memCS

This is the source code for our paper: Yunrui Jiao, Han Zhao, Jianshi Tang, et al., A memristor-based energy-efficient compressed sensing accelerator with hardware-software co-optimization for edge computing, National Science Review, 2025; nwaf499.

code

AMP algorithm

• The source code of the approximate message passing (AMP) algorithm comes from the formulas in this paper.

MMM & SE strategies

• These two strategies are used to optimize the measurement matrix $\Phi$ and the sparse transform matrix $\Psi$ in the AMP algorithm, respectively.

classification

• The dataset and source code for this section are saved here.

pretrained_models

• A pre-trained convolutional neural network (default: ResNet50) is used here to perform classification inference on the reconstructed images.

ImageNet

• For the source of the ImageNet-1k dataset, see here.
• A tool code (./ImageNet/parquet_to_image.ipynb) is used to extract images from a specific data structure for experiments.