The physics of learning machines

Abstract

ABSTRACT A learning machine, like all machines, is an open system driven far from thermal equilibrium by access to a low entropy source of free energy. We discuss the connection between machines that learn, with low probability of error, and the optimal use of thermodynamic resources for both classical and quantum machines. Both fixed point and spiking perceptrons are discussed in the context of possible physical implementations. An example of a single photon quantum kernel evaluation illustrates the important role for quantum coherence in data representation. Machine learning algorithms, implemented on conventional complementary metal oxide semiconductor (CMOS) devices, currently consume large amounts of energy. By focussing on the physical constraints of learning machines rather than algorithms, we suggest that a more efficient means of implementing learning may be possible based on quantum switches operating at very low power. Single photon kernel evaluation is an example of the energy efficiency that might be possible.