Guarantees and limitations for warm starts and iterative methods in variational quantum computing

Guarantees and limitations for warm starts and iterative methods in variational quantum computing

Seminar author:Ricard Puig

Event date and time:03/20/2025 02:30:pm

Event location:Seminar Room, GIQ

Event contact:

Barren plateaus are fundamentally a statement about quantum loss landscapes on average but there can exist patches of barren plateau landscapes with substantial gradients. This has motivated the study of warm starts whereby the algorithm is cleverly initialized closer to a minimum. Numerical studies indicate that these methods may be promising. In parallel, analytic studies have proven that small angle initializations, whereby the parameterized quantum circuit is initialized in a small region typically around identity or a Clifford, can exhibit non-exponentially vanishing gradients. However, a good solution may be far from this region and thus these methods can (in full generality) only work on a vanishing fraction of problem instances. We present general analysis of warm starts for physically-motivated ansatze and iterative training strategies.  Our work thus suggests that while there are hopes to be able to warm-start variational quantum algorithms, any initialization strategy that cannot get increasingly close to the region of attraction with increasing problem size is likely to prove challenging to train. 

Refs: https://arxiv.org/abs/2404.10044,  https://arxiv.org/abs/2502.07889