Instilling Social to Physical: Co-Regularized Heterogeneous Transfer Learning
Authors: Ying Wei, Yin Zhu, Cane Leung, Yangqiu Song, Qiang Yang
AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our proposed method outperforms state-of-the-art methods on two ubiquitous computing tasks, namely human activity recognition and region function discovery. Experiments We verify the effectiveness of social knowledge transfer on two ubiquitous computing tasks, namely human activity recognition and region function discovery. In the activity recognition task, we collected 232 sensor records through cellphones from 10 volunteers. |
| Researcher Affiliation | Collaboration | Hong Kong University of Science and Technology, Hong Kong Wisers Research, Hong Kong West Virginia University, Morgantown, WV, USA |
| Pseudocode | No | The optimization process is described through iterative steps like 'Fix U, V1, V2:', 'Fix W, V1, V2:', 'Fix U, W, V2:', and 'Fix U, W, V1:', including mathematical equations for gradients and solutions, but it is not presented in a formal pseudocode block or labeled as an algorithm. |
| Open Source Code | No | The paper does not provide any statement or link regarding the open-source availability of the code for the described methodology. |
| Open Datasets | No | In the activity recognition task, we collected 232 sensor records through cellphones from 10 volunteers. In the region function discovery task, our dataset is a collection of taxis trajectories generated by 182 taxis within one month in a big city of South China. We obtain social knowledge from Sina Weibo, a Twitter-like microblogging service in China. The full dataset contains tweets from about 10 million users. The paper does not provide links or clear indications of public availability for these datasets. |
| Dataset Splits | Yes | The trade-off parameter C of linear SVM is set according to 10-fold cross validation for each model. In this experiment, we randomly sample 100 sensor records as training samples to perform 9-class classification, and the other 132 sensor records as test data. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory) used to conduct the experiments. It only mentions 'cellphones' in the context of data collection, not the computational environment. |
| Software Dependencies | No | The paper mentions using 'linear SVM (Chang and Lin 2011)', 'word2vec (Mikolov et al. 2013)', 'Latent Dirichlet Allocation (LDA) (Blei, Ng, and Jordan 2003)', and 't-SNE (Van der Maaten and Hinton 2008)', citing the original works. However, it does not specify version numbers for these or any other software libraries or dependencies used in the implementation. |
| Experiment Setup | Yes | The trade-off parameter C of linear SVM is set according to 10-fold cross validation for each model. We perform grid search on β by fixing u. Co HTL gains the best average accuracy at β = 0.1 as Figure 6(a) shows. We adopt β = 0.1, u = 300 in our experiments. |