Periodic Reporting for period 1 - QuaIL (Quantum Information Learning)
Période du rapport: 2020-01-01 au 2021-12-31
Have you ever wondered how will space-faring humans make video calls with their friends on Earth, stream their favourite TV series or communicate important messages?
Optical communications technology uses light propagating in free space and optical fiber to transmit data for telecommunications and networking. When the communication takes place over very long distances the light signals get extremely damped and the message they transmit becomes very difficult to read. Actually, the signals are so weak that the message cannot be read without errors, even when all the technical conditions are perfectly controlled and no external sources of noise are present. This is the realm where Quantum Physics enters the game: according to Heisenberg’s Uncertainty Principle, the position and velocity of a particle cannot be measured perfectly at the same time. The same principle applies to properties of the photons, when we use them to transfer a message over very long distances such as in optical-fiber, satellite or deep-space links.
This inherent quantum noise constitutes a new challenge for the manipulation of information that cannot be solved with classical methods. Luckily, quantum physics also offers us new instruments to face this kind of challenges.
The QuaIL project investigated how to build reliable quantum measurement devices, or decoders, that would allow us to increase current communication rates. These fundamentally deviate from classical measurement devices since they require to perform collective measurements of multiple signals at once. The project was based on the idea of combining basic building blocks of quantum optical circuits into a complex architecture, and optimize its structure with the help of machine-learning.
The project implemented several algorithms for the optimization and control of quantum devices, applying them to the specific problem of finding an optical decoder capable of reading information written into optical coherent-state signals with minimum error.
In particular, using reinforcement-learning methods, we implemented algorithms that were able to reach an exciting result: the calibration of an unknown quantum device by trial-and-error, based only on making an efficient use of limited communication experiments.
Furthermore, we were able to discover decoders of medium-sized messages that increase up to 3 times the success rate with respect to single-signal receivers and are only 7%-away from the theoretically optimal decoder. Importantly, the decoders we found can be implemented with state-of-the-art optical technology.
Finally, the project identified two communication settings of practical relevance where a quantum advantage can be guaranteed: communication in the absence of a phase reference; and communication with optical amplifiers, where we showed that up to 50% of the amplification energy cost can be spared by using a quantum optimal measurement.
Being aware of the growing interest that the general public starts to have in quantum technologies, we also realized a popular science comic on the subject, with comic writer Martoz, titled “Erwyn and the photonic flea”.
Optical communications technology uses light propagating in free space and optical fiber to transmit data for telecommunications and networking. When the communication takes place over very long distances the light signals get extremely damped and the message they transmit becomes very difficult to read. Actually, the signals are so weak that the message cannot be read without errors, even when all the technical conditions are perfectly controlled and no external sources of noise are present. This is the realm where Quantum Physics enters the game: according to Heisenberg’s Uncertainty Principle, the position and velocity of a particle cannot be measured perfectly at the same time. The same principle applies to properties of the photons, when we use them to transfer a message over very long distances such as in optical-fiber, satellite or deep-space links.
This inherent quantum noise constitutes a new challenge for the manipulation of information that cannot be solved with classical methods. Luckily, quantum physics also offers us new instruments to face this kind of challenges.
The QuaIL project investigated how to build reliable quantum measurement devices, or decoders, that would allow us to increase current communication rates. These fundamentally deviate from classical measurement devices since they require to perform collective measurements of multiple signals at once. The project was based on the idea of combining basic building blocks of quantum optical circuits into a complex architecture, and optimize its structure with the help of machine-learning.
The project implemented several algorithms for the optimization and control of quantum devices, applying them to the specific problem of finding an optical decoder capable of reading information written into optical coherent-state signals with minimum error.
In particular, using reinforcement-learning methods, we implemented algorithms that were able to reach an exciting result: the calibration of an unknown quantum device by trial-and-error, based only on making an efficient use of limited communication experiments.
Furthermore, we were able to discover decoders of medium-sized messages that increase up to 3 times the success rate with respect to single-signal receivers and are only 7%-away from the theoretically optimal decoder. Importantly, the decoders we found can be implemented with state-of-the-art optical technology.
Finally, the project identified two communication settings of practical relevance where a quantum advantage can be guaranteed: communication in the absence of a phase reference; and communication with optical amplifiers, where we showed that up to 50% of the amplification energy cost can be spared by using a quantum optimal measurement.
