Relative stability toward diffeomorphisms indicates performance in deep nets
| Main Authors: | Leonardo Petrini, Alessandro Favero, Mario Geiger, Matthieu Wyart |
|---|---|
| Format: | Proceeding Journal |
| Bahasa: | eng |
| Terbitan: |
, 2021
|
| Subjects: | |
| Online Access: |
https://zenodo.org/record/5589870 |
Daftar Isi:
- Code and pre-trained models used for the paper Relative stability toward diffeomorphisms indicates performance in deep nets, Petrini L. et al., NeurIPS2021. Fifteen different network architectures have been pre-trained on 4 benchmark datasets of images (MNIST, FashionMNIST, CIFAR10, SVHN) for different initialization seeds and train-set sizes for a total of 2'000+ pretrained models. Details on the available models can be found in the Pandas dataframe: pretrained_models_dataframe.pkl More details about the trainings are available in the reference paper (Appendix E) and at the GitHub repository: https://github.com/leonardopetrini/diffeo-sota For more info contact leonardo.petrini at epfl.ch