Deep Subdomain Adaptation Network for Image Classification

By Yongchun Zhu, Fuzhen Zhuang, Jindong Wang, Guolin Ke, Jingwu Chen, Jiang Bian, Hui Xiong, Qing He. 2020

1,091 citations recorded by the source corpus.

For a target task where the labeled data are unavailable, domain adaptation can transfer a learner from a different source domain. Previous deep domain adaptation methods mainly learn a global domain shift, i.e., align the global source and target distributions without considering the relationships between two subdomains within the same category of different domains, leading to unsatisfying transfer learning performance without capturing the fine-grained information. Recently, more and more researchers pay attention to subdomain adaptation that focuses on accurately aligning the distributions of the relevant subdomains. However, most of them are adversarial methods that contain several loss functions and converge slowly. Based on this, we present a deep subdomain adaptation network (DSAN) that learns a transfer network by aligning the relevant subdomain distributions of domain-specific layer activations across different domains based on a local maximum mean discrepancy (LMMD). Our DSAN is very simple but effective, which does not need adversarial training and converges fast. The adaptation can be achieved easily with most feedforward network models by extending them with LMMD loss, which can be trained efficiently via backpropagation. Experiments demonstrate that DSAN can achieve remarkable results on both object recognition tasks and digit classification tasks. Our code will be available at https://github.com/easezyc/deep-transfer-learning.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Jindong Wang
2. Dongrui Wu
3. Danfeng Hong
4. Qian Du
5. Tao Qin
6. Wei Li
7. Wenjie Feng
8. Xiaojun Chang

Citations in the corpus

• A Comprehensive Survey of Neural Architecture Search (2020, background): “completely different from the artificial network, so as to break through the existing human mindset. Artificially designed networks have made breakthroughs in many fields, including image recognition [4, 7, 112], machine translation [1, 3, 141, 149], semantic segmentation [142, 143, 148], object det”
• AdaRNN: Adaptive Learning and Forecasting of Time Series (2021, background): “DA often performs instance re-weighting or feature transfer to reduce the distribution divergence in training and test data [12, 44, 47, 48, 56].”
• Graph Information Aggregation Cross-Domain Few-Shot Learning for Hyperspectral Image Classification (2022, methodology): “At present, many deep learning methods design adaptive layers and metric criteria to reduce the deviation of SD and TD data distribution [16]–[18].”
• Generalizing to Unseen Domains: A Survey on Domain Generalization (2021, methodology): “Some methods explicitly minimized the feature distribution divergence by minimizing the maximum mean discrepancy (MMD) [181], [188], [189], [190], second order correlation [191], [192], [193], both mean and variance (moment matching) [183], Wasserstein distance [182], etc, of domains for either doma”
• A Survey on Negative Transfer (2020, methodology): “Similar ideas were also investigated in [73], [76].”
• Single-Source Domain Expansion Network for Cross-Scene Hyperspectral Image Classification (2022, methodology): “To evaluate the performance of SDEnet with only SD used for training, relevant algorithms including DAAN, MRAN, DSAN, HTCNN, PDEN, LDSDG and SagNet are used for comparison.”
• Cross-City Matters: A Multimodal Remote Sensing Benchmark Dataset for Cross-City Semantic Segmentation using High-Resolution Domain Adaptation Networks (2023, background): “They are DeepLabv3 [23], SegNet [17], FastFCN [16], AdaptSeg [29], deep subdomain adaptation network (DSAN) [31], Dual-stream HR-Net (DualHR) [24], SegFormer [12], and our proposed HighDAN.”