Physicist Andrea Liu: Simulating Brains with Spring Networks for Ultra-Low-Power Local Learning
量子位 · wechat · 2026-08-19
University of Pennsylvania physicist Andrea Liu discussed with Charles Zhang the jamming transition in soft matter physics and its applications in brain-like intelligent materials. Liu explained how disordered solids gain rigidity and shared how her team maps mechanical spring networks to neural networks, achieving allosteric control by adjusting spring parameters. The discussion highlighted using physical constraints (like energy minimization) to replace global gradient computation, proposing an update rule relying solely on local information to train networks. This mechanism mimics brain operations and consumes significantly less energy than standard AI tasks.
More from Embodied
- China adopts human-in-the-loop convoys: lead truck driven, self-driving followers — rohanpaul_ai · 2026-08-19
- Nomura Assigns Unitree Robotics 25x 2027 P/S, $370 Target Price — zephyr_z9 · 2026-08-19
- Robert Scoble: my robot is driving me and my wife around San Francisco — Scobleizer · 2026-08-19
- AI in Pharma: Precision Automation Fills 200 Bottles Per Minute — Olivier__OG · 2026-08-19
- Humanoid robot spotted walking on San Francisco streets — davidpattersonx · 2026-08-19
- CoRL 2026 to host workshop on continually self-improving robots — chris_j_paxton · 2026-08-19