Being aware of the growing interest that the general public starts to have in quantum technologies, we also realized a popular science comic on the subject, with comic writer Martoz, titled “Erwyn and the photonic flea”.
The work was structured along two main lines.
Numerical efforts were concentrated on the implementation and testing of several families of algorithms for the discovery of minimum-error optical decoders.
For the optimization of adaptive architectures, where the signals are partially measured and then, based on the outcome, updated and measured again, in a recursive way, we found that reinforcement-learning methods such as Q-learning and Thompson sampling were able to find near-optimal setups and exhibited good real-time performance while training. This family of algorithms will be worth investigating further in future research, since it can be applied to train real quantum computers and devices based only on experimental data. A scalable use of these resources will however require the introduction of deep-learning methodologies.
For the optimization multi-signal architectures we found useful instead to work with simpler optimization methods, which are based on the knowledge of the architecture itself. In order to encounter a good device setup for discriminating multiple messages sent through the communication channel, we trained the decoder on random messages by minimizing the decoding error probability. We discovered near-optimal decoder setups for random and linear codes, advancing the state of the art.
Theoretical efforts were focused on the study of specific communication settings where a quantum advantage is obtained by a quantum encoder or decoder. In particular we found that the fundamental quantum effect known as optical squeezing can increase the communication rate on dephasing channels, that describe a variety of practical situations, with respect to standard communication methods. Moreover, we discovered a striking reduction of energy consumption in optically amplified communication lines using a quantum decoder. Finally, we established a theoretical learning framework for optical quantum circuits.
The results have been and will be presented at international conferences and workshops and will allow for further exploitation via theoretical study and experimental tests of the discovered decoders.
Numerical efforts were concentrated on the implementation and testing of several families of algorithms for the discovery of minimum-error optical decoders.
For the optimization of adaptive architectures, where the signals are partially measured and then, based on the outcome, updated and measured again, in a recursive way, we found that reinforcement-learning methods such as Q-learning and Thompson sampling were able to find near-optimal setups and exhibited good real-time performance while training. This family of algorithms will be worth investigating further in future research, since it can be applied to train real quantum computers and devices based only on experimental data. A scalable use of these resources will however require the introduction of deep-learning methodologies.
For the optimization multi-signal architectures we found useful instead to work with simpler optimization methods, which are based on the knowledge of the architecture itself. In order to encounter a good device setup for discriminating multiple messages sent through the communication channel, we trained the decoder on random messages by minimizing the decoding error probability. We discovered near-optimal decoder setups for random and linear codes, advancing the state of the art.
Theoretical efforts were focused on the study of specific communication settings where a quantum advantage is obtained by a quantum encoder or decoder. In particular we found that the fundamental quantum effect known as optical squeezing can increase the communication rate on dephasing channels, that describe a variety of practical situations, with respect to standard communication methods. Moreover, we discovered a striking reduction of energy consumption in optically amplified communication lines using a quantum decoder. Finally, we established a theoretical learning framework for optical quantum circuits.
The results have been and will be presented at international conferences and workshops and will allow for further exploitation via theoretical study and experimental tests of the discovered decoders.
We advanced the state-of-the-art of quantum optical decoders, providing, for the first time to our knowledge, a systematic study of practically realizable near-optimal decoders for small/medium-sized coherent-state codes. In doing so, we introduced a family of machine-learning-inspired algorithms for the manipulation of quantum devices that be of independent interest.
The control algorithms we developed constitute a step towards an efficient integration of classical and quantum information processing devices, an essential step in the large-scale application of quantum-assisted technologies. On the other hand, the decoder setups we discovered might lead in the future to attain higher communication rates and reduce energy consumption in long-distance communication on optical-fiber and deep-space links.
Our popular science comic will lead to a better understanding of the power and limits of quantum technologies, whose actual capabilities can be easily misunderstood and should be more precisely stated. It will also act as a friendly entry-point to the world of quantum mechanics and its applications.
The control algorithms we developed constitute a step towards an efficient integration of classical and quantum information processing devices, an essential step in the large-scale application of quantum-assisted technologies. On the other hand, the decoder setups we discovered might lead in the future to attain higher communication rates and reduce energy consumption in long-distance communication on optical-fiber and deep-space links.
Our popular science comic will lead to a better understanding of the power and limits of quantum technologies, whose actual capabilities can be easily misunderstood and should be more precisely stated. It will also act as a friendly entry-point to the world of quantum mechanics and its applications.