Adversarial Deep Embedded Clustering: on a better trade-off between\n Feature Randomness and Feature Drift

Clustering using deep autoencoders has been thoroughly investigated in recent\nyears. Current approaches rely on simultaneously learning embedded features and\nclustering the data points in the latent space. Although numerous deep\nclustering approaches outperform the shallow models in achieving favorable\nresults on several high-semantic datasets, a critical weakness of such models\nhas been overlooked. In the absence of concrete supervisory signals, the\nembedded clustering objective function may distort the latent space by learning\nfrom unreliable pseudo-labels. Thus, the network can learn non-representative\nfeatures, which in turn undermines the discriminative ability, yielding worse\npseudo-labels. In order to alleviate the effect of random discriminative\nfeatures, modern autoencoder-based clustering papers propose to use the\nreconstruction loss for pretraining and as a regularizer during the clustering\nphase. Nevertheless, a clustering-reconstruction trade-off can cause the\n\\textit{Feature Drift} phenomena. In this paper, we propose ADEC (Adversarial\nDeep Embedded Clustering) a novel autoencoder-based clustering model, which\naddresses a dual problem, namely, \\textit{Feature Randomness} and\n\\textit{Feature Drift}, using adversarial training. We empirically demonstrate\nthe suitability of our model on handling these problems using benchmark real\ndatasets. Experimental results validate that our model outperforms\nstate-of-the-art autoencoder-based clustering methods.\n

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.