HoMM: Higher-Order Moment Matching for Unsupervised Domain Adaptation
Authors: Chao Chen, Zhihang Fu, Zhihong Chen, Sheng Jin, Zhaowei Cheng, Xinyu Jin, Xian-sheng Hua3422-3429
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments are conducted, showing that our proposed Ho MM consistently outperforms the existing moment matching methods by a large margin. |
| Researcher Affiliation | Collaboration | 1Zhejiang University, 2Alibaba DAMO Academy, Alibaba Group |
| Pseudocode | No | The paper describes its methods using mathematical equations and textual explanations, but it does not provide any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Codes are available at https://github.com/chenchao666/Ho MM-Master |
| Open Datasets | Yes | We conduct experiments on three public visual adaptation datasets: digits recognition dataset, Office-31 dataset, and Office-Home dataset. The digits recognition dataset includes four widely used benchmarks: MNIST, USPS, Street View House Numbers (SVHN), and SYN (synthetic digits dataset). |
| Dataset Splits | No | The paper mentions using 'all the labeled source domain samples and all the unlabeled target domain samples for training' and refers to hyperparameter selection, but it does not explicitly provide specific training/validation/test splits with percentages, sample counts, or explicit instructions for reproducibility of splits beyond using standard dataset configurations. |
| Hardware Specification | No | The paper mentions using 'modified Le Net' and 'Res Net-50' as backbone networks, but it does not provide specific details about the hardware (e.g., CPU/GPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper states, 'Our model is trained with Adam Optimizer based on Tensorflow.' However, it does not specify the version number for TensorFlow or any other software dependencies. |
| Experiment Setup | Yes | Specifically, the trade-off parameters are selected from λd = {1, 10, 102, ..., 108}, λdc {0.01, 0.03, 0.1, 0.3, 1.0}. For the digits recognition tasks, the hyper-parameter λd is set to 104 for third-order Ho MM and set to 107 for fourth-order Ho MM. For the experiments on Office-31 and Office-Home, λd is set to 300 for the third-order Ho MM and set to 3000 for the fourth-order Ho MM. Besides, the hyper-parameter γ in RBF kernel is set to 1e-4 across the experiments, the learning rate of the centers is set to α = 0.5 for digits dataset and set to α = 0.3 for Office-31 and Office-Home dataset. The threshold η of the predicted probability is chosen from {0.6, 0.65, 0.7, 0.75, 0.8, 0.85, 0.9, 0.95}. ... we set the batch size as 128 for each domain and set the learning rate as 1e-4 throughout the experiments. ... We set the batchsize as 70 for each domain. The learning rate of the fc layer parameters is set as 3e-4 and the learning rate of the conv layer (scale5/block3) parameters is set as 3e-5. |