/SMCL

PyTorch implementation of Stochastic Multiple Choice Learning for Training Diverse Deep Ensembles

Primary LanguagePythonMIT LicenseMIT

SMCL

This is a PyTorch implementation of Stochastic Multiple Choice Learning for Training Diverse Deep Ensembles.

Results

We provide the results for image classification using CIFAR10 dataset. We have used ResNet-50 network for our experiments. Here ensemble size M = 5 and the number of predictors K = 1. The code can be easily extended for any dataset and experiments as well as for semantic segmentation and image captioning tasks.

Ensemble Size M Oracle Accuracy
1 95.42
2 97.71
3 98.11
4 98.40
5 98.79