ANÁLISE DE COMPORTAMENTO DO SVM QUÂNTICO

Abstract

Quantum computing is considered a possible substitute for classic computing in certain tasks and problems, that leads to the propositions of quantum versions of conventional algorithms and methods, and that’s something that requires a lot of research yet. In context, this work’s main goal is to verify the accuracy of a quantum Support Vector Machine (SVM) for binary classification and compare it with its classic counterpart, all aiming to explore its inner workings, verifying  its limitations and possible parameters that can be modified. For such a task, unique datasets of differing structures and values were generated, all with their own levels of quality. The classical SVM was applied to each dataset and the data about execution and results were extracted. After that the quantum SVM was put together, implemented using almost exclusively the quantum circuits, and once again it was put to classify the data in each datasets, this time with a simulated quantum computer and later using a real one. After both executions, their data regarding operation and results were also extracted, with the intent to compare them with the classical SVM, these numerical results were the F1-Score, Recall Score, ROC curve and McNemar’s test. After analysis and premature conclusions, the focus became the code and structure of the QSVM, not only to distinguish the datasets that had higher compatibility, as well as hyperparameters that could be adjusted, or structures to be removed, added or altered.