29–31 Oct 2024
Padova
Europe/Rome timezone

Early exiting from Quantum Neural Network as a noise mitigation strategy in NISQ devices, a preliminary study

29 Oct 2024, 17:20
20m
Sala Elettra (Palazzo della Salute)

Sala Elettra

Palazzo della Salute

Via San Francesco, 90 - Padova

Speaker

Giacomo Vittori (Sapienza Università di Roma)

Description

Quantum Neural Networks hold great promise for addressing computational challenges, but noise in near-term quantum devices remains a significant obstacle for circuit depth. In this work, we propose a preliminary study on a novel noise mitigation strategy based on early exit, traditionally used in classical deep learning to improve computational efficiency. Experiments have been conducted on a classification task over MNIST dataset, where early exit mechanism has been implemented through mid circuit measurements. The proposed methodology shows promising results under coherent noise, while requiring further refinement under incoherent noise conditions. Despite these limitations, the approach offers a promising path toward enhancing the robustness of QNN on near-term quantum devices.

Sessione Quantum Machine Learning

Primary authors

Andrea Ciardiello (Istituto Nazionale di Fisica Nucleare) Giacomo Vittori (Sapienza Università di Roma) Simone Scardapane (Sapienza Università di Roma and INFN) Stefano Giagu (Sapienza Università di Roma and Istituto Nazionale di Fisica Nucleare)

Presentation materials