Method and apparatus for training layered belief propagation-based deep learning model for decoding quantum error correction code

A deep learning model using concatenated NBP models with quantization activation functions addresses high decoding complexity and latency in QLDPC codes by integrating degeneracy, achieving efficient and fast quantum error correction.

WO2026116650A1PCT designated stage Publication Date: 2026-06-04KOREA ADVANCED INST OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
KOREA ADVANCED INST OF SCI & TECH
Filing Date
2025-07-08
Publication Date
2026-06-04

AI Technical Summary

Technical Problem

Existing quantum error correction decoding methods, particularly for QLDPC codes, face challenges in high decoding complexity and latency due to the lack of integration of degeneracy properties and inefficient message scheduling in hierarchical trust propagation algorithms.

Method used

A deep learning model is designed using neural belief propagation (NBP) models concatenated with quantization activation functions, trained stepwise to minimize loss functions considering layered constraints and degeneracy, determining optimal message passing orders without post-processing.

Benefits of technology

This approach reduces decoding delay time and complexity, enhancing the performance of quantum error correction by reflecting degeneracy characteristics and improving convergence speed.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure KR2025009808_04062026_PF_FP_ABST
    Figure KR2025009808_04062026_PF_FP_ABST
Patent Text Reader

Abstract

A method for training a layered belief propagation-based deep learning model for decoding a quantum error correction code, according to an embodiment of the present invention, comprises the steps of: initializing a weight assigned to the deep learning model including at least one neural belief propagation (NBP) model; and training the deep learning model so as to minimize a loss function determined in consideration of layered constraints and degeneracy, wherein the deep learning model includes at least one first deep learning model designed by concatenating the at least one NBP model by the number of first layers, and the at least one first deep learning model is configured by being concatenated by the number of second layers corresponding to a preset number of repetitions.
Need to check novelty before this filing date? Find Prior Art