21–26 Jun 2026
U. Ottawa - Learning Crossroads (CRX) Building
America/Toronto timezone
Welcome to the 2026 CAP Congress Program website! / Bienvenue au siteweb du programme du Congrès de l'ACP 2026!

Adaptively Reconstructing a Quantum State Using Artificial Neural Networks

22 Jun 2026, 15:00
15m
U. Ottawa - Learning Crossroads (CRX) Building

U. Ottawa - Learning Crossroads (CRX) Building

100 Louis-Pasteur Private, Ottawa, ON K1N 9N3
Oral Competition (Graduate Student) / Compétition orale (Étudiant(e) du 2e ou 3e cycle) Division for Quantum Information / Division de l'information quantique (DQI / DIQ) (DQI) M2-5 | (DIQ)

Speaker

Riza Fazili (University of Ottawa)

Description

Quantum state tomography (QST) aims to reconstruct an unknown quantum state from measurement statistics and is a central tool for characterization and validation in quantum information and quantum computing. A key practical limitation in QST is sample cost: reconstructing a quantum state requires data from an informationally complete measurement set, and the number of distinct measurement settings scales exponentially with the system size. Adaptive QST mitigates this burden by using interim data to choose subsequent measurement settings, allocating samples to the most informative settings and reducing the number of settings that must be sampled exhaustively.
We investigate an adaptive QST strategy that uses Restricted Boltzmann Machines (RBMs), energy-based generative neural networks trained on measurement results, to represent quantum states. A committee of independently trained RBMs guides measurement selection by ranking candidate measurement settings by prioritizing measurements that are expected to reduce the model uncertainty most. To quantify the benefit of adaptivity under controlled conditions, we implement a candidate pool of Pauli measurement settings and simulate measurements outcomes in Qiskit (an open-source quantum computing software development kit). Target states are drawn from random ensembles spanning both pure and mixed states for one- to five-qubit systems.
For each total sample budget, we compare three reconstruction pipelines: (i) non-adaptive RBM tomography with a fixed measurement schedule, (ii) adaptive RBM tomography with data-driven measurement selection, and (iii) conventional maximum-likelihood estimation (MLE), which reconstructs the quantum state by maximizing the likelihood of the observed measurement data subject to physicality constraints. We quantify the performance by evaluating the infidelity (a difference measure between quantum states) for varying numbers of total samples, while also comparing total computational runtimes. This study aims to delineate the regimes in which adaptive RBM-based QST provides measurable gains over the non-adaptive design and MLE under identical measurement constraints.

Keyword-1 Quantum State Tomography
Keyword-2 Adaptive Quantum State tomoghy

Author

Riza Fazili (University of Ottawa)

Co-authors

Prof. Jeff Lundeen (University of Ottawa) Prof. Stefanie Czischek (University of Ottawa)

Presentation materials

There are no materials yet.