Deep Learning Through the Lens of Example Difficulty

Authors: Robert Baldock, Hartmut Maennel, Behnam Neyshabur

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To ensure that our results are robust to the choice of architectures and datasets, we report empirical findings for Res Net18 (He et al., 2016), VGG16 (Simonyan and Zisserman, 2015) and MLP architectures trained on CIFAR10, CIFAR100 (Krizhevsky et al., 2009), Fashion MNIST (FMNIST) (Xiao et al., 2017) and SVHN (Netzer et al., 2011) datasets.
Researcher Affiliation Industry Robert J. N. Baldock Google Research, Brain Team rjnbaldock@gmail.com Hartmut Maennel Google Research, Brain Team hartmutm@google.com Behnam Neyshabur Google Research, Blueshift Team neyshabur@google.com
Pseudocode No No explicitly labeled pseudocode or algorithm blocks were found.
Open Source Code No The paper does not provide a statement about releasing code or a link to a code repository for the described methodology.
Open Datasets Yes report empirical findings for Res Net18 (He et al., 2016), VGG16 (Simonyan and Zisserman, 2015) and MLP architectures trained on CIFAR10, CIFAR100 (Krizhevsky et al., 2009), Fashion MNIST (FMNIST) (Xiao et al., 2017) and SVHN (Netzer et al., 2011) datasets.
Dataset Splits Yes We trained 250 Res Net18 models on CIFAR10, with 90:10% random train:validation splits as described in Appendix A.
Hardware Specification No All calculations were performed using Google s computer infrastructure.
Software Dependencies No All models were trained using SGD with momentum. Our MLP comprises 7 hidden layers of width 2048 with Re LU activations.
Experiment Setup Yes All models were trained using SGD with momentum. Our MLP comprises 7 hidden layers of width 2048 with Re LU activations. Details of the datasets, architectures, and hyperparameters used can be found in Appendix A.