AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Feature Dropout: Revisiting the Role of Augmentations in Contrastive Learning

Alex Tamkin, Margalit Glasgow, Xiluo He, Noah Goodman

NeurIPS 2023 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Inductive bias and minimal representations · 1 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 learned viewmaker augmentations instead of expert SimCLR augmentationsSimCLR ResNet-18 · C+Digit, CIFAR-10 transferaccuracy: 58.8 -> 69.3Improved
use viewmaker instead of expert augmentationsSimCLR ResNet-18 · C+Shape, CIFAR-10 transferaccuracy: 76.0 -> 79.8Improved
use viewmaker instead of expert augmentationsSimCLR ResNet-18 · C+Digit, digit transferaccuracy: 86.7 -> 94.3Improved
use viewmaker instead of expert augmentationsSimCLR ResNet-18 · C+Letter, letter transferaccuracy: 94.1 -> 96.9Improved
use viewmaker instead of expert augmentationsSimCLR ResNet-18 · C+Letter, CIFAR-10 transferaccuracy: 74.8 -> 71.9Hurt
use viewmaker instead of expert augmentationsSimCLR ResNet-18 · CIFAR-10 only, CIFAR-10 transferaccuracy: 86.2 -> 84.5Hurt
use viewmaker instead of spectral expert viewsSimCLR ResNet-18 · Speech + Sound pretrain, speech accuracyaccuracy: 10.1 -> 60.8Improved
use viewmaker instead of waveform expert viewsSimCLR ResNet-18 · Speech + Sound pretrain, sound accuracyaccuracy: 43.3 -> 97.0Improved
use spectral expert views instead of viewmakerSimCLR ResNet-18 · Speech only, speech accuracyaccuracy: 92.4 -> 97.0Improved
mask viewmaker perturbation so it cannot destroy the object label (ablation)Viewmaker · C+Shape, CIFAR-10 transferaccuracy: 79.8 -> 26.0Hurt
mask viewmaker perturbation so it cannot destroy the object label (ablation)Viewmaker · C+Digit, CIFAR-10 transferaccuracy: 69.3 -> 50.7Hurt
mask viewmaker perturbation so it cannot destroy the object label (ablation)Viewmaker · C+Letter, CIFAR-10 transferaccuracy: 71.9 -> 23.2Hurt

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