AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

Loss Functions for Predictor-based Neural Architecture Search

Han Ji, Yuqi Feng, Jiahao Fan, Yanan Sun

ICCV 2025 · Architectures and efficiency › Neural architecture search · approach: Efficient NAS search strategies · 3 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 PW loss (ListMLE then WARP) instead of single MSE loss (ablation)PWLNAS · NAS-Bench-101test error (%): 6.06 -> 5.80Improved
use PW loss instead of best single loss HR (ablation)PWLNAS · NAS-Bench-101test error (%): 5.83 -> 5.80Improved
use PW loss (HR then MAPE) instead of MSE (ablation)PWLNAS · DARTS CIFAR-10test error (%): 2.74 -> 2.47Improved
use PW loss instead of MAPE (ablation)PWLNAS · DARTS CIFAR-10test error (%): 2.53 -> 2.47Improved
use PW loss instead of WARP (ablation)PWLNAS · TransNAS-Bench-101 Micro Class Sceneaccuracy: 54.76 -> 54.83Improved
use PW loss instead of MSE (ablation)PWLNAS · TransNAS-Bench-101 Micro Jigsawaccuracy: 94.76 -> 94.88Improved
use WARP loss instead of MSE for PINAT predictorPINAT · NAS-Bench-201 (1% data)N@10: 146.60 -> 3.78Improved
use WARP loss instead of HR for MLP-based AP predictorAP · NAS-Bench-201 (1% data)N@10: 23.58 -> 113.20Hurt
use ListMLE instead of MSE for AP predictorAP · NAS-Bench-201 (1% data)N@10: 250.94 -> 22.74Improved
combine MSE with SR ranking loss versus single ranking lossGCN predictor · NAS-Bench-101/201/TransNAS/GraphPrecision@0.5, N@10, Kendall TauHurt
increase training data portionGCN predictor with regression/ranking losses · several NAS search spacesPrecision@0.5 and N@10Hurt
use GT or exp(GT) weights instead of rank-based weights in weighted losses (ablation)WARP and EW · NAS-Bench-201 (1% data)Precision@T and N@KImproved

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