AI Paper Atlas

Research map / Learning theory and algorithms / Kernels and regression theory

Model Similarity Mitigates Test Set Overuse

Horia Mania, John Miller, Ludwig Schmidt, Moritz Hardt, Benjamin Recht

NeurIPS 2019 · Learning theory and algorithms › Kernels and regression theory · approach: Statistical perspectives on generalization · 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
Use measured pairwise model similarity instead of treating model losses as independentImageNet ILSVRC models · ImageNet ILSVRCpairwise loss agreement: 0.62 -> 0.85Improved
Use a similarity-based refined union bound rather than the standard union boundImageNet model test-set reuse · ImageNet validationtestable models: 4 \times improvementImproved
Use pairwise similarity to bound test-set reuseDARTS random configurations · CIFAR-10increase in testable models: 9.9 \timesImproved
Assume Naive Bayes structure in addition to using similarityDARTS random configurations · CIFAR-10increase in testable models: 1.6 \cdot 10^4 \timesImproved
Use pairwise similarity to bound test-set reuseDARTS highest-scoring configurations · CIFAR-10increase in testable models: 12.0 \timesImproved
Assume Naive Bayes structure in addition to using similarityDARTS highest-scoring configurations · CIFAR-10increase in testable models: 3.4 \cdot 10^4 \timesImproved
Account for model similarity in the theoretical boundNon-adaptive evaluation of classifiers · General classificationoverfitting guaranteeImproved

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