SSOG Attention Challenges SDPA with Subquadratic Complexity in Vision Transformers

Researchers propose SSOG (Sum of Separable Gaussians) attention, a subquadratic alternative to SDPA that achieves better performance on Vision Transformers with near-linear scaling.

2026-08-16 ~ 2026-08-17 · 2 related posts

• SSOG Attention: Sub-Quadratic Alternative to SDPA with Separable Gaussians, Beats on CIFAR-100 — 4rtemi5 · 2026-08-16
• SSOG Challenges SDPA in ViTs, Offering Faster Convergence and Near-Linear Scaling — serrjoa · 2026-08-17