Quantum Limits of Knowledge 2021 Wednesday 31st March, 2021 General purpose analogue machine learning has a history that predates digital computing with the first important results published by Shannon in 1941. In this talk I discuss the thermodynamic constraints on analogue machine learning based on Hopf bifurcations in both classical and quantum systems. This makes clocks central to machine learning. In the classical case dissipation and thermal noise are essential to learning. In the quantum case learning can proceed at zero temperature due entirely to quantum noise in the form of spontaneous emission and tunnelling and new modes of 12 operation emerge based on using quantum sensors and actuators. Coherent quantum feedback offers uniquely quantum protocols not based on measurement feedback.