This repository is the official implementation of the following paper:

• Towards Exact Gradient-based Training on Analog In-memory Computing (NeurIPS 2024, ArXiv preprint).
• Analog In-memory Training on General Non-ideal Resistive Elements: The Impact of Response Functions (NeurIPS 2025, ArXiv preprint)

Requirements

This project is built on the analog in-memory computing open-source library IBM Analog Hardware Acceleration Kit, AIHWKit.

Conda installation (Recommanded)

To install requirements:

conda create -n analog python=3.10
conda activate analog
conda install -y pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia
conda install -c conda-forge aihwkit-gpu
pip install tensorboard matplotlib numpy

Pip installation

pip install -r requirements.txt

Running

See the commands in each directories.