Summary
Quantum neural networks are a young research field, that has been rapidly expanding due to their potential to attain revolutionary computing capacities and the possibility to learn on quantum data, inaccessible to classical computers. However, despite impressive proof-of-concept results, currently existing approaches that rely on sparsely coupled qubits, are not scalable to network sizes and connectivities with tunable weights required for state-of-the art tasks. In qDynnet, I will adopt a completely new and unexplored approach that uses parametrically coupled superconducting quantum oscillators instead of physically coupled qubits, that will allow me to obtain quantum neural networks of unprecedented size, connectivity and tunability. To do this, I will shift the paradigm by implementing neurons as basis states of dynamically coupled multi-level quantum oscillators, and connections between neurons as transition rates obtained through different dynamical processes such as parametric coupling, resonant drives and dissipation. I will implement experimentally quantum neural network architectures that were only simulated until now and use them to demonstrate data classification with basis state neurons. In order to go towards more complex tasks, I will use parametric coupling to introduce tunable connections between neurons. I will develop new training methods that will allow me to tune connections in such dynamical quantum neural networks and use them to demonstrate learning to recognize quantum states. I will develop circuit geometries that will be scalable to large quantum neural networks with millions of neurons and tunable connections. The qDynnet project will provide understanding of physics, and methods for dynamical coupling and training, that will have a broad impact across quantum computing fields and serve as a foundation for a whole new family of large-scale dynamical quantum neural networks.
Unfold all
/
Fold all
More information & hyperlinks
| Web resources: | https://cordis.europa.eu/project/id/101076898 |
| Start date: | 01-03-2023 |
| End date: | 29-02-2028 |
| Total budget - Public funding: | 1 497 536,00 Euro - 1 497 536,00 Euro |
Cordis data
Original description
Quantum neural networks are a young research field, that has been rapidly expanding due to their potential to attain revolutionary computing capacities and the possibility to learn on quantum data, inaccessible to classical computers. However, despite impressive proof-of-concept results, currently existing approaches that rely on sparsely coupled qubits, are not scalable to network sizes and connectivities with tunable weights required for state-of-the art tasks. In qDynnet, I will adopt a completely new and unexplored approach that uses parametrically coupled superconducting quantum oscillators instead of physically coupled qubits, that will allow me to obtain quantum neural networks of unprecedented size, connectivity and tunability. To do this, I will shift the paradigm by implementing neurons as basis states of dynamically coupled multi-level quantum oscillators, and connections between neurons as transition rates obtained through different dynamical processes such as parametric coupling, resonant drives and dissipation. I will implement experimentally quantum neural network architectures that were only simulated until now and use them to demonstrate data classification with basis state neurons. In order to go towards more complex tasks, I will use parametric coupling to introduce tunable connections between neurons. I will develop new training methods that will allow me to tune connections in such dynamical quantum neural networks and use them to demonstrate learning to recognize quantum states. I will develop circuit geometries that will be scalable to large quantum neural networks with millions of neurons and tunable connections. The qDynnet project will provide understanding of physics, and methods for dynamical coupling and training, that will have a broad impact across quantum computing fields and serve as a foundation for a whole new family of large-scale dynamical quantum neural networks.Status
SIGNEDCall topic
ERC-2022-STGUpdate Date
09-02-2023
Images
No images available.
Geographical location(s)
Search
