Scalable Deep Subspace Clustering Network

TL;DR

SDSNet is a scalable deep subspace clustering framework that reduces computational complexity to linear time using landmark-based approximation and joint optimization, maintaining high clustering quality.

Contribution

It introduces SDSNet, a novel deep clustering method that achieves linear complexity with landmark approximation and joint auto-encoder and self-expression optimization.

Findings

Achieves comparable clustering accuracy to state-of-the-art methods.

Significantly reduces computational time and resources.

Maintains clustering quality with linear complexity.

Abstract

Subspace clustering methods face inherent scalability limits due to the $O(n^3)$ cost (with $n$ denoting the number of data samples) of constructing full $n\times n$ affinities and performing spectral decomposition. While deep learning-based approaches improve feature extraction, they maintain this computational bottleneck through exhaustive pairwise similarity computations. We propose SDSNet (Scalable Deep Subspace Network), a deep subspace clustering framework that achieves $\mathcal{O}(n)$ complexity through (1) landmark-based approximation, avoiding full affinity matrices, (2) joint optimization of auto-encoder reconstruction with self-expression objectives, and (3) direct spectral clustering on factorized representations. The framework combines convolutional auto-encoders with subspace-preserving constraints. Experimental results demonstrate that SDSNet achieves comparable clustering quality to state-of-the-art methods with significantly improved computational efficiency.