Machine learning and condensed matter
I am interested how machine learning and condensed matter insights can be useful in quantum information. We combined two challenges important for further scaling of quantum devices: error correction and device verification. We introduced a new scalable approach to quantum error correction implemented on Hamiltonian level (Phys. Rev. Research 1, 033092 (2019)). We also created unsupervised algorithm that helps to determine presence of topological order from experimentally accessible data (New J. Phys. 22 045003 (2020)). We also proposed a method to make quantum neural network states more expressive by adding relevant correlators directly to the neural network ( Phys. Rev. R 4, L012010 (2022)).
Adding a little bit of physics to make neural networks more expressive
Neural nets became a powerful ansatz for solving variational problems in condensed matter physics. They are flexible, physics-agnostic and can capture a volume-law entanglement. At the same time, without a lot of customization, they do not always converge, in Phys. Rev. R 4, L012010 (2022) we showed that adding a little extra physics can significanly help! Specifically, we found a way to encode relevant correlators into the structure of the neural network and showed pretty great energy accuracy for some challenging problems.
Unsupervised learning of phase transitions
Detecting exotic phase transitions from measurement data alone is hard. Can machine learning help us understand these phase transitions better? The answer is yes: we built predictive models that can automatically find the boundaries of these exotic phases.
Specifically, analysis of how well can predictive models determine a known, experimentally accessible tuning parameter from data can help us determine a position of the topological phase transition from measured data alone.
Read more here: New J. Phys. 22 045003 (2020)
Hamiltonian learning error correction strategies
Contemporary quantum computers are facing two outstanding challenges: the first is correcting errors and the second is the verification that quantum devices perform the intended tasks.
In this work, we bring these two important directions of research together and create a new paradigm for error correction through verification of a quantum device. Our algorithm can be trained on a classical computer and then be deployed on real quantum devices. This new approach employs precise engineering driven by machine learning and opens a new field of research that may lead to the reformulation of quantum error correction.
Read more here: Phys. Rev. Research 1, 033092 (2019)