Recorded as part of the CFvW Colloquium on July 11, 2024
Machine Learning meets Quantum Mechanics - Prof. Dr. Roberto Giuntini, University of Cagliari und Technical University of Munich
Talk abstract:
Research in the broad area of pattern recognition, machine learning, and quantum computing has inspired new ideas about some important general problems that arise in several disciplines, including information theory (classical and quantum), logic, cognitive science and neuroscience, and philosophy.
One of the fundamental questions that these disciplines often face is the following: How are abstract concepts formed and recognized on the basis of
previous (natural or artificial) experiences? This problem has been studied, with a variety of methods and tools, both in the context of human intelligence and artificial intelligence.In this seminar, the problem will be addressed within the framework of machine learning and quantum computing.
Machine learning can be defined as the art and science of making computers learn from data how to solve problems (or recognize and classify new objects) without being explicitly programmed. Quantum computing describes the processing of information using tools based on the laws of quantum theory. Today, we are witnessing a dramatic explosion of data, and the problem of extracting and recognizing only "useful information" from these data is crucial but extremely resource consuming. On the other hand, quantum computing has shown that there exist quantum algorithms that allow a formidable acceleration in solving problems that, in their current state, would require exponential times. The realization of the so-called noisy intermediate-scale quantum (NISQ) computers is now a reality. Therefore, the combination of machine learning and quantum computing appears inevitable. This "marriage" is favored by the fact that one of the fundamental features of quantum theory is that it can deal with incomplete information in a particularly natural and efficient way, a feature that is of primary importance in machine learning. The approach that I will present in this seminar (called Quantum-Inspired Machine Learning) consists of formally translating the process of (supervised) classification of (classical) machine learning by using the formalism of quantum theory in such a way that the resulting classification algorithms can be implemented on non-necessarily quantum computers. In particular, I will address the problem of binary classification of classical datasets, presenting a classifier (called the Helstrom Quantum Classifier (HQC), based on the Helstrom protocol, which is used to distinguish between two quantum states (mathematically represented by density matrices). HQC acts on density matrices, which, in our model, encode the patterns of a classical dataset. Experimental benchmark results show that, in many cases, the accuracy of HQC is superior to that of many classical classifiers. Finally, we will show how the improvement in HQC performance is positively correlated with the increase in the number of "quantum copies" of each (encoded) classical pattern.
Explore our colloquium schedule on our website: [ Ссылка ]
