In small-sample science, regularized regression and random forests often beat deep learning
bravo_abad · x · 2026-09-08
The most useful ML model in a scientific lab is often not a neural network. A biologist with 50 patient samples, a chemist with 100 compounds, or a physicist with a few hundred expensive measurements works in a very different regime than the datasets that made deep learning famous. In that regime, regularized regression, random forests, SVMs or Gaussian processes can be better tools — especially when uncertainty and honest validation matter as much as raw accuracy. This is a core idea of the author's new textbook "Inteligencia artificial para estudiantes de ciencias" (Ediciones Pirámide, 2026).
Related event: In Small-Data Science, Classic ML Often Beats Deep Learning(2 posts)→
More from Research
- Mathematician argues papers should shift from proving conjectures to sharing insight — tak3sh8 · 2026-09-08
- Niantic's AutoCompass trains accurate visual localization from noisy GPS labels — ducha_aiki · 2026-09-08
- Apex unveils Astra, an automated AI research system aimed at AI improving AI — thetripathi58 · 2026-09-08
- More Ukrainian teams join CV research map with USDZ-prediction approach to object alignment and animation — ducha_aiki · 2026-09-08
- JustRL II adds token-level critic to GRPO for 128k long-CoT RL, lifting 2B model AIME25 from 61 to 81 — zibuyu9 · 2026-09-08
- Helsinki's Luigi Acerbi recruiting PhD/postdoc on amortized probabilistic ML for decision-making — LucaAmb · 2026-09-08