Wasserstein Distance Guided Representation Learning for Domain Adaptation

Authors: Jian Shen, Yanru Qu, Weinan Zhang, Yong Yu

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical studies on common sentiment and image classification adaptation datasets demonstrate that our proposed WDGRL outperforms the state-of-the-art domain invariant representation learning approaches.
Researcher Affiliation Academia Jian Shen, Yanru Qu, Weinan Zhang, Yong Yu Shanghai Jiao Tong University {rockyshen, kevinqu, wnzhang, yyu}@apex.sjtu.edu.cn
Pseudocode Yes A detailed algorithm of the combination is given in Algorithm 1.
Open Source Code Yes Experiment code: https://github.com/RockySJ/WDGRL.
Open Datasets Yes Amazon review benchmark dataset. The Amazon review dataset2 (Blitzer et al. 2007) is one of the most widely used benchmarks for domain adaptation and sentiment analysis. It is collected from product reviews from Amazon.com and contains four types (domains), namely books (B), DVDs (D), electronics (E) and kitchen appliances (K).
Dataset Splits No The paper mentions total labeled and unlabeled data amounts per domain (e.g., '2,000 labeled reviews and approximately 4,000 unlabeled reviews') and refers to an 'evaluation protocol in (Long et al. 2013)', but does not explicitly provide specific numerical training/validation/test splits (percentages or sample counts) within the text.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory, or cloud instances) used for running experiments are mentioned in the paper.
Software Dependencies No The paper mentions using 'Tensor Flow' and 'Adam optimizer' but does not provide specific version numbers for these software components.
Experiment Setup Yes For each approach we use a batch size of 64 samples in total with 32 samples from each domain, and a fixed learning rate 10 4. All compared approaches are combined with a discriminator to learn both domain invariant and discriminative representations and to conduct the classification task. We use standard multi-layer perceptron (MLP) as the basic network architecture. MLP is sufficient to handle all the problems in our experiments. For Amazon review dataset the network is designed with one hidden layer of 500 nodes, relu activation function and softmax output function, while the network for Office-Caltech dataset has two hidden layers of 500 and 100 nodes.