Domain Invariant Representation Learning with Domain Density Transformations

Authors: A. Tuan Nguyen, Toan Tran, Yarin Gal, Atilim Gunes Baydin

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the effectiveness of our method on several widely used datasets for the domain generalization problem, on all of which we achieve competitive results with state-of-the-art models.
Researcher Affiliation Collaboration A. Tuan Nguyen University of Oxford; Vin AI Research Oxford, United Kingdom tuan@robots.ox.ac.uk Toan Tran Vin AI Research Hanoi, Vietnam v.toantm3@vinai.io Yarin Gal University of Oxford Oxford, United Kingdom yarin@cs.ox.ac.uk Atilim Gunes Baydin University of Oxford Oxford, United Kingdom gunes@robots.ox.ac.uk
Pseudocode No The paper describes the theoretical approach and practical implementation using GANs but does not include any explicit pseudocode or algorithm blocks.
Open Source Code Yes Our code is available at https://github.com/atuannguyen/DIRT.
Open Datasets Yes Rotated MNIST [18]: In this dataset, 1,000 MNIST images (100 per class) [29] are chosen to form the first domain (denoted M0), then counter-clockwise rotations of 15◦, 30◦, 45◦, 60◦ and 75◦ are applied to create five additional domains, denoted M15, M30, M45, M60 and M75. The task is classification with ten classes (digits 0 to 9). [29] Y. Le Cun and C. Cortes. MNIST handwritten digit database. 2010. URL http://yann.lecun. com/exdb/mnist/.
Dataset Splits Yes Following standard practice, we use 90% of available data as training data and 10% as validation data, except for the Rotated MNIST experiment where we do not use a validation set and just report the performance of the last epoch.
Hardware Specification Yes We train our model on a NVIDIA Quadro RTX 6000.
Software Dependencies No The paper mentions optimizers (Adam, SGD), network architectures (Alexnet, Resnet18), and the Star GAN model, but does not provide specific version numbers for any software dependencies or libraries.
Experiment Setup Yes We train our network for 500 epochs with the Adam optimizer [26], using the learning rate 0.001 and minibatch size 64... We train the network for 100 epochs with plain stochastic gradient descent (SGD) using learning rate 0.001, momentum 0.9, minibatch size 64, and weight decay 0.001.