Quantum Machine Learning: Quantum Information Encoding and the Compressed Quantum Artificial Neuron (McCulloch-Pitts Unit)

The study of quantum computation, which aims at information processing via the manipulation of quantum matter, is driven, on top of curiosity, by the expectation for a new computational scheme that is much faster than the traditional classical one. In recent years there is a growing effort to harness the power of quantum computation to solving problems in machine learning and vice versa. My research takes part in this effort.

A central question in information processing is: what is the most desirable way to encode the information in the first place? In classical computation, for example, a bit is represented by levels of electric voltages. Similarly, the quantum bit (qubit) is often encoded via quantum degree of freedom, such as energy or spin. Yet, in quantum computation, there are infinitely many conceivable possibilities to encode information - by exploiting the superposition structure of the quantum state – most of which are overlooked. Such encodings are the subject of this talk.

The talk will begin with an introduction to quantum computation and quantum machine learning. Then, in the second part of the talk, I will present two alternatives for information encoding on quantum states: one that is quite uncommon and another that is entirely new. I will further show how such encoding can be used to simulate an exponentially compressed quantum McCulloch-Pitts unit, the most primitive artificial neural network. Throughout the talk, I will give a brief overview of the quantum machine learning research done in my group. I will describe the potential of such research, but - just as important - mention what should be taken with a grain of salt.

Remark: the talk is self-contained and intended for non-experts. Engineering and exact sciences students, including those with no background in neither quantum mechanics nor machine learning, are very welcome to join.