decision. Which explanation format — token-level highlights, retrieved counter-evidence, or a short natural-language rationale — most improves a moderator’s decision accuracy and review time, and how faithfully does each reflect the classifier’s true reasoning?
When comparing RQs, check whether they are distinct from each other, and which one is more tractable under the constraints found earlier.
Drafting a strong Problem Statement
The draft below is a problem statement for RQ 1. The notes after it explain the choices in it.
Online health misinformation poses a growing public health challenge in Singapore, where online communities frequently mix standard English with Singlish — a local creole not well-represented in existing NLP benchmarks. Most health misinformation detection models are trained on English-only corpora and may not generalise reliably to this code-mixed context, leaving human moderators without dependable AI support on Singapore-facing platforms. This project investigates the extent to which pre-trained language models can be adapted for Singlish health misinformation detection under constrained data and compute budgets, using cross-lingual transfer and lightweight fine-tuning as candidate approaches.
- It names a specific gap. “Misinformation is a problem” is too broad. “Existing models do not cover Singlish” is concrete.
- It does not commit to a solution. It “investigates … using X as candidate approaches”. If those approaches do not work, that is still a valid research outcome.
- It connects the technical gap to the human motivation. Singlish underrepresentation is the technical fact, and “leaving moderators without dependable AI support” is the human consequence. A problem statement needs both.
- It keeps the scope realistic. “Constrained data and compute budgets” reflects the 5-week timeline.
Using the constraints you pulled from the EROP brief, write two distinct research questions as a team. For each, write a sentence or two framing the core question. Then pick one, draft a 100–150 word problem statement like the one above, and sketch a rough outline of the approach with your mentor.
Checkpoint Deliverable: Day 1
At the end of Day 1, the team has chosen its problem from the research brief and has a draft solution outline: a problem statement covering the motivation, the gap and the general direction; one or two research questions; and a rough idea of the approach.