README.md for code accompanying the paper "Semi-Supervised Partial Label Learning via Confidence-Rated Margin Maximization"

PARM

This repository is the official implementation of the PARM algorithm of the paper "Semi-Supervised Partial Label Learning via Confidence-Rated Margin Maximization" and technical details of this algorithm can be found in the paper.

Requirements

• MATLAB, version 2014a and higher.
• MOSEK toolbox package, which can be download from https://www.mosek.com/.

Note that the MOSEK toolbox has to be licensed and the license can be applied for from https://www.mosek.com/products/academic-licenses/ for free for academic purpose. It has to be reset in the appropriate location as instructed by the MATLAB command we you run the demo code.

To start,

• Create a directory of your choice and copy the toolbox there.
• Set the path in your MATLAB to add the directory you just created.

Then, run this command to enter the MATLAB environment:

matlab

code structure

To see the structure of the source code, run this command in MATLAB command:

help PARM_code

To see the interface details of the training function, run this command in MATLAB command:

help PARM_train

To see the interface details of the testing function, run this command in MATLAB command:

help PARM_predict

demo

This repository provides a demo, i.e. demo.m, which shows the training and testing phase of PARM. Before run demo.m, please rename the variable 'mosek_path' as the path containing quadprog.m of the MOSEK toolbox package. The detailed comments can be found in demo.m.

To run demo.m, run this command in MATLAB command:

demo

hyperparameter setting

The performance of PARM is somewhat sensitive w.r.t. $\lambda$ and $\mu$, whose values are chosen from {0.001,0.005,0.01,0.05,0.1,0.5,1,5,10} via cross validation on the training set. Here we provide some hyperparameter settings for the data sets in our paper:

Data set name	$\lambda$	$\mu$
Deter	10	0.5
Vehicle	10	1
Abalone	0.01	0.01
Satimage	5	0.1
Lost	0.001	0.001
Mirflickr	0.1	0.01
BirdSong	5	1
LYN10	10	1
LYN20	10	1

citation

@inproceedings{NeurIPS20Wang,
    author = {Wang, Wei and Zhang, Min-Ling},
    title = {Semi-supervised partial label learning via confidence-rated margin maximization},
    booktitle = {Advances in Neural Information Processing Systems 33},
    address = {Virtual Event},
    year = {2020},
    pages = {6982-6993}
}