RoDA: Robust Domain Alignment for Cross-Domain Retrieval Against Label Noise
Authors: Ziniu Yin, Yanglin Feng, Ming Yan, Xiaomin Song, Dezhong Peng, Xu Wang
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The effectiveness and robustness of our Ro DA framework are comprehensively validated through extensive experiments across three multi-domain benchmarks. ... We evaluate the effectiveness of our Ro DA with all baselines on three datasets by conducting extensive NCIR experiments using the same network architecture. ... We conduct experiments on twelve cross-domain tasks with the Office-Home dataset (Venkateswara et al. 2017) to evaluate the contribution of each part within our framework. ... The experimental results demonstrate the effectiveness and robustness of our Ro DA across a spectrum of noise conditions within NCIR tasks. |
| Researcher Affiliation | Collaboration | 1The College of Computer Science, Sichuan University, Chengdu, China 2Tianfu Engineering-oriented Numerical Simulation & Software Innovation Center, Chengdu, China 3Centre for Frontier AI Research (CFAR), A*STAR, Singapore 4Sichuan National Innovation New Vision UHD Video Technology Co., Ltd., Chengdu, China |
| Pseudocode | No | The paper describes the methodology using mathematical equations and textual explanations (e.g., equations for Lwarmup, LI, Lcln, Vd, ˆyi, Wi, Lrlb, Lsdal, Md, Lacc), but it does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code https://github.com/yznovo/Ro DA |
| Open Datasets | Yes | The experiment utilized three datasets: Office-31 (Saenko et al. 2010), Office-Home (Venkateswara et al. 2017), Adaptiope (Ringwald and Stiefelhagen 2021). |
| Dataset Splits | No | The paper mentions applying various ratios of label noise (20%, 40%, 60%, and 80%) to datasets, and conducting 'cross-domain tasks on all domains of datasets' and 'twelve cross-domain tasks with the Office-Home dataset'. However, it does not specify explicit training, validation, or test splits for these datasets, nor does it refer to standard splits with citations for reproducibility beyond the initial dataset citations. |
| Hardware Specification | No | The paper states: "we utilize two models, both of which consist of a feature extractor and a classifier. Specifically, we take the ResNet-50 network pre-trained on ImageNet as the feature extractor and a fully connected layer as the classifier." This describes the model architecture, but no specific hardware (e.g., GPU, CPU models, memory) used for training or evaluation is provided. |
| Software Dependencies | No | The paper mentions using "ResNet-50 network pre-trained on ImageNet" and "Adam optimizer." It does not provide specific version numbers for these or any other software libraries, programming languages (e.g., Python, PyTorch/TensorFlow), or operating systems. |
| Experiment Setup | Yes | In the experiments, we utilize two models, both of which consist of a feature extractor and a classifier. Specifically, we take the Res Net-50 network pre-trained on Image Net as the feature extractor and a fully connected layer as the classifier. We adopt the Adam optimizer and set the learning rate to 1e-5. ... Besides, the training batch size and max epoch are set to 16 and 50 with all methods. The hyperparameters: β, γ, θ, α and λ are set to 2, 0.5, 0.7, 0.9 and 1.0 respectively. |