AI Paper Atlas

Research map / Learning theory and algorithms / Submodular and game algorithms

EigenCache: Rethinking Diffusion Acceleration as Covariance-Optimal Forecasting and Submodular Information Allocation

Chenyang Xu, Dezhen Wang, Lin Chen, Kepeng Lin, Hao Wang

ICML 2026 · Learning theory and algorithms › Submodular and game algorithms · approach: Submodular influence maximization

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace Hermite/Taylor polynomial forecasting with Kriging covariance predictorEigenCache vs HiCache · FLUX.1-dev moderate reductionImageReward: 0.9979 -> 0.9985Improved
replace HiCache with EigenCache at aggressive accelerationHiCache (N=9) · FLUX.1-dev aggressive reductionImageReward: 0.9113 -> 0.9120No clear effect
replace HiCache with EigenCache at aggressive accelerationHiCache (N=9) · FLUX.1-dev aggressive reductionLPIPS: 0.4374 -> 0.4150Improved
replace HiCache with EigenCache at aggressive accelerationHiCache (N=9) · FLUX.1-dev aggressive reductionPSNR: 28.647 -> 28.850Improved
use greedy information scheduling instead of fixed interval (ablation)EigenCache vs HiCache · FLUX.1-dev N=7latency (s): 4.84 -> 4.55Improved
use greedy scheduling instead of uniform scheduling (ablation)EigenCache scheduler · target posterior variance 0.01anchors needed: 12 -> 8Improved
replace HiCache with EigenCacheHiCache · HunyuanVideo 3.5xnormalized flickering score: 0.955 -> 0.972Improved
replace TaylorSeer with EigenCacheTaylorSeer · HunyuanVideo 3.5xnormalized flickering score: 0.915 -> 0.972Improved
replace DeepCache with EigenCacheDeepCache · SDXL COCO-2017FID: 18.45 -> 17.82Improved
replace ToCa with EigenCacheToCa · DiT-XL/2 standard scheduleFID-50K: 2.95 -> 2.35Improved
use OUSAC guidance schedule with EigenCacheEigenCache · DiT-XL/2FID-50K: 2.35 -> 2.21No clear effect
use base-model kernel instead of per-LoRA recalibrated oracle kernel (ablation)EigenCache · FLUX.1-dev with Anime LoRAImageReward: 0.9245 -> 0.9238No clear effect

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