AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Flow-based Spectral Kernel Learning for Nonstationary Attention

Zi Yang, Ying Li, Alejandro Lancho, Dario C. Larese, Pablo Olmos, Michael Minyi Zhang

NeurIPS 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse and structured attention · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes flow-based spectral kernel attention (FSKA), which learns expressive nonstationary attention kernels to adapt similarity structure beyond fixed or stationary kernels.

Problem

Existing kernelized attention relies on fixed or stationary kernels, which limits its ability to adapt similarity structure to complex token interactions.

Approach

Learns nonstationary attention kernels via a flow-based spectral parameterization, replacing fixed or stationary kernels in kernelized attention with learned ones.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
remove query/key and value projections from a standard Transformer (ablation)Transformer · MNISTaccuracy: 0.8556 -> 0.5267Hurt
remove query-key projections and fix value projection, with learned flow kernel (ablation)FSKA · MNISTaccuracy: 0.8656 (IKAN) -> 0.8722Improved
make value projection learnable (ablation)FSKA · MNISTaccuracy: 0.8722 -> 0.8725No clear effect
use positive random features (ELU+1) instead of direct RFF (ablation)FSKA · MNISTaccuracy: 0.3956 -> 0.8722Improved
increase RFF dimension M from 32 to 128 (ablation)FSKA · MNISTaccuracy: 0.8656 -> 0.8822Improved
increase NGSM mixture count Q from 1 to 10 (ablation)FSKA · MNISTaccuracy: 0.8483 -> 0.8743Improved
increase NGSM mixture count Q beyond 10 (ablation)FSKA · MNISTaccuracy: 0.8743 -> 0.8431 (Q=20)No clear effect
increase flow depth N_f from 2 to 10 (ablation)FSKA · MNISTaccuracy: 0.8503 -> 0.8707Improved
replace Transformer attention with FSKATransformer · SST-2F1-macro: 76.07 -> 78.74Improved
replace Transformer attention with FSKATransformer · SST-5F1-macro: 35.11 -> 38.16Improved
use FSKA instead of Mamba on full documentsMamba · SCOTUSaccuracy: 59.84 -> 58.91Hurt
use FSKA instead of Mamba on full documentsMamba · SCOTUSmacro-F1: 37.91 -> 39.68Improved

Experimental setup

Builds on: kernelized attention

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy