Quantum Techniques in Machine Learning

Low Overhead Precise Gate Importance Estimation for Pruning Parametrized Quantum Circuits

Not scheduled

20m

Jan Mouton Learning Centre (Stellenbosch University)

Poster

Speaker

Rodrigo Martínez Sanz (Universidad Politécnica de Valencia)

Description

Pruning parametrized quantum circuits requires quantifying the local contribution of each gate to the cost function, denoted as the importance in the classical domain, a quantity that existing methods approximate through first- or second-order Taylor expansions. We show that this approximation can be avoided by deriving a closed-form expression for the exact importance at zero additional cost. Through VQE simulations, we demonstrate that our criterion outperforms prior pruning methods, and we further analyze how the trainability and implementability of the circuit improve as increasing the number of pruned gates.

Presentation materials

AbstractQTML_RodrigoMSanz.pdf