AI Paper Atlas

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

CaCo: Both Positive and Negative Samples are Directly Learnable via Cooperative-Adversarial Contrastive Learning

Xiao Wang, Yuhang Huang, Dan Zeng, Guo-Jun Qi

TPAMI 2023 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Inductive bias and minimal representations · 31 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
take positives from the shared memory bank in addition to negatives (ablation)CaCo (batch 1024, 200 epochs) · ImageNet-1k linear evaltop-1 accuracy: 70.6 -> 71.3Improved
take negatives from the adversarially trained memory bank in addition to bank positives (ablation)CaCo (batch 1024, 200 epochs) · ImageNet-1k linear evaltop-1 accuracy: 70.7 -> 71.3Improved
use a memory bank for positives only vs none (ablation)CaCo (batch 1024, 200 epochs) · ImageNet-1k linear evaltop-1 accuracy: 70.2 -> 70.7Improved
use a memory bank for negatives only vs none (ablation)CaCo (batch 1024, 200 epochs) · ImageNet-1k linear evaltop-1 accuracy: 70.2 -> 70.6Improved
increase memory bank from 4M to 12M parameters (37268 to 98304 negatives) (ablation)CaCo (batch 4096, 200 epochs) · ImageNet-1k linear evaltop-1 accuracy: 71.7 -> 72.1Improved
increase batch size from 256 to 4096 (ablation)CaCo (200 epochs) · ImageNet-1k linear evaltop-1 accuracy: 70.9 -> 72.0Improved
replace AdCo adversarial-only memory bank with cooperative-adversarial bankAdCo · ImageNet-1k linear eval (200 epochs, batch 256)top-1 accuracy: 68.2 -> 70.9Improved
train longer, 200 to 800 epochs at batch 4096CaCo · ImageNet-1k linear evaltop-1 accuracy: 72.0 -> 75.3Improved
add multi-crop augmentation, batch 2048CaCo (800 epochs) · ImageNet-1k linear evaltop-1 accuracy: 75.3 -> 75.7Improved
compare CaCo to NNCLR under single crop at batch 4096NNCLR · ImageNet-1k linear eval (800 epochs)top-1 accuracy: 74.9 -> 75.3Hurt
use CaCo features instead of NNCLR for transferNNCLR · Transfer to CIFAR100linear accuracy: 79.0 -> 76.0Mixed
use CaCo features instead of MoCo v2 for transferMoCo v2 · Transfer to Aircraftlinear accuracy: 50.9 -> 66.0Improved

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