Speaker
Description
Hybrid quantum-classical algorithms rely on an efficient back-and-forth between executing quantum circuit operations and processing information on classical computers (e.g., variational optimization or mid-circuit measurements). In this context, compilation is becoming increasingly important to enhance the performance and flexibility of these hybrid pipelines.
During my Xanadu Residency in 2025, I worked on the implementation of a (q)jit-compatible Quantum Natural Gradient optimizer in PennyLane, enabling hybrid programs to run within a fully differentiable and compiled workflow. A key part of this effort was to develop a JAX-native version of the Quantum Natural Gradient optimization and its momentum-based extension. This required rethinking how the metric tensor and parameter updates are represented so that they can be expressed as JAX data structures and integrate smoothly with Catalyst, Xanadu lower-level compiler for hybrid quantum-classical programs.
The new optimizers now run as a streamlined and accelerator-friendly execution pipeline, enabling faster and more scalable training of variational quantum algorithms. These improvements highlight how the hybrid tools developed within the PennyLane open-source software ecosystem support more efficient workflows for near-term algorithms.
| Sessions | Technological aspects |
|---|---|
| Invited | No |