Particle Competition and Cooperation in Networks for Semi-Supervised Learning

Abstract

Semi-supervised learning is one of the important topics in machine learning, concerning with pattern classification where only a small subset of data is labeled. In this paper, a new network-based (or graph-based) semi-supervised classification model is proposed. It employs a combined random-greedy walk of particles, with competition and cooperation mechanisms, to propagate class labels to the whole network. Due to the competition mechanism, the proposed model has a local label spreading fashion, i.e., each particle only visits a portion of nodes potentially belonging to it, while it is not allowed to visit those nodes definitely occupied by particles of other classes. In this way, a "divide-and-conquer" effect is naturally embedded in the model. As a result, the proposed model can achieve a good classification rate while exhibiting low computational complexity order in comparison to other network-based semi-supervised algorithms. Computer simulations carried out for synthetic and real-world data sets provide a numeric quantification of the performance of the method.

Preventing Error Propagation In Semi-Supervised Learning Using Teams Of Walking Particles

“…These features allow particles to leave mislabeled nodes, which will be usually inside other class neighborhood, and help their teammates in the neighborhood of its respective class. In our experiments, when there is no mislabeled nodes, the proposed algorithm correct classification rates are compatible with those achieved by the particles competition and cooperation method [26] in the same data set. The advantage of the proposed algorithm appears when there are mislabeled samples in the data sets, in these cases it performs better than the particles competition and cooperation method and other representative semi-supervised learning graph-based methods.…”

“…The next step is to compare the proposed method with other representative semi-supervised learning graph-based methods when applied to real-world networks with mislabeled data. Here the performance of the proposed method is compared to those of Local and Global Consistency (LGC) [13], Label Propagation (LP) [14], Linear Neighborhood Propagation (LNP) [15], and the original Particle Competition and Cooperation (PCC) method [26]. The σ parameters of the LGC and the LP methods, and the k parameters of LNP, PCC and the proposed method, are all optimized using the genetic algorithm available in the Global Optimization Toolbox of MATLAB.…”

“…Particles compete with each other in order to possess nodes of the network, naturally confining themselves within a cluster. Later, this approach was extended to realize semi-supervised learning [26,27]. In this version, there are team of particles which cooperates with their teammates and compete against other teams.…”

“…Those particle walking based methods provide classification results similar to those from some well known methods, but with lower order of computational complexity. The particle competition and cooperation approach [26] naturally has some tolerance to mislabeled data [28], though it was not designed to solve this particular problem. In this paper, we present a new semi-supervised learning particle walking algorithm which is specifically designed to handle data sets with large amounts of mislabeled data.…”

An Oscillatory Correlation Model for Semi-Supervised Classification

“…Among these approaches, network-based techniques (which is the topic explored in this paper) have received increased attention in the recent years [5][6][7][8][9][10][11][12][13][14]. The main feature of these techniques lies in the way data is represented: the network nodes represent the data objects whereas the edges represent the distances (similarities) among objects.…”