AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

Understanding Deep Contrastive Learning via Coordinate-wise Optimization

Yuandong Tian

NeurIPS 2022 · Learning theory and algorithms › Prediction and decision losses · approach: Loss landscape analysis · 4 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
backprop through alpha(theta) instead of coordinate-wise optimization (ablation)alpha-CL with InfoNCE alpha · CIFAR-10 500 epochstop-1 accuracy: 87.86 -> 87.48Hurt
directly set alpha from pairwise distances (alpha-CL-direct) vs InfoNCEResNet18 · CIFAR-10 100 epochstop-1 accuracy: 84.06 -> 85.11Improved
alpha-CL-direct vs InfoNCEResNet18 · STL-10 100 epochstop-1 accuracy: 78.46 -> 79.32Improved
alpha-CL-direct vs InfoNCEResNet18 · CIFAR-100 100 epochstop-1 accuracy: 55.70 -> 57.63Improved
alpha-CL-direct vs InfoNCEResNet50 · CIFAR-100 100 epochstop-1 accuracy: 60.16 -> 62.93Improved
alpha-CL-direct vs InfoNCE at 500 epochsResNet18 · CIFAR-10top-1 accuracy: 87.86 -> 88.09No clear effect
alpha-CL-direct vs InfoNCE at 500 epochsResNet50 · STL-10top-1 accuracy: 87.90 -> 87.85No clear effect
replace InfoNCE regularizer with r_Halpha-CL · CIFAR-10 500 epochstop-1 accuracy: 87.86 -> 87.92No clear effect
use squared regularizer r_s instead of InfoNCE (ablation)alpha-CL · STL-10 500 epochstop-1 accuracy: 83.70 -> 81.65Hurt
use quadratic loss instead of InfoNCEcontrastive loss · CIFAR-10 500 epochstop-1 accuracy: 87.86 -> 73.58Hurt
use r_gamma regularizer instead of InfoNCEalpha-CL · STL-10 500 epochstop-1 accuracy: 83.70 -> 83.47No clear effect
switch backbone ResNet18 to ResNet50InfoNCE · CIFAR-10 500 epochstop-1 accuracy: 87.86 -> 90.19Improved

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