[4A1] [General Session] 2. Machine Learning

[4A1-04] Clustering by Deep Mixture Models

〇Kaede Hayashi¹, Tomoharu Iwata², Tadahiro Taniguchi¹ (1. Ritsumeikan University, 2. NTT Communication Science Laboratories)

Keywords:Clustering, Deep learning, Probabilistic generative model

Clustering is an important task in the field of machine learning and artificial intelligence. Since probabilistic generative models have strong assumptions on data‘, feature engineering ’has been required for clustering with Gaussian mixture models (GMM). In the last few years, research on clustering complicated data with a model combining Variational Autoencoder (VAE) and GMM has attracted attention. In this paper, we propose Deep Mixture Models (DMM). In DMM, a latent vector is first generated by GMMs, then latent vector is transfered into an observation. DMMs are trained by maximizing the lower bound of the marginal likelihood. In the experiment, our proposal model shows the best performance compared to baseline methods for the data which are difficult to obtain clusters by GMMs.