QUBO Formulation for Sparse Sensor Placement for Classification

The demand for facial recognition technology has grown in various sectors over the past decade, but the need for efficient feature selection methods is crucial due to high-dimensional data complexity. This paper explores the potential of quantum computing for Sparse Sensor Placement optimisation (SSPO) in facial image classification. It studies a well known Filter Approach, based on statistical measures like Pearson correlation and Mutual Information, as it offers computational simplicity and speed. The proposed Quadratic Unconstrained Binary optimisation (QUBO) formulation for SSPO, inspired by the Quadratic Programming Feature Selection approach, aims to select a sparse set of relevant features while minimising redundancy. QUBO formulations can be solved by simulated annealing and by quantum annealing. Two experiments were conducted to compare the QUBO with a machine learning (ML) approach. The results showed that the QUBO approach, utilising simulated annealing, achieved an accuracy between random placed sensors and ML based sensors. The ML algorithm outperformed the QUBO approach, likely due to its ability to capture relevant features more effectively. The QUBO approach’s advantage lies in its much shorter running time. The study suggests potential improvements by using Mutual Information instead of Pearson correlation as a measure of feature relevance. Additionally, it highlights the limitations of quantum annealers’ current connectivity and the need for further advancements in quantum hardware.