Dongshu showed that learning algorithms based on error backpropagation are compatible with unsupervised learning. All you have to do is to define an “unsupervised target”. Preprint coming soon!
In July 2024, with the help of Bluefors engineers, we have installed our first dilution refrigerator and started measuring superconducting circuits in our lab!
Quantum neural networks have the potential to achieve unprecedented computational capabilities, as well as to efficiently recognize quantum states, a task inaccessible to classical computers. However, existing approaches that rely on their implementation with qubits are limited by the latter’s poor connectivity.
Danijela Marković’s project qDynnet takes a new approach, which uses parametrically coupled quantum oscillators instead of physically coupled qubits. This will allow for realization of quantum neural networks of unprecedented size, connectivity and tunability. For this, neurons are implemented as ground states of a set of coupled quantum oscillators, and connections between neurons as transitions between these states. In qDynnet project these networks will be experimentally realized with superconducting circuits and used to demonstrate automatic recognition of quantum states.
The qDynnet project will provide the understanding of physics of dynamical connections, and develop new dynamical learning methods, that will serve as a foundation for a whole new family of dynamical quantum networks.
Julie Grollier receives the IEEE Magnetics Scociety Mid-Career Award for her contributions to the development of spintronic devices and their use in neuromorphic computing. The ceremony will take place in Sendai in Japan, during the 2023 Intermag conference.
Jérémie Laydevant has defended his PhD thesis on 20 October 2022. He will be joining the Peter McMahon group at Cornell for his postdoc. Congratulations Jérémie!
Our preprint on quantum reservoir computing with superconducting resonators is out! We show in simulations that quantum reservoir can be implemented on a superconducting circuit called a Josephson mixer, and that it can solve non-linearly separable machine learning tasks. We show that a smaller number of such quantum neurons can achieve the same performance as a larger number of classical neurons.
J. Dudas, E. Plouet, A. Mizrahi, J. Grollier, & D. Marković, Quantum reservoir neural network implementation on a Josephson mixer. [arXiv]
Neuromorphic physics team is a part of the CNRS/Thales laboratory, associated with University Paris-Saclay. Our main research interest is neuromorphic computing with nanodevices, quantum neuromorphic computing and neuromorphic algorithms.
We are currently looking for post-docs, contact us if you are interested!