Quantum learning for large dimensional quantum systems

Project description

Statistical inference and learning play an increasing role in Quantum Engineering and Quantum Metrology. The efficient statistical reconstruction of quantum states is a crucial enabling tool for current quantum engineering experiments in which multiple qubits can be prepared in exotic states. However, standard estimation methods such as maximum likelihood become practically unfeasible for systems of merely 10 qubits, due to the exponential growth in size of the Hilbert space.

The aim of this project is to develop mathematical theory and investigate new methods for learning quantum states of large dimensional quantum systems. This stems from ongoing collaborations with Theo Kypraios and Ian Dryden (Statistics group, Nottingham), Cristina Butucea (Univesite Paris Est), Michael Nussbaum (Cornell), Jonas Kahn (Toulouse) and Richard Kueng (Caltech). In [1,2] we proposed and analysed faster estimation methods with close to optimal accuracy. The first goal is to better understand the behaviour of the estimators with respect to different measurement scenarios. Next, we would like to equip them with reliable confidence regions (error bars) which are crucial for experimental applications. Going beyond "full state tomography" new methods are needed which are able to "learn" the structure of the quantum state by making use of prior information encoded in physically relevant low dimensional models. Possible directions to be explored include models based on matrix product states, neural networks, quantum time series, compressed sensing [3] and the study of the asymptotical structure of the statistical models [4]. The project will involve both theoretical and computational work at the overlap between quantum information theory and modern statistical inference.