M2AFI DATA

Incomplete multi-view clustering (IMC) is prevalentin machine learning since multi-view data from nature andsociety are difficult to fully observed. Comparing to multi-viewclustering (MVC), IMC is more complicated since it simulta?neously addresses data restoration and clustering. Current algo?rithms are criticized for sequently executing feature learning andclustering, and time consuming for data restoration. To addressthese limitations, we propose a novel Maximization-MinimizationAdversarial learning and Feature Imputation algorithm (M2AFI)for IMC, which simultaneously enhances relations of views,feature imputation, and dynamically interaction between featurelearning and clustering. Specifically, M2AFI learns features ofsamples with nonnegative matrix factorization, and employsSchatten p-norm to enhance relations of views. Moreover, itavoids data restoration of missing samples by imputing featuresof missing samples, which reduces time complexity of algorithms.Moreover, M2AFI utilizes adversarial learning to dynamicallyinteract features of samples and clusters, thereby improvingdiscriminative of features. The experimental results prove thatM2AFI outperforms state-of-the-art baselines in terms accuracyand robustness.