AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

Progressive Neural Architecture Generation

Caiyang Yu, Chen Huang, Yun Liu, Chenwei Tang, Wei Ju, Jiancheng Lv

CVPR 2026 · Architectures and efficiency › Neural architecture search · approach: Bi-level differentiable NAS

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 diffusion-style iterative network inference with quantization-based multi-scale sub-architecture generation (MSQ)PNAG vs DiffusionNAG · NB201 CIFAR-10 generationtime per architecture: 8 s -> 0.006 s (1300x)Improved
add step-wise consistency constraint (SCC) on more generation steps (ablation)PNAG · CIFAR10 on NB201 and MBV3validity and accuracyImproved
use PNAG instead of DiffusionNAG for generationNB201 Aircraft · NB201 Aircraftaccuracy (%): 58.83 -> 66.99Improved
use PNAG instead of DiffusionNAGMBV3 CIFAR100 · MBV3 CIFAR100accuracy (%): 86.07 -> 87.20Improved
use PNAG instead of DiffusionNAG on CIFAR-10 and CIFAR-100 in NB201NB201 · NB201 CIFAR-10/CIFAR-100accuracy (%): 94.37 -> 94.37; 73.51 -> 73.51No clear effect
use PNAG instead of DiffusionNAG for DARTS-space CIFAR-100DARTS space · DARTS space CIFAR-100test accuracy (%): PNAG 83.54 vs OStr-DARTS 84.22Mixed
add step-wise validity supervision (PNAG) vs final-step-only constraint (DiffusionNAG)PNAG · NB201 validityvalidity (%): 98.97 -> 100.00Improved
swap surrogate model to match target objective (fClean, fAPGD, fBlur)PNAG conditional generation · NB201 CIFAR10 Clean/APGD/Blurmax accuracy (%): APGD 34.52; Blur 48.37; Clean 92.17Improved
apply PNAG to Transformer search space (AutoFormer) TinyAutoFormer space · ImageNet Tinytop-1 accuracy (%): 74.7 -> 76.6Improved
apply PNAG to AutoFormer BaseAutoFormer space · ImageNet Basetop-1 accuracy (%): 82.4 -> 82.4No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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