arXiv:2609.35163
Deployment of Large IQP Circuit Born Machines on Qubit-Limited Hardware Using Noise Injection
Ju-Young Ryu, Wooyeong Song, Kwangil Bae, Wonhyuk Lee, Ilkwon Sohn
Abstract
Quantum circuit Born machines (QCBMs) built with instantaneous quantum polynomial (IQP) circuits have loss functions that can be efficiently estimated classically, but require quantum computers during inference. Thus, classical computers can train large IQP QCBMs that exceed the qubit numbers of available quantum hardware. This motivates research into how qubit-limited quantum computers can be used to sample from larger IQP QCBMs. Leveraging insights from classical simulation of noisy IQP circuits, we propose a deployment protocol for IQP QCBMs that injects noise at the logical level to probabilistically decompose the original circuit into smaller clusters that can be sampled on qubit-limited quantum processors. We also apply lightweight resampling of known noisy qubits to mitigate the impact of the introduced error. Numerical simulations of our protocol applied to shallow IQP QCBMs and IBM Heron experiments deploying a 316-qubit IQP circuit onto 156-qubit hardware show that mitigation improves the similarity of the empirical distribution to the target distribution. The proposed deployment protocol also crucially preserves correlation structures more accurately than a classical baseline that has accurate single-qubit marginals.
AI Summary
IQP circuit Born machines can be trained classically even when their qubit count exceeds available hardware. This paper asks how to approximately sample such models using smaller processors, without interprocessor entanglement or quasiprobability circuit cutting. The protocol deliberately injects logical dephasing, making selected qubits maximally mixed and breaking the interaction graph into independently executable fragments. Configuration-model percolation estimates, followed by empirical tuning, choose the removal probability to meet a hardware-capacity success target. Removed output bits are then resampled using ideal single-bit marginals (ProdEM) or marginals conditioned on surviving neighbors (LocalCondEM), estimated classically. LocalCondEM requires at most $2^{\deg(i)+1}$ expectation values per removed node. With exact marginals and noiseless fragment execution, its edge-correlator bias is bounded by $2p_p^2$ under independent removal. Simulations use six 28-qubit graph-generation models and fragment capacities of 15, 18, and 21 qubits. Mitigation brings graph-quality metrics near an independent-marginal baseline while reducing errors on graph-local correlators. A hardware demonstration approximates a 316-qubit random IQP circuit using a 156-qubit IBM Heron processor: 1,015 accepted samples require 3,362 hardware calls. Mitigation reduces $\mathrm{MMD}^2$ from $4.78\times10^{-4}$ to $3.51\times10^{-4}$, versus $3.38\times10^{-4}$ for the baseline, while improving selected local correlators. The protocol samples a deliberately distorted distribution, not the noiseless target. Stated limitations include circuit-structure dependence, restrictive hardware-noise assumptions, weak mitigation guarantees, and likely unsuitability for denser or deeper models. The hardware test is a deployment demonstration rather than a generative-quality benchmark; its bootstrap intervals exclude drift, calibration variability, and reference-estimation uncertainty.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.