
Nutrition5k: Towards Automatic Nutritional Understanding of Generic Food
What the Nutrition5k food-image dataset contains, what its small human comparison shows, and why benchmark results do not establish consumer-app accuracy.
A closer look at the science behind nutrition.
Evidence-based articles exploring the science of nutrition, metabolism, and health technology.

What the Nutrition5k food-image dataset contains, what its small human comparison shows, and why benchmark results do not establish consumer-app accuracy.

How HEI-2020 and AHEI-2010 differ, what diet-quality cohort studies show, and why a score is not an individual diagnosis.

The four NOVA food groups, a controlled feeding trial, and the strengths and limits of evidence linking ultra-processed foods with health outcomes.

What EAR, RDA, AI, and UL mean, how Daily Values differ, and why a single day of estimated intake cannot diagnose nutrient deficiency.

A dated comparison of WHO, American Heart Association, and 2025-2030 US guidance on sugar, sodium, and saturated fat, with definitions and primary sources.

How sleep history, circadian timing, and sleep inertia enter an alertness model, and what the original papers do and do not validate.

What the SAFTE fatigue model represents, how sleep-restriction experiments inform it, and why its outputs are not direct measures of individual capacity.

What two meta-analyses found about dehydration and cognition, including corrected effect estimates and uncertainty about a universal 2% threshold.

What PREDICT 1 measured in UK and US adults, what individual metabolic variation means, and the limits of claims about personalized nutrition.

Randomized trials and a 2026 follow-up show why meal-timing results depend on the comparison, population, duration, and outcome.

What studies of glucose variability actually measured, why evidence from older adults with diabetes has limits, and what remains unproven about cognition.

Evidence on caffeine, alertness, sleep timing, and individual variation, without a universal optimal dose or a guaranteed safe limit.