Understanding Clustering in Slot Attention via Particle Dynamics

2026-10-03

AI Summary

This study addresses the limitation that neural network components in existing Slot Attention obscure the intrinsic dynamical nature of clustering behavior, leaving object-centric representations without mechanistic interpretability. To this end, this work proposes a parameter-free Simplified Slot Attention (SSA) that eliminates all learnable components and establishes theoretical connections to soft k-means and particle dynamics. By doing so, it elucidates the formation process of object-centric representations from a mechanistic perspective, demonstrating that competitive object-centric segmentation can be achieved without neural networks. Evaluated on Pascal VOC, SSA attains performance comparable to standard Slot Attention, offering a novel perspective for understanding attention-based clustering mechanisms.

Abstract

Studying attention through the lens of interacting particle dynamics has shown how token clustering can emerge from the underlying dynamics. We extend this perspective to slot attention, a method for object-centric image segmentation and representation learning in which learned components obscure how much of the clustering behaviour is intrinsic to the attention dynamics. We therefore introduce simplified slot attention (SSA), a parameter-free variant whose dynamics are connected to soft $k$-means clustering and which provides a straightforward mechanistic explanation for the emergence of object-centric representations. On the Pascal VOC dataset, SSA achieves performance comparable to that of slot attention, demonstrating that competitive object-centric segmentation can be achieved without learned neural-network components.

Problem

• Slot Attention
• Clustering
• Particle Dynamics
• Object-centric Representation

Innovation

• Slot Attention
• Particle Dynamics
• Soft k-means Clustering
• Object-centric Representation
• Parameter-free