AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

FreeAugment: Data Augmentation Search Across All Degrees of Freedom

Tom Bekor, Niv Nayman, Lihi Zelnik-Manor

ECCV 2024 · Architectures and efficiency › Neural architecture search · approach: Bi-level differentiable NAS · 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
jointly learn all four degrees of freedom instead of freezing magnitudes to uniform (ablation)FreeAugment · CIFAR100top-1 accuracy (%): 79.64 -> 80.04Improved
learn types and order instead of freezing them to uniform (ablation)FreeAugment · CIFAR100top-1 accuracy (%): 79.54 -> 80.04Improved
learn policy depth instead of freezing it to uniform (ablation)FreeAugment · CIFAR100top-1 accuracy (%): 79.61 -> 80.04Improved
use Gumbel-Sinkhorn permutation sampling instead of independent Gumbel-Softmax per layer (ablation)FreeAugment · CIFAR10 WRN-40-2repeated transformation rateImproved
increase Sinkhorn iterations L up to 20 (ablation)FreeAugment · CIFAR100top-1 accuracy (%)Improved
sample a distinct augmentation per image instead of per batch (ablation)FreeAugment · CIFAR100top-1 accuracy (%): 79.43 -> 80.04Improved
add warm-up for policy parameters (ablation)FreeAugment · CIFAR100top-1 accuracy (%): 79.53 -> 80.04Improved
use uniform instead of Gaussian magnitude distribution (ablation)FreeAugment · CIFAR100top-1 accuracy (%): 79.9 -> 80.04No clear effect
learn depth with Gumbel-Softmax over depths instead of application probabilities (ablation)FreeAugment · CIFAR100top-1 accuracy (%): 79.11 -> 80.04Improved
train evaluation classifier from scratch instead of from end-of-search weights (ablation)FreeAugment · CIFAR100top-1 accuracy (%): 79.85 -> 80.04Improved
apply FreeAugment policy versus TA(Wide) on CIFAR100 WRN-28-10WRN-28-10 · CIFAR100top-1 accuracy (%): 84.33 -> 84.13Hurt
apply FreeAugment policy found on ImageNet-100 to full ImageNetResNet-50 · ImageNettop-1 accuracy (%): 76.3 -> 77.63Improved

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