Comparing Loss Functions for Partial Label Learning

This repository is a fork of the public repository LW loss. This repository is part of the BESS project which explores interactions between machine learning and logic. In particular, it contains all experiments related to partial label learning that appear on the paper.

Command to reproduce experiments with the large consistent synthetic dataset (Section 6.2, Figure 7)

python zs29.py

Command to reproduce experiments with the CIFAR10 (Section 6.3, Figure 8)

python zs31.py

Command to reproduce experiments with the CIFAR100 (Section 6.3, Figure 9)

python zs30.py

Command to reproduce experiments with real PLL datasets (Section 6.5, Table 6)

python zs33.py

Reproducing plots

Once the experiments terminated, the plots (Figures 7, 8, 9) can be generated from the logs by running

python plot.py