A Foundry scientist trained neural networks to operate simulated fluctuating nanosystems in an energy-efficient way, converting information into heat or work.
Significance and Impact
The approach could be adapted for use with physical computing devices, in order to reduce energy consumption and waste.
- Used genetic algorithms to train neural-network feedback-control protocols for simulated fluctuating nanosystems
- Demonstrated procedure on model systems of a particle pulled through a viscous medium by an optical trap and an Ising model undergoing magnetization reversal
- When provided with feedback from the system, neural networks learned protocols to extract work or store heat