AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Let’s Agree to Agree: Neural Networks Share Classification Order on Real Datasets

Guy Hacohen, Leshem Choshen, Daphna Weinshall

ICML 2020 · Label-efficient and robust learning › Recognition and long-tail · approach: Loss design and classifier theory · 5 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
Train independently initialized networks of the same architecture on the same real datasetNeural networks on real classification benchmarks · ImageNet, CIFAR, MNIST, Fashion-MNIST, and Stack Overflowsimilarity of example-wise learning orderImproved
Use a stronger architecture that can continue learning after matching examples learned by the weaker modelNeural networks with different architectures · ImageNetsimilarity of learning order: 0.87Improved
Train on separate random partitions sampled from the same data distributionNeural networks of the same architecture · Shared test setsimilarity of accessibility-score orderImproved
Replace nonlinear layers with linear equivalents (ablation)st-VGG · Small mammalsaverage accuracy: 0.56 -> 0.43Hurt
Replace nonlinear layers with linear equivalents (ablation)st-VGG · Small mammalsshared learning-order behaviorNo clear effect
Train neural networks on synthetic Gabor-patch data instead of a real benchmark dataset (ablation)st-VGG · Gabor datasetshared learning-order behaviorHurt
Train on overlapping Gaussian classes (ablation)Fully connected neural networks · Gaussian classification taskshared learning-order behaviorHurt
Shuffle labels so that examples cannot generalize (ablation)st-VGG · Small mammals with shuffled labelsshared learning-order behavior: 1.07 train; 1.35 testHurt
Use AdaBoost with linear weak classifiers instead of neural networksCIFAR-10 classifiers · CIFAR-10learning-order correlation: 0.35Hurt
Give AdaBoost Inception-V3 penultimate-layer features (ablation)AdaBoost compared with st-VGG · CIFAR-10learning-order correlation: 0.35 -> 0.05Hurt
Vary optimization, initialization, batch size, dropout, architecture, or activation settings (ablation)Neural networks on real datasets · Multiple real benchmarksqualitative learning-order similarityNo clear effect

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