Undirected Neural Networks

This repository contains the code for the paper "Modeling Structure with Undirected Neural Networks" by Tsvetomila Mihaylova, Vlad Niculae and André Martins. accepted at ICML 2022.

Contents

• Folder attn contains experiments for sequence completion with undirected self-attention.

• Folder mnist contains experiments for image classification and visualization with undirected convolutional networks.

• Folder parsing contains the experiments for parsing with structured UNN.

Requirements

• Python 3 (>3.6; prefered 3.7)

• Install the latest PyTorch version

• Execute pip install -r requirements.txt to install the required libraries.

There are additional requirements for the parsing experiments, please check the file parsing/parser/README.md.

Citation

@misc{unn2022, url = {https://arxiv.org/abs/2202.03760}, author = {Mihaylova, Tsvetomila and Niculae, Vlad and Martins, Andr'{e} F.T.}, title = {Modeling Structure with Undirected Neural Networks}, year = {2022}, }