Connectionists: Research Associate Position in Quantum Algorithms for Cognitive Healthcare

[Mark Herbster]

Research Associate Position in Quantum Algorithms for Cognitive Healthcare
University College London 
Department of Computer Science

The aim of this 1 year Research Assistant position is to design and analyse efficient algorithms for quantum machine learning.  We are interested in both algorithms that may be implemented with currently envisioned quantum computing devices as well as the efficient implementation on classical computers of quantum-inspired algorithms.  The postdoc will be carried out in collaboration with Siemens UK. Two research areas that will be considered for the grant are the following, Quantum Machine Learning for regularised risk minimisation and for Adiabatic Annealing.  The project will be supervised by Mark Herbster (m.herbster at cs.ucl.ac.uk) and Simone Severini (s.severini at ucl.ac.uk) at UCL and Peter Mountney (peter.mountney at siemens.com) at Siemens UK.

To apply see http://bit.ly/2lKKqMq and for details see below.

Quantum Machine Learning for regularised risk minimisation.

Recently a new family of quantum algorithms has emerged, based on the HHL (Quantum algorithm for solving linear systems of equations (2008)). HHL can approximately solve a system of linear equations with potentially exponential speedups over classical approaches. These algorithms are driving a revolution in quantum machine learning by extending HHL or using it as a subroutine. A common framework for machine learning is regularised risk minimisation. Given a dataset we should select a hypothesis which trades off "fit to the data" and "complexity" of the hypothesis. Standard examples of regularisers for linear prediction include the 2-norm, 1-norm and maximum entropy. A number of problems such as object detection can be framed as "matrix learning" or non-negative matrix factorization including object detection and tracking. The aim of this work is to apply and extend the quantum algorithms based on HHL to matrix learning via spectral regularization. In this work we will develop an approach to automatically detect and track surgical tools in medical images. We will investigate the feasibility of this in software.


Quantum Machine Learning via Adiabatic Annealing. 

Early approaches to quantum machine learning, such as QBoost, focused on using the quantum annealer as an optimizer for quadratic unconstrained binary optimization problems with limited success. However, recent work has re-imagined devices for quantum annealing as a quantum sampler and physical implementation of a Restricted Boltzmann Machine. The Boltzmann machine is one of the fundamental techniques powering the deep learning revolution. Initial work on Quantum Boltzmann Machines for training Deep Neural Networks and Deep Belief Networks has shown comparable or superior results with significantly fewer iterations of generative training.  Deep Neural Networks will play a large role in the Cognitive OR of the future by helping to detect, model and track everything form anatomy and tools to people and equipment.  We will also investigate the feasibility of this approach to detect tools in medical images.