Discriminative Adversarial Domain Adaptation

Authors: Hui Tang, Kui Jia5940-5947

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments show the efficacy of our proposed methods and we achieve the new state of the art for all the three settings on benchmark datasets.
Researcher Affiliation Academia Hui Tang, Kui Jia South China University of Technology eehuitang@mail.scut.edu.cn, kuijia@scut.edu.cn
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes We implement all our methods by Py Torch. The code will be available at https://github.com/huitangtang/DADA-AAAI2020.
Open Datasets Yes Office-31 (Saenko et al. 2010) is a popular benchmark domain adaptation dataset... Syn2Real (Peng et al. 2018) is the largest benchmark.
Dataset Splits Yes We follow standard evaluation protocols for unsupervised domain adaptation (Ganin et al. 2016; Wang et al. 2019): we use all labeled source and all unlabeled target instances as the training data. ... We use the training domain as the source domain and validation one as the target domain.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper states 'We implement all our methods by Py Torch' but does not provide specific version numbers for PyTorch or any other software dependencies.
Experiment Setup Yes We follow DANN (Ganin et al. 2016) to use the SGD training schedule: the learning rate is adjusted by ηp = η0 (1+αp)β , where p denotes the process of training iterations that is normalized to be in [0, 1], and we set η0 = 0.0001, α = 10, and β = 0.75; the hyper-parameter λ is initialized at 0 and is gradually increased to 1 by λp = 2 1+exp( γp) 1, where we set γ = 10. We empirically set q = 0.1.