Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Predicting Neural Network Accuracy from Weights

About

We show experimentally that the accuracy of a trained neural network can be predicted surprisingly well by looking only at its weights, without evaluating it on input data. We motivate this task and introduce a formal setting for it. Even when using simple statistics of the weights, the predictors are able to rank neural networks by their performance with very high accuracy (R2 score more than 0.98). Furthermore, the predictors are able to rank networks trained on different, unobserved datasets and with different architectures. We release a collection of 120k convolutional neural networks trained on four different datasets to encourage further research in this area, with the goal of understanding network training and performance better.

Thomas Unterthiner, Daniel Keysers, Sylvain Gelly, Olivier Bousquet, Ilya Tolstikhin• 2020

Related benchmarks

TaskDatasetResultRank
Generalization predictionSmallCNN Zoo CIFAR-10-GS ReLU (test)
Kendall's Tau0.914
6
Generalization predictionSmallCNN Zoo SVHN-GS ReLU (test)
Kendall's Tau0.8463
6
Generalization predictionSmallCNN Zoo CIFAR-10-GS Tanh (test)
Kendall's Tau0.914
6
Generalization predictionSmallCNN Zoo SVHN-GS Tanh (test)
Kendall's Tau0.844
6
Generalization predictionSmallCNN Zoo CIFAR-10-GS both activations ReLU Tanh (test)
Kendall's Tau91.5
6
Predicting image classifier test accuracySmall ResNet CIFAR-10 trained (test)
R^20.976
4
Predicting image classifier test accuracyW-Asymmetric ResNet CIFAR-10 (test)
R^20.978
4
Training dataset class predictionModel-J ResNet split
Accuracy63.1
4
Training dataset class predictionModel-J DINO
Accuracy51.1
4
Training dataset class predictionModel-J MAE
Accuracy0.502
4
Showing 10 of 18 rows

Other info

Follow for update