Deep Discriminative Clustering Analysis

TL;DR

Deep Discriminative Clustering (DDC) leverages a deep neural network with global and local constraints to learn discriminative representations, significantly improving clustering performance across multiple data modalities.

Contribution

This paper introduces DDC, a novel deep clustering method that models relationships between patterns with constraints, enabling end-to-end learning of discriminative features for clustering.

Findings

DDC outperforms existing methods on eight diverse datasets.

The method converges theoretically during iterative training.

Discriminative representations serve as effective clustering centers.

Abstract

Traditional clustering methods often perform clustering with low-level indiscriminative representations and ignore relationships between patterns, resulting in slight achievements in the era of deep learning. To handle this problem, we develop Deep Discriminative Clustering (DDC) that models the clustering task by investigating relationships between patterns with a deep neural network. Technically, a global constraint is introduced to adaptively estimate the relationships, and a local constraint is developed to endow the network with the capability of learning high-level discriminative representations. By iteratively training the network and estimating the relationships in a mini-batch manner, DDC theoretically converges and the trained network enables to generate a group of discriminative representations that can be treated as clustering centers for straightway clustering. Extensive experiments strongly demonstrate that DDC outperforms current methods on eight image, text and audio datasets concurrently.