Return of Frustratingly Easy Domain Adaptation
Authors: Baochen Sun, Jiashi Feng, Kate Saenko
AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our method on object recognition (Saenko et al. 2010) and sentiment analysis (Blitzer, Dredze, and Pereira 2007) with both shallow and deep features, using standard benchmarks and protocols. In all experiments we assume the target domain is unlabeled. We follow the standard procedure (Fernando et al. 2013; Donahue et al. 2014) and use a linear SVM as the base classifier. |
| Researcher Affiliation | Academia | Baochen Sun Department of Computer Science University of Massachusetts Lowell Lowell, MA 01854, USA bsun@cs.uml.edu Jiashi Feng Department of EECS, UC Berkeley, USA & Department of ECE, National University of Singapore, Singapore elefjia@nus.edu.sg Kate Saenko Department of Computer Science University of Massachusetts Lowell Lowell, MA 01854, USA saenko@cs.uml.edu |
| Pseudocode | Yes | Algorithm 1 CORAL for Unsupervised Domain Adaptation |
| Open Source Code | No | The paper describes its algorithm and mentions that it can be implemented in 'four lines of MATLAB code' (Algorithm 1), but it does not provide any link or explicit statement for the public release of its source code. |
| Open Datasets | Yes | We evaluate our method on object recognition (Saenko et al. 2010) and sentiment analysis (Blitzer, Dredze, and Pereira 2007) with both shallow and deep features, using standard benchmarks and protocols. [...] standard Office (Saenko et al. 2010) and extended Office-Caltech10 (Gong et al. 2012) datasets [...] Cross-Dataset Testbed (Tommasi and Tuytelaars 2014) dataset [...] standard Amazon review dataset (Blitzer, Dredze, and Pereira 2007; Gong, Grauman, and Sha 2013) |
| Dataset Splits | Yes | The model selection approach of (Fernando et al. 2013) is used to set the C parameter for the SVM by doing cross-validation on the source domain. [...] In each trial, we use the standard setting (...) and randomly sample the same number (20 for Amazon, Caltech, and Webcam; 8 for DSLR as there are only 8 images per category in the DSLR domain) of labelled images in the source domain as training set, and use all the unlabelled data in the target domain as the test set. |
| Hardware Specification | Yes | The Tesla K40 used for this research was donated by the NVIDIA Corporation. |
| Software Dependencies | No | The paper states that the algorithm can be implemented in 'four lines of Matlab code' and mentions other tools like 'Alex Net' and 'linear SVM' but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | The model selection approach of (Fernando et al. 2013) is used to set the C parameter for the SVM by doing cross-validation on the source domain. [...] In this paper, we set λ to 1. |