Models May Learn Who Generated the Data: Robotic Surgery Study Shows AUC Drops 0.89 to 0.57
bravo_abad · x · 2026-09-30
Ueki et al. show models can pick up habits of the data generator: analyzing 98 robotic-surgery procedures by 16 surgeons via 25 features from 3D hand trajectories, a random forest hit AUC 0.89 with random splits but only 0.57 when each test surgeon's procedures were excluded from training — a warning that random train-test splits can overstate model usefulness wherever the data source correlates with the label.
Related event: Study Finds Models Can Learn the Habits of Data Generators(2 posts)→
More from Embodied
- a16z leads $7.5M pre-seed for Aleph Surgery to build a scalable robotic surgeon — a16z · 2026-09-30
- Shield AI names Kansas and Washington as flight test and production homes for X-BAT fighter jet — BrettKrieger12 · 2026-09-30
- YC hosts SF evening on physical AI with Physical Intelligence co-founder — ycombinator · 2026-09-30
- Reddit user demos 'robo-dots' robot project in new video — etherd0t · 2026-09-30
- DynaRobotics' full-stack approach to workflow automation wins industry praise — chris_j_paxton · 2026-09-30
- Flexiv arm localizes touch using only joint torque sensors, no skin sensors — chris_j_paxton · 2026-09-30