** TALK CANCELLED ** The physics of learning machines

Our planet is full of learning machines which are optimized by evolution and experience to exploit limited thermodynamic resources. What physical principles underly this and how might they be used to make efficient artificial learning agents? Learning machines like all machines are open systems driven from thermal equilibrium. As far as we know, there are no quantum learning machines in nature. How can we define a class of learning machines driven by quantum effects?

In this talk, I will introduce the topic of learning machines, and will give a brief history of the subject. I will review the connection between learning machines, the physical devices that learn, and mathematical machine learning algorithms. Optimised learners minimise the power dissipated as they minimize error rate. This is true for both classical (thermal) and quantum learning machines. I will discuss the formulation of stochastic relations of elementary learning machines and the average error so that they can be extended to the fully quantum regime where the temperature is very low. This enables us to define a class of quantum learning machines that are driven by quantum noise, such as quantum tunneling noise or spontaneous emission noise, rather than thermal noise.