DiEC: Diffusion Embedded Clustering

TL;DR

DiEC introduces an unsupervised clustering framework that leverages optimal intermediate representations from pretrained diffusion models, systematically identifying the most clustering-friendly layer and timestep to enhance performance without additional data augmentation.

Contribution

The paper proposes a novel method to select optimal diffusion model representations for clustering, improving efficiency and effectiveness over traditional single-representation approaches.

Findings

Achieves state-of-the-art clustering results on multiple benchmarks.

Effectively identifies clustering-optimal layer and timestep in diffusion models.

Reduces computational overhead compared to existing methods.

Abstract

Deep clustering methods typically rely on a single, well-defined representation for clustering. In contrast, pretrained diffusion models provide abundant and diverse multi-scale representations across network layers and noise timesteps. However, a key challenge is how to efficiently identify the most clustering-friendly representation in the layer*timestep space. To address this issue, we propose Diffusion Embedded Clustering (DiEC), an unsupervised framework that performs clustering by leveraging optimal intermediate representations from pretrained diffusion models. DiEC systematically evaluates the clusterability of representations along the trajectory of network depth and noise timesteps. Meanwhile, an unsupervised search strategy is designed for recognizing the Clustering-optimal Layer (COL) and Clustering-optimal Timestep (COT) in the layer*timestep space of pretrained diffusion models, aiming to promote clustering performance and reduce computational overhead. DiEC is fine-tuned primarily with a structure-preserving DEC-style KL-divergence objective at the fixed COL + COT, together with a random-timestep diffusion denoising objective to maintain the generative capability of the pretrained model. Without relying on augmentation-based consistency constraints or contrastive learning, DiEC achieves excellent clustering performance across multiple benchmark datasets.