The l2,1-Norm Stacked Robust Autoencoders for Domain Adaptation

Authors: Wenhao Jiang, Hongchang Gao, Fu-lai Chung, Heng Huang

AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experimental results demonstrate that the proposed method is very effective in multiple cross domain classification datasets which include Amazon review dataset, spam dataset from ECML/PKDD discovery challenge 2006 and 20 newsgroups dataset.
Researcher Affiliation Academia 1Department of Computer Science and Engineering, University of Texas at Arlington, Arlington, TX, USA 2Department of Computing, Hong Kong Polytechnic University, Hung Hom, Hong Kong, China
Pseudocode Yes Algorithm 1 ℓ2,1-norm Robust Autoencoder
Open Source Code No The paper does not provide any links to its own source code or explicitly state that it has been made publicly available.
Open Datasets Yes We test and analyze the proposed method on Amazon review dataset 2 (Blitzer, Dredze, and Pereira 2007), ECML/PKDD 2006 spam dataset 3 (Bickel 2008) and 20 newsgroups dataset 4. [Footnote 2: http://www.cs.jhu.edu/~mdredze/datasets/sentiment/, Footnote 3: http://www.ecmlpkdd2006.org/challenge.html, Footnote 4: http://qwone.com/~jason/20Newsgroups/]
Dataset Splits Yes Hence, we simply use a validation set containing a small number of labeled samples selected randomly from target domain to select parameters for feature learning algorithms.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments, such as GPU/CPU models or memory.
Software Dependencies No The paper mentions using a 'linear SVM (Chang and Lin 2011)' and discusses other methods but does not provide specific version numbers for any software dependencies needed for reproducibility.
Experiment Setup Yes There are three parameters in our method: the intensity of non-linear transformation α, the regularizer coefficient λ and the number of layers. We study the effects of these parameters on B D, Public U0 and Comp vs. Rec datasets. We fixed the number of layers and plotted the accuracies with different values of α and λ in Figure 1. The number of layers are 5, 3 and 3 for B D, Public U0 and Comp vs. Rec datasets respectively.