Scale-Equivariant Steerable Networks

Authors: Ivan Sosnovik, Michał Szmaja, Arnold Smeulders

ICLR 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate state-of-the-art results on the MNIST-scale dataset and on the STL-10 dataset in the supervised learning setting. In this section we conduct the experiments and compare various methods for working with scale variations in input data.
Researcher Affiliation Collaboration Ivan Sosnovik , Michał Szmaja , Arnold Smeulders Uv A-Bosch Delta Lab University of Amsterdam
Pseudocode No We illustrate all algorithms in Figure 1.
Open Source Code Yes Source code is available at http://github.com/isosnovik/sesn
Open Datasets Yes We conduct the experiments on the MNIST-scale dataset. We rescale the images of the MNIST dataset Le Cun et al. (1998)...
Dataset Splits Yes The obtained dataset is then split into 10,000 for training, 2,000 for evaluation and 50,000 for testing.
Hardware Specification Yes We used 1 Nvidia Ge Force GTX 1080Ti GPU for training the models.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes All models are trained with the Adam optimizer Kingma & Ba (2014) for 60 epochs with a batch size of 128. Initial learning rate is set to 0.01 and divided by 10 after 20 and 40 epochs. All models are trained for 1000 epochs with a batch size of 128. We use SGD optimizer with Nesterov momentum of 0.9 and weight decay of 5 10 4. The initial learning rate is set to 0.1 and divided by 5 after 300, 400, 600 and 800 epochs.