Speaker
Description
Variational quantum algorithms (VQAs) — including VQE, QAOA, and VQLS — share
a classical optimisation problem: minimising a parameterised cost function over
barren plateaus, noisy high-dimensional loss landscapes. While prior work has
benchmarked optimisers within individual VQA families, systematic comparison
across task families, noise models, and qubit counts remains limited. This work
extends the Pellow-Jarman et al. (2021) VQLS optimiser comparison study into a
broader benchmarking system, evaluating how optimiser performance generalises
across VQA variants and realistic hardware conditions.
Drawing on the findings of Novak et al. (2025), we benchmark an expanding set
of gradient-based and derivative-free optimisers spanning evolutionary,
physics-inspired, and bio-inspired strategies across VQE, QAOA, and VQLS,
multiple ansatz architectures, and noise models from shot noise to
hardware-like conditions.
A core contribution is a retrospective early stopping analysis motivated by
barren plateau theory. Arrasmith et al. (2021) show that as qubit count grows,
the cost function signal becomes harder to detect, causing fixed patience
thresholds to trigger prematurely even when real progress is still happening.
Rather than imposing early stopping during runs, we record the index of the
last significant cost improvement and study how it scales with qubit count,
noise level, and optimiser choice.
Finally, we emphasise open and modular software design — each component is
independently extendable to support community reuse — and propose an
AI-assisted interactive platform where users configure experiments, observe
optimisation in real time, and receive plain-language explanations from an
integrated LLM assistant.
| Topic | Quantum variational circuits and their applications |
|---|