Collaborative and Adversarial Network for Unsupervised Domain Adaptation

By Weichen Zhang, Wanli Ouyang, Wen Li, Dong Xu. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018

485 citations recorded by the source corpus.

In this paper, we propose a new unsupervised domain adaptation approach called Collaborative and Adversarial Network (CAN) through domain-collaborative and domain-adversarial training of neural networks. We add several domain classifiers on multiple CNN feature extraction blocks1, in which each domain classifier is connected to the hidden representations from one block and one loss function is defined based on the hidden presentation and the domain labels (e.g., source and target). We design a new loss function by integrating the losses from all blocks in order to learn domain informative representations from lower blocks through collaborative learning and learn domain uninformative representations from higher blocks through adversarial learning. We further extend our CAN method as Incremental CAN (iCAN), in which we iteratively select a set of pseudo-labelled target samples based on the image classifier and the last domain classifier from the previous training epoch and re-train our CAN model by using the enlarged training set. Comprehensive experiments on two benchmark datasets Office and ImageCLEF-DA clearly demonstrate the effectiveness of our newly proposed approaches CAN and iCAN for unsupervised domain adaptation.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging

View Paper: https://www.semanticscholar.org/paper/b2b0e40f2bc62a6e87430e94ab6af9ddc67a5c72

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

In Computer Vision and Pattern Recognition

In Radiology, Nuclear Medicine and Imaging

Citations in the corpus

Unsupervised Domain Adaptation via Structurally Regularized Deep Clustering (2020, background): “A typical line of works [16, 43, 53, 63] leverages a domain-adversarial task to align the source and target domains as a whole so that class labels can be transferred from the source domain to the unlabeled target one.”
Universal Domain Adaptation (2019, background): “deep adaptation methods further explore architecture designs [19, 43, 36, 26, 20, 41, 16, 47, 25, 4, 18].”
Domain-Specific Batch Normalization for Unsupervised Domain Adaptation (2019, background): “Collaborative and Adversarial Network (CAN) [30] proposes a new loss function that enforces lower-level layers of the network to have high domain-specificity and low domain-invariance while making higher-level layers to have the opposite properties.”
Collaborative Learning for Faster StyleGAN Embedding (2020, unknown): “image classification [7, 3, 41], unsupervised domain adaptation [40], subspace clustering [39] and reinforcement learning [21].”
Deep Subdomain Adaptation Network for Image Classification (2020, methodology): “In addition, most state-of-the-art approaches [14], [29], [31] are domain adversarial net-based adaptation methods.”
Transfer Learning with Dynamic Adversarial Adaptation Network (2019, background): “We compare our proposed Dynamic Adversarial Adaptation Network (DAAN) with several state-of-the-art deep unsupervised domain adaptation methods: • Deep residual learning [1] • Deep Domain Confusion (DDC) [33] • Deep Adaptation Network (DAN) [32] • Residual Transfer Network (RTN) [41] • Domain Advers”
Do We Really Need to Access the Source Data? Source Hypothesis Transfer for Unsupervised Domain Adaptation (2020, methodology): “For DA methods, (Zhang et al., 2018b; Choi et al., 2019) directly incorporate pseudo labeling as a regularization while (Long et al., 2017; 2018) leverage pseudo labels in the adaptation module to pursue discriminative (conditional) distribution alignment.”
Racial Faces in the Wild: Reducing Racial Bias by Information Maximization Adaptation Network (2018, methodology): “Previous UDA methods apply the source classifier to predict pseudo-labels in the target domain, by which the network can be fine-tuned using supervised losses [51, 69, 16, 18, 66].”
Larger Norm More Transferable: An Adaptive Feature Norm Approach for Unsupervised Domain Adaptation (2018, methodology): “iCAN [50] iteratively applies sample selection on pseudo-labeled target samples and retrains the network.”
Cross-Domain Gradient Discrepancy Minimization for Unsupervised Domain Adaptation (2021, methodology): “Some previous studies [43, 6, 39] directly incorporate naive pseudo labeling strategy into their methods.”
On Data Augmentation for GAN Training (2020, background): “GAN has achieved remarkable results in various tasks, for example: image synthesis [2]–[5], image transformation [6]–[8], super-resolution [9], [10], text to image [11], [12], video captioning [13], image dehazing [14], domain adaptation [15], anomaly detection [16], [17].”
A review of applications in federated learning (2020, methodology): “Such method like Collaborative and Adversarial Network (CAN), a novel unsupervised domain adaptation approach, shows effectiveness and high performance (Zhang et al., 2018).”
DLOW: Domain Flow for Adaptation and Generalization (2018, background): “Domain adaptation aims to utilize a labeled source domain to learn a model that performs well on an unlabeled target domain [13, 18, 12, 55, 29, 3, 31, 16, 6, 61, 57].”
Cluster Alignment With a Teacher for Unsupervised Domain Adaptation (2019, background): “[50] successfully extend RevGrad [7] to consider each domain’s characteristics using collaborative games.”
Universal Source-Free Domain Adaptation (2020, background): “The available DA techniques heavily rely on the adversarial discriminative [46, 56, 38] strategy.”