PG-Learn (Parallel Graph Learning) offers a new way of doing graph-based semi-supervised classification by combining graph construction and label inference in a loop to enable supervised hyper-parameter tuning.
PG-Learn infers the edge weights towards the best performance of a specific classification task. To this end, PG-Learn optimizes a learning-to-rank validation loss function with a gradient-based, parallel, and resource-adaptive scheme.
PG-Learn is proved to outperform a variety of existing graph construction methods in accuracy (per fixed time budget for hyperparameter tuning), and scales more effectively to high dimensional, potentially noisy feature space.
Please cite the following paper should you use any of the code, datasets, or ideas from our work:
Xuan Wu∗, Lingxiao Zhao∗ and Leman Akoglu. (∗ equal contribution) A Quest for Structure: Jointly Learning the Graph Structure and Semi-Supervised Classification. In 27th ACM International Conference on Information and Knowledge Management (CIKM) . 2018.