Balancing Fidelity and Diversity in Diffusion Models via Symmetric Attention Decomposition: Hopfield Perspective

Researchers have introduced a method to balance fidelity and diversity in diffusion models by decomposing the attention matrix in transformers into symmetric and skew-symmetric components, interpreting the symmetric part as governing energy landscape structure and the skew-symmetric part as driving circulation. By deriving Hopfield-style stability measures from the symmetric component, the team observed correlations with fidelity-diversity trade-offs in generation and proposed a controllable knob to modulate this balance through circulation modification. The approach, detailed in a new arXiv paper with accompanying code on GitHub, offers a principled way to tune generative model outputs.

arXiv:2605.27476v1 Announce Type: new Abstract: We characterize the pre-softmax attention matrix $\mathbf{QK^\top}$ in transformers as an associative memory matrix encoding pairwise associations between input features. By decomposing this matrix into its symmetric and skew-symmetric parts, we interpret the symmetric component as governing the structure of the energy landscape, and the skew-symmetric component as driving circulation on that landscape. Leveraging the energy formulation induced by the symmetric component, we derive Hopfield-style stability measures that quantify the stability of retrieved features. We observe meaningful correlations between Hopfield-style stability measures and the fidelity-diversity trade-offs in generation. Finally, we propose a controllable knob to modulate this trade-off by modifying the circulation of the underlying dynamics. Code is available at our GitHub https://github.com/hyeon-cho/Attention-Symmetric-Decomposition .