Research

Shared research skills

Study these before analyzing specialization papers: they teach how to turn interest into falsifiable questions, find the evidence behind claims, design fair experiments, report uncertainty and explain work at different lengths.

1. From interest to falsifiable research question

   Turning a broad interest into a precise question with a falsifiable hypothesis is the first research skill a supervisor or research interviewer probes. A good question names the system, the conditions, the metric, the comparison and the result that would prove you wrong.

   Not started3 h 30 min
2. Prior work, positioning and novelty

   Positioning means knowing the closest prior work well enough to state exactly what your contribution adds and under which assumptions. It decides whether an idea is research or engineering, and research interviewers routinely ask how an idea differs from the obvious related systems.

   Not started4 h
3. Reading papers and locating the evidence for claims

   Reading a paper well means reconstructing its problem, assumptions, mechanism and claims, and then finding the specific evidence behind each claim. It is the core skill behind paper discussions in research interviews, reading groups and design reviews.

   Not started5 h
4. Experimental design: controls, baselines and ablations

   Experimental design decides whether a result can be attributed to your idea rather than to tuning effort, hardware, workload choice or chance. Controls, fair baselines, ablations, sensitivity analysis and error analysis are what interviewers probe when they ask how you would evaluate something.

   Not started5 h
5. Statistical uncertainty, seeds and paired comparisons

   Performance and ML results are noisy: runs differ because of seeds, scheduling, caches, co-tenants and data order. Knowing what counts as an independent sample, how to pair comparisons and how to report a confidence interval is what makes a claimed improvement believable.

   Not started5 h
6. Reproducibility, provenance and artifacts

   Provenance is the record of exactly how every number and figure was produced: code version, configuration, data, environment and seeds. It lets you, a reviewer or a colleague regenerate results, and artifact evaluation and ML checklists formalize it.

   Not started4 h
7. Benchmark realism and external validity

   A benchmark result matters only if its workload, environment and metric resemble the setting you claim to improve. This topic covers synthetic versus trace-driven workloads, load-generator design, hidden confounders and how far results generalize in systems, cloud, networking and ML evaluation.

   Not started4 h
8. Negative results, failed reproductions and revising hypotheses

   Many research attempts produce results that do not support the original hypothesis. Ruling out bugs and low power, documenting failed reproductions precisely and revising hypotheses honestly is a mark of research maturity that supervisors and research interviewers look for.

   Not started2 h 30 min
9. Correctness guarantees versus empirical evidence

   Papers and design documents mix two kinds of claims: guarantees that follow from a proof, invariant or model-checked specification under stated assumptions, and empirical results that hold only for the measured setup. Keeping them apart is essential when discussing consensus, transactions, isolation or performance.

   Not started3 h 30 min
10. Explaining a paper or project at 2, 10 and 45 minutes

    Research conversations require explaining the same work at very different depths: a two-minute summary, a ten-minute talk and a longer technical discussion driven by questions. Templates for each format, honest attribution and accurately stated evidence make an explanation credible in thesis and research interviews.

    Not started4 h
11. Research coding: prototypes, reference checks and instrumentation

    Research code must be fast to change and trustworthy in its numbers. Readable prototypes, numerical care, reference implementations with differential tests and logging of every configuration are what research-engineering interviews and real projects check.

    Not started5 h
12. Research internship and Master's thesis interview preparation

    Thesis and research-internship interviews assess research interests, a plausible problem, motivation for the specific lab or team, knowledge of relevant papers, feasibility and how you explain prior technical work. Academic labs do not share a standardized interview loop, so preparation is a checklist and a question bank rather than a drilled format.

    Not started5 h