Research

Not All Dimensions Are Equal: Outlier-Aware Attention Linearization with Grouped Quadratic Kernels

Single source1 source12 Oct 2026

arXiv:2610.13674v1 Announce Type: cross Abstract: Linearizing softmax attention offers a route to efficient Transformer inference, but preserving pretrained attention remains challenging.

Reported by 1 source

arxiv.orghttps://arxiv.org/abs/2610.13674

← Back to all developments · Signals are a development reported by more than one source. See the methodology.