RobPicker in Nature Methods: meta-learning fixes noisy labels, protein detection 0.24 to 0.47
bravo_abad · x · 2026-11-09
Liang et al. publish RobPicker in Nature Methods for robust identification of macromolecules in cryo-electron tomograms. Small protein complexes are hard to spot and hard to annotate, so emphasizing mistakes can backfire on wrong labels. RobPicker uses a small carefully annotated reference set and meta-learning to adjust both training labels and per-example weights, rewarded by detection gains on the clean set — lifting ATP synthase detection in an algal dataset from 0.24 to 0.47.
More from Research
- USC paper shows AI agents leak secrets via the "mosaic effect" without reading classified files — 2C_ornot2C · 2026-11-09
- Tsinghua & Moonshot's TCSExam: GPT-6 Astra proves 70% of theorems, solves 0 open conjectures — rohanpaul_ai · 2026-11-09
- TCSExam benchmark: GPT-6 Astra tops Full at 71.9, Opus 5.5 leads Lite at 72.3 — rohanpaul_ai · 2026-11-09
- Meta teases entry into AI-for-science race, joining frontier labs push — kimmonismus · 2026-11-09
- Study: 3g Daily Creatine for 4 Weeks Improved Cognition in Women, Most in Older Adults — 2C_ornot2C · 2026-11-09
- Brynjolfsson, Korinek, Agrawal publish Economics of Transformative AI research agenda in CACM — erikbryn · 2026-11-09