Field notes
Honest notes on how research really gets done.
Practical guides for PhD students and researchers, from the first year to the accepted paper, and clear explanations of trustworthy AI. Written from my own work as a researcher, a reviewer, and a teacher of research methodology.
Series 01
PhD basics
Start here if you are beginning a doctorate, or wish someone had told you this earlier.
Your first year of a PhD: what it really is, and how to use it well
What a doctorate actually trains you to do, the three jobs of year one, and the habits that make the rest of the PhD easier.
7 min readPhD basicsHow to choose a PhD research question that holds up to scrutiny
Turn a vague topic into a question that is new, answerable, and defensible, with five tests and a four step method.
8 min readPhD basicsHow to read a research paper efficiently: the three pass method for PhD students
The three pass method, a reviewer's checklist for judging evidence, and a note template that turns reading into writing.
8 min readSeries 02
Trustworthy AI made clear
What it means for a model to be reliable, explained for researchers and clinicians who use these systems.
Uncertainty quantification explained for researchers
A plain map of uncertainty quantification: what it is, the main methods and their cost, how to test them, and what to report. Free PDF kit included.
9 min readTrustworthy AI made clearA heatmap is not a confidence score: why medical AI needs explainability and uncertainty
Explainability shows where a model looked. Uncertainty says how sure it is. Why medical imaging AI needs both, and what to check in a paper.
6 min readTrustworthy AI made clearAleatoric vs epistemic uncertainty: what a model does not know, and why
Two kinds of "I am not sure": noise in the data, and gaps in the model. Plain examples, how to estimate each, and what to do with them.
8 min readTrustworthy AI made clearSelective prediction: when a medical AI should say "I am not sure"
Letting a model abstain on uncertain cases. How risk and coverage trade off, and six checks before you claim it works.
8 min readTrustworthy AI made clearWhy a model's confidence is often wrong: calibration and ECE explained
Accuracy says how often a model is right. Calibration says whether its confidence can be believed. How to measure it, and what to report.
8 min readSeries 03
Research in the age of AI
How to use AI tools in research without giving up the judgment that makes the work hold up.
Series 04
Getting papers accepted
How peer review really works, and how to answer reviewers so your paper gets to accepted.
Free downloads, no sign-up.
The checklist, the rebuttal template and the UQ kit are free to download. Follow new field notes with the RSS feed.