This repository contains the code for Interface Laplace Learning: Learnable Interface Term Helps Semi-Supervised Learning
NumPy, SciPy and PyTorch. Scipy is for sparse matrix calculation. PyTorch is for acceleration of matrix multiplication on gpu. The code is tested on the following version:
numpy==1.26.4
scipy==1.11.4
torch==2.2.1
data/ contains pretrained extracted features of each dataset. The files are collected from GraphLearning package without any change, but renamed for clarity.
inter_laplace.py includes the main function to perform training and inference.
utils.py includes utility functions.
preprocess.py includes the preprocess functions to get T, interface index, A and lambda.
The optimal parameters are provided in the appendix. Should take less than 1 second for each trial on gpu.
python inter_laplace.py --dataset mnist --label_num 1 --k_hop 4 --target_mse 0.20