Graph transduction as a non-cooperative game

Authors:
Aykut Erdem;Marcello Pelillo
Affiliations:
Hacettepe University, Beytepe, Ankara, Turkey;"Ca' Foscari" University of Venice, Mestre, Venezia, Italy
Venue:
GbRPR'11 Proceedings of the 8th international conference on Graph-based representations in pattern recognition
Year:
2011

Citing 8
Cited 0

The Dynamics of Nonlinear Relaxation Labeling Processes

Journal of Mathematical Imaging and Vision
From Few to Many: Illumination Cone Models for Face Recognition under Variable Lighting and Pose

IEEE Transactions on Pattern Analysis and Machine Intelligence
Clustering through ranking on manifolds

ICML '05 Proceedings of the 22nd international conference on Machine learning
Learning from labeled and unlabeled data on a directed graph

ICML '05 Proceedings of the 22nd international conference on Machine learning
Manifold Regularization: A Geometric Framework for Learning from Labeled and Unlabeled Examples

The Journal of Machine Learning Research
The complexity of computing a Nash equilibrium

Communications of the ACM - Inspiring Women in Computing
Semi-supervised learning by mixed label propagation

AAAI'07 Proceedings of the 22nd national conference on Artificial intelligence - Volume 1
Semi-Supervised Learning

Semi-Supervised Learning

Quantified Score

Hi-index	0.00

Visualization

Abstract

Graph transduction is a popular class of semi-supervised learning techniques, which aims to estimate a classification function defined over a graph of labeled and unlabeled data points. The general idea is to propagate the provided label information to unlabeled nodes in a consistent way. In contrast to the traditional view, in which the process of label propagation is defined as a graph Laplacian regularization, here we propose a radically different perspective that is based on game-theoretic notions. Within our framework, the transduction problem is formulated in terms of a non-cooperative multi-player game where any equilibrium of the proposed game corresponds to a consistent labeling of the data. An attractive feature of our formulation is that it is inherently a multi-class approach and imposes no constraint whatsoever on the structure of the pairwise similarity matrix, being able to naturally deal with asymmetric and negative similarities alike. We evaluated our approach on some real-world problems involving symmetric or asymmetric similarities and obtained competitive results against state-of-the-art algorithms.