Semi-Supervised Learning with Competitive Infection Models
Abstract
The goal of semi supervised learning methods is to effectively combine labeled and unlabeled data to arrive at a better model. Many such methods rely on graph-based approaches, where continuous labels are propagated through a graph on the input points. Here we argue that it is more effective to consider infection processes on these graphs, whereby at any point in time nodes can infect other nodes with their labels. Since the dynamics of these processes is stochastic, we develop algorithms for efficiently estimating the expected labels over time. We show that our approach addresses many of the limitations of graph based learning, and is also empirically effective.
Cite
Text
Rosenfeld and Globerson. "Semi-Supervised Learning with Competitive Infection Models." International Conference on Artificial Intelligence and Statistics, 2018.Markdown
[Rosenfeld and Globerson. "Semi-Supervised Learning with Competitive Infection Models." International Conference on Artificial Intelligence and Statistics, 2018.](https://mlanthology.org/aistats/2018/rosenfeld2018aistats-semi/)BibTeX
@inproceedings{rosenfeld2018aistats-semi,
title = {{Semi-Supervised Learning with Competitive Infection Models}},
author = {Rosenfeld, Nir and Globerson, Amir},
booktitle = {International Conference on Artificial Intelligence and Statistics},
year = {2018},
pages = {336-346},
url = {https://mlanthology.org/aistats/2018/rosenfeld2018aistats-semi/}
}