Cooperative and Adversarial Learning: Co-enhancing Discriminability and Transferability in Domain Adaptation
Authors: Hui Sun, Zheng Xie, Xin-Ye Li, Ming Li
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments to show the effectiveness of the proposed CALE framework. ... Experimental results on Office-Home and Vis DA-2017 are shown in Table 1, and Table 2, respectively. ... The bolded and underlined numbers denote the best and the second best performance. |
| Researcher Affiliation | Academia | Hui Sun, Zheng Xie, Xin-Ye Li, Ming Li National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing 210023, China {sunh,xiez,lixy,lim}@lamda.nju.edu.cn |
| Pseudocode | Yes | Algorithm 1: Cooperative and Adversarial Learning (CALE) |
| Open Source Code | Yes | More details about code and datasets can be found at https://github.com/sunh-23/CALE. |
| Open Datasets | Yes | We evaluate the CALE model and the baselines on three benchmark visual datasets: Office-31 contains 4110 images from 31 categories of three distant domains, including Amazon (A), Webcam (W), and DSLR (D). Office Home, a more challenging dataset, consists of 15588 images of 65 object classes in office and home environments, forming four extremely dissimilar domains: Artistic (Ar), Clip Art (Cl), Product (Pr), and Real World (Rw). Vis DA-2017, a large dataset with 152397 Synthetic 3D rendered images and 55388 Real-world photos across 12 categories. |
| Dataset Splits | No | The paper mentions using specific datasets (Office-31, Office Home, Vis DA-2017) and self-training, but it does not provide explicit details about the proportions or counts of training, validation, and test splits for these datasets. |
| Hardware Specification | No | The paper does not specify the hardware used for running experiments, such as GPU or CPU models, memory, or cloud computing instances. |
| Software Dependencies | No | The paper mentions using Res Net, Dei T, SGD, and Fix Match, but it does not provide specific version numbers for these software components or any other libraries and frameworks used for implementation. |
| Experiment Setup | Yes | Throughout the experiments, the trade-off parameter of CALE regularization λ is set to 1, and the self-training threshold τ in Equation 8 is set to 0.95. The distance function Dist. is set as cross-entropy in Office-31 and Office-Home, KL-divergence in Vis DA-2017. |