AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

Auto-GAS: Automated Proxy Discovery for Training-free Generative Architecture Search

Lujun Li, Haosen Sun, Shiwen Li, Peijie Dong, Wenhan Luo, Wei Xue, Qifeng Liu, Yike Guo

ECCV 2024 · Architectures and efficiency › Neural architecture search · approach: Bi-level differentiable NAS · 14 citations (OpenAlex)

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
use the evolved information-aware proxy instead of existing zero-cost proxiesAuto-GAS · TransNAS-Bench-101 AutoencodingKendall's Tau (%): 51.28% -> 61.14% (vs NWOT)Improved
use parameter-level proxies (Synflow, SNIP, Params) for generator rankingzero-cost proxies · TransNAS-Bench-101 AutoencodingKendall's Tau (%): Synflow -50.91%; Params -12.97%Hurt
evolve proxy with search instead of vanilla Gram-matrix (ablation)Auto-GAS proxy · CIFAR-10FID: 17.88 -> 12.21Improved
include encoding operators in proxy search space (ablation)Auto-GAS proxy · TransNAS-Bench-101 AutoencodingKendall's Tau (%): 51.64% -> 61.14%Improved
include reduction operators in proxy search space (ablation)Auto-GAS proxy · TransNAS-Bench-101 AutoencodingKendall's Tau (%): 52.78% -> 61.14%Improved
include transform operators in proxy search space (ablation)Auto-GAS proxy · TransNAS-Bench-101 AutoencodingKendall's Tau (%): 55.78% -> 61.14%Improved
include augment operators in proxy search space (ablation)Auto-GAS proxy · TransNAS-Bench-101 AutoencodingKendall's Tau (%): 58.40% -> 61.14%Improved
use evolutionary proxy search instead of random search (ablation)Auto-GAS · TransNAS-Bench-101 AutoencodingKendall's Tau (%): 29.49% -> 61.14%Improved
use Pearson or Hamming encoding instead of Gram-matrix (ablation)Auto-GAS proxy · CIFAR-10FID: 12.21 -> 12.85 (Pearson); 12.21 -> 12.35 (Hamming)No clear effect
increase EA population from 16 to 64 or batch size from 20 to 40 (ablation)Auto-GAS search · CIFAR-10FID: 12.21 -> 16.32 (64 individuals); 12.28 -> 16.52 (batch 40)Hurt
replace Gram-matrix-based proxy by L2 norm, entropy, or variance statistics (ablation)Auto-GAS proxy · CIFAR-10FID: 12.21 -> 24.62 (L2 norm)Hurt
use training-free proxy search instead of training-based GAN searchAuto-GAS · GAS search costGPU-days: EAGAN 0.8 -> Auto-GAS 0.09Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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