marlin-codes / hie Goto Github PK

Authors: Menglin Yang, Min Zhou, Rex Ying, Yankai Chen, Irwin King
Codes: https://github.com/marlin-codes/HIE

The non-Euclidean geometry of hyperbolic spaces has recently garnered considerable attention in the realm of representation learning. Current endeavors in hyperbolic representation largely presuppose that the underlying hierarchies can be automatically inferred and preserved through the adaptive optimization process. This assumption, however, is questionable and requires further validation. In this work, we first introduce a position-tracking mechanism to scrutinize existing prevalent hyperbolic neural networks, revealing that the learned representations are sub-optimal and unsatisfactory. To address this, we propose a simple yet effective method, hyperbolic informed embedding (HIE), by incorporating cost-free hierarchical information deduced from the hyperbolic distance of the node to origin (i.e., induced hyperbolic norm) to advance existing hyperbolic neural networks. The proposed HIE method is both task-agnostic and model-agnostic, enabling its seamless integration with a broad spectrum of models and tasks.

This repository is an official PyTorch Implementation for "Hyperbolic Representation Learning: Revisiting and Advancing (ICML 2023)"

Note: this project is built upon HGCN. We thank the authors for their wonderful work.

The code was developed and tested on the following python environment, other similar environment should also work.

python 3.8.10
scikit-learn 1.3.2
numpy 1.24.2
scipy 1.10.1
tqdm 4.61.2
torch                    2.0.0+cu118
torch-cluster            1.6.1
torch-geometric          2.4.0
torch-scatter            2.1.2
torch-sparse             0.6.18
torch-spline-conv        1.2.2
geoopt 0.5.0

bash examples/icml2023/hgnn/HGCN/batch_cora_8.sh
results: 0.7880, 0.8230, 0.8310

bash examples/icml2023/hgnn/HGCN/batch_citeseer_8.sh
results: 0.6250, 0.7360, 0.7230

bash examples/icml2023/hgnn/HGCN/batch_pubmed_8.sh
results: 0.7780, 0.7990, 0.7900

bash examples/icml2023/hgnn/HGCN/batch_disease_256.sh
results: 0.9389, 0.9427, 0.9447

bash examples/icml2023/hgnn/HGCN/batch_disease_8.sh
results: 0.8936, 0.8696, 0.9318

Please check the folder of examples for more usages of the codes.

If you find this code useful in your research, please cite the following paper:

@inproceedings{yang2023hyperbolic,
author = {Yang, Menglin and Zhou, Min and Ying, Rex and Chen, Yankai and King, Irwin},
title = {Hyperbolic Representation Learning: Revisiting and Advancing},
year = {2023},
booktitle = {Proceedings of the 40th International Conference on Machine Learning},
articleno = {1654},
numpages = {21},
location = {Honolulu, Hawaii, USA},
series = {ICML'23}
}