Understanding the nature of dressed states in superconducting transmon qubit systems is central to implementing high-fidelity quantum computing with this hardware . Most frequency and coupler configurations between transmons lead to localized states, where most of the excitations exist at a single transmon site. However, finding dressed states that are jointly excited at two distant sites is much more challenging and interesting, as they are inherently entangled states. Studying these states and the configurations that cause them can help us mitigate unwanted crosstalk and better understand quantum transport in the system.
We approach the problem of discovering new bare frequencies that produce such dressed states by utilizing generative machine learning. Specifically, we train a denoising diffusion model to produce these frequencies by learning from known examples, allowing us to sidestep the exponentially increasing computation cost from directly computing Hamiltonians as the system size grows. Preliminary results have shown promising capabilities for generating interesting dressed states in small, 1D systems, with more exciting work ahead for larger, more complex architectures.