#1 out of 3
science1d ago
Scientists have found foam bubbles follow the same maths as AI
- Penn researchers show wet foam keeps wandering among many nearly equal states, not settling on one final arrangement.
- The study links foam dynamics to gradient-descent style optimization used in deep learning for better generalization.
- Researchers say flattening the loss landscape helps models generalize, a concept now seen in foam behavior.
- The findings may apply to other adaptive structures, like cellular cytoskeleton, to stay organized while changing.
- Foam is not settling into one best state, but remains dynamic while appearing stable to an observer.
- The research draws on the 2025 PNAS paper Slow relaxation and landscape-driven dynamics in viscous ripening foams.
- The team expects the maths to guide adaptive materials that keep shape yet respond to surroundings.
- The work compares foam evolution with AI training to stay in flat regions rather than deepest valleys.
- The study uses simulations of wet foam to reveal the dynamic behavior described above.
Vote 0


