ETD

Archivio digitale delle tesi discusse presso l'Università di Pisa

Tesi etd-09162021-090424


Tipo di tesi
Tesi di laurea magistrale
Autore
PELLICANO', MATTIA PAOLO
URN
etd-09162021-090424
Titolo
Artificial Intelligence for Open Quantum Systems and Quantum Neural Networks for AI
Dipartimento
INGEGNERIA CIVILE E INDUSTRIALE
Corso di studi
MATERIALS AND NANOTECHNOLOGY
Relatori
relatore La Rocca, Giuseppe Carlo
correlatore Ciuti, Cristiano
Parole chiave
  • open quantum systems
  • quantum computation
  • Quantum reservoir computing
Data inizio appello
01/10/2021
Consultabilità
Non consultabile
Data di rilascio
01/10/2061
Riassunto
In recent years, researchers are investing more and more resources in understanding to what extent quantum computing can improve the performance of classical Machine Learning algorithms. The basic idea of what they want to achieve is more efficient execution of those algorithms with high computational cost, employing quantum computing, translating the original classical logic within the quantum theory. Since the global volume of collected data continuously increases each year, and the algorithm architectures become increasingly sophisticated, the need to find innovative approaches to performing better is also overgrowing. This urgency is leading many research areas to evaluate more deeply the potential of quantum computers and quantum computation to speed up the previously stated Data Mining and Machine Learning algorithms, but not just that. Quantum computers could impact society by providing a new paradigm for privacy and security. They could help in all those calculations and simulations of complex quantum systems, from a single molecule to a large-scale chemical reaction, with obvious benefits for pharmaceuticals discovery and development. Many works conducted mainly by physicists and computer science experts have recently demonstrated the impressive computational power of quantum systems due to their intrinsic quantum nature. Today, we possess various examples of how quantum algorithms speed up over the best classical methods used on the same problem. Many experts guarantee that it is only a matter of time to have the technologies stable enough to test all the theoretical proposals.
In this context, even a branch of algorithms and architectures named Reservoir Computing takes on a quantum declination. Reservoir Computing refers to a set of methodologies for the design and training of that class of neural network called Recurrent Neural Networks. The term was introduced in 2007 to emphasize how these methodologies share the same basic fundamental idea, the separation between the recurrent (dynamic) part of the network, the reservoir, from the non-recurrent part, the readout. Even if it is currently a classical paradigm used as a Machine Learning model, Reservoir Computing, thanks to its feasible architecture, is being studied for more efficient hardware implementations in classical and quantum versions. A Quantum Reservoir is essentially a network of qubits that interact among themselves. The interaction, the link weight, between two qubits, or rather two nodes, is randomly generated, just as the classical counterpart. In many recent works about this topic, quantum reservoirs have been used for classical tasks, such as time series analysis or digit recognition, as well as few quantum tasks, like quantum state recognition and preparation. In this thesis, numerical simulations have been done to see if a bosonic quantum reservoir, subject to dissipation and interacting with a qubit, can perform a quantum gate on that qubit through this non-trivial and rich interaction. In other words, the tempting idea is to use these interactions to make the qubit evolve temporally until a specific desired state is reached, and the hope is to find a system used as a component in future quantum devices or computers.
The basic knowledge for understanding this work will be provided in the first two chapters, while we will expose the methods and results of the experiment conducted in the third chapter. The first chapter will serve as an overview of quantum mechanics. It will consist of a brief foundation course on quantum mechanics and quantum information topics, from the definition of quantum state optics, passing through the second quantization and Fock formalism, and arriving at some glimpses of open quantum systems theory. In the second chapter, Machine Learning and Neural Network will be briefly illustrated to accompany the reader to the concept of Reservoir Computing, which will be introduced along with Classical Reservoir Computing examples and recently studied aspects of Quantum Reservoir Computing. Finally, in the third chapter, the adopted theoretical model describing the system under consideration, a bosonic reservoir and a qubit, will be explained in detail and the methodology used to train the couplings between the qubit and the reservoir. The outcome resulting will be in the end commented.
File