Przemysław Sadowski
Institute of Theoretical and Applied Informatics, Polish Academy of Sciences
Przemysław Sadowski is an assistant professor at Institute of Theoretical and Applied Informatics, Polish Academy of Sciences. He is working on a range of topics within the field of quantum algorithms, currently focusing on quantum optimization and quantum machine learning.
Introduction
Recent progress in developing machine learning techniques that harness the power of quantum processing units is very dynamic, but the devices available in the near-term provide limited resources to harness the proposed methods. The main goal of this work is to consider ways to improve quantum distance-based classification in terms of needed resources and to develop new features that extend its usefulness as well as boost its efficiency.
Methods
First, we show that the resources used to implement the distance-based classifier can be significantly reduced. The quantum operations performed during classification apply only to the ancillary and feature vectors registers, and the rest of the information may be classical. We use quantum channel description and distinguish classical and quantum dynamics, obtaining a scheme with memory size constant with training set size.
Then, we use Open Quantum Walk Model to describe the classification, obtaining a scheme that allows distributed information and state recycling.
Results
We use classification scheme described by a quantum channel and perform Iris dataset analysis. We show that our algorithm allows to consider all 3 classes, all 4 features and all of the samples on a 5 qubit computer, instead of 2 classes, 2 features and 2 samples as in the related works (Schuld et al.).
Then, we show the classification behavior when the training information is specially distributed. In our toy model we consider agents arranged in a cycle with various class representatives arrangement (Figure 1).
Finally, we show that probability of correct classification increases if we do not discard the state at the end of the protocol, and use it for further analysis instead (Figure 2).
Discussion
The most basic consequence of the introduced hybrid classical-quantum method is parallel computation potential, a number of small QPUs can be used to obtain higher classification accuracy.
Moreover, joining quantum cryptographic protocols with distributed information classification could bring novel applications for QPUs.
Finally, we have shown that the classifier can achieve better accuracy when a state after the final measurement is preserved and processed. The foundations of the observed improvement will be an object of our further study.
