Scholarlink

<p>The transmission of classical data through quantum channels is susceptible to errors that can cause irrecoverable information loss. In image processing, erasure errors completely destroy pixel data in a region, making traditional denoising insufficient. A successful recovery requires the model to predict the missing content generatively. We introduce a hybrid quantum-classical autoencoder designed for this task of reconstruction. Our model is engineered to reconstruct images that suffer from both corruption (bit and phase flips) and irrecoverable loss (erasures). The core of our architecture is a variational quantum circuit that functions as a generative model in the latent space, learning the underlying data distribution to predict missing information from the surrounding context. Across MNIST, CIFAR-10, and Omniglot datasets, our model demonstrates robust predictive reconstruction, maintaining high perceptual fidelity even when combined error and erasure rates are extreme. This work presents a novel application of hybrid architectures, using a quantum circuit’s expressive power to solve the challenging problem of generative reconstruction for quantum-corrupted data.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Generative reconstruction for irrecoverable erasure errors in quantum-corrupted images

  • 期刊论文

  • 发表日期：

    2025年11月10日

查看全文
  • Prokash Chandra Roy,
  • Saikat Barua,
  • Md. Razwanul Islam Tanvir,
  • Freyana Zanain,
  • M. R. C. Mahdy

摘要

The transmission of classical data through quantum channels is susceptible to errors that can cause irrecoverable information loss. In image processing, erasure errors completely destroy pixel data in a region, making traditional denoising insufficient. A successful recovery requires the model to predict the missing content generatively. We introduce a hybrid quantum-classical autoencoder designed for this task of reconstruction. Our model is engineered to reconstruct images that suffer from both corruption (bit and phase flips) and irrecoverable loss (erasures). The core of our architecture is a variational quantum circuit that functions as a generative model in the latent space, learning the underlying data distribution to predict missing information from the surrounding context. Across MNIST, CIFAR-10, and Omniglot datasets, our model demonstrates robust predictive reconstruction, maintaining high perceptual fidelity even when combined error and erasure rates are extreme. This work presents a novel application of hybrid architectures, using a quantum circuit’s expressive power to solve the challenging problem of generative reconstruction for quantum-corrupted data.