Hands-On Deep Learning for IoT

Neural networks for clustering analysis

Several variants of k-means have been proposed to address issues with higher-dimensional input spaces. However, they are fundamentally limited to linear embedding. Hence, we cannot model non-linear relationships. Nevertheless, fine-tuning in these approaches is based on only cluster assignment hardening loss (see later in this section). Therefore, a fine-grained clustering accuracy cannot be achieved. Since the quality of the clustering results is dependent on the data distribution, deep architecture can help the model learn mapping from the data space to a lower-dimensional feature space in which it iteratively optimizes a clustering objective. Several approaches have been proposed over the last few ...