Orthogonalizing Convolutional Layers with the Cayley Transform

Authors: Asher Trockman, J Zico Kolter

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments have two goals: First, we show that our layer remains orthogonal in practice. Second, we compare the performance of our layer versus alternatives (particularly BCOP) on two adversarial robustness tasks on CIFAR-10
Researcher Affiliation Collaboration Asher Trockman Computer Science Department Carnegie Mellon University ashert@cs.cmu.edu J. Zico Kolter Computer Science Department Carnegie Mellon University & Bosch Center for AI zkolter@cs.cmu.edu
Pseudocode Yes Algorithm 1: Orthogonal convolution via the Cayley transform.
Open Source Code Yes Code is available at https://github.com/locuslab/orthogonal-convolutions.
Open Datasets Yes For all experiments, we used CIFAR-10 with standard augmentation, i.e., random cropping and flipping.
Dataset Splits No The paper mentions using CIFAR-10 but does not specify the training, validation, or test split percentages or sample counts.
Hardware Specification Yes Width 1 was on a Nvidia RTX 2080 Ti, while 2, 3, 6, and 8 were on a Nvidia Quadro RTX 8000.
Software Dependencies Yes In Py Torch 1.8, our layer can be implemented as follows.
Experiment Setup Yes For each architecture/convolution pair, we tried learning rates in {10 5, 10 4, 10 3, 10 2, 10 1}, choosing the one with the best test accuracy. Most often, 0.001 is appropriate. ... Adam (Kingma & Ba, 2014) showed a significant improvement over plain SGD, and we used it for all experiments. ... The batch size was fixed at 128 for all graphs