Physics Does Digital Optimization—which we call Onsager Computing— for Machine Learning, Control Theory, Backpropagation, etc.

Eli Yablonovitch
University of California, Berkeley (UC Berkeley)
Electrical Engineering

Optimization is vital to Engineering, Artificial Intelligence, and to many areas of Science. Mathematically, we usually employ steepest-descent, or other digital algorithms. But, Physics itself, performs optimizations in the normal course of dynamical evolution. Nature provides us with the following optimization principles:
1. The Principle of Least Action;
2. The Variational Principle of Quantum Mechanics;
3. The Principle of Minimum Entropy Generation;
4. The First Mode to Threshold method;
5. The Principle of Least Time;
6. The Adiabatic Evolution method;
7. Quantum & Classical Annealing
In effect, Physics can provide machines which solve digital optimization problems much faster than any digital computer. Of these physics principles, “Minimum Entropy Generation” in the form of bistable electrical or optical circuits is particularly adaptable toward offering digital Optimization. For example, we provide the electrical circuit which can address the challenging Ising problem, binary magnet energy minimization.
Since Onsager, 1930, introduced the Principle of Minimum Entropy Generation we call this Onsager Computing, as opposed to conventional Von Neumann Computing. Electrical Onsager Computers run ~10000 times faster have ~10000 times less energy-to-solution, than conventional machines.


Back to New Trends in Scientific Computing