Do We Really Need to Access the Source Data? Source Hypothesis Transfer for Unsupervised Domain Adaptation

By Jian Liang, Dapeng Hu, Jiashi Feng. 2020

1,724 citations recorded by the source corpus.

Unsupervised domain adaptation (UDA) aims to leverage the knowledge learned from a labeled source dataset to solve similar tasks in a new unlabeled domain. Prior UDA methods typically require to access the source data when learning to adapt the model, making them risky and inefficient for decentralized private data. This work tackles a practical setting where only a trained source model is available and investigates how we can effectively utilize such a model without source data to solve UDA problems. We propose a simple yet generic representation learning framework, named \emph{Source HypOthesis Transfer} (SHOT). SHOT freezes the classifier module (hypothesis) of the source model and learns the target-specific feature extraction module by exploiting both information maximization and self-supervised pseudo-labeling to implicitly align representations from the target domains to the source hypothesis. To verify its versatility, we evaluate SHOT in a variety of adaptation cases including closed-set, partial-set, and open-set domain adaptation. Experiments indicate that SHOT yields state-of-the-art results among multiple domain adaptation benchmarks.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging, Representation & Self-Supervised Learning

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

In Representation & Self-Supervised Learning

Citations in the corpus

Refining Language Models with Compositional Explanations (2021, methodology): “, due to privacy constraints [27], or access restriction [29]) but unlabeled data in the target domain (DT ) are readily accessible, during model refinement.”
Prototype-Guided Continual Adaptation for Class-Incremental Unsupervised Domain Adaptation (2022, methodology): “Step 3 [20,21,22,23,24,25,26,27,28,29] Postit Notes, Mug, Webcam, Desk Lamp, Telephone, Helmet, Mouse, Pen, Monitor, Mop”
Class-incremental Novel Class Discovery (2022, background): “Being meaningful, such source-free model adaptation has been explored in the related areas of domain adaptation [28,35].”
Novel Class Discovery in Semantic Segmentation (2021, background): “Entropy measurement is one of the key approaches to address uncertainty, which plays an important role in utilizing unlabeled target domain in UDA [18, 21, 33].”
Algorithm fairness in artificial intelligence for medicine and healthcare (2023, methodology): “Federated learning applied to multi-site domain adaptation across distributed clients would naturally mitigate many instances of dataset shift 232–236 .”
Sketch3T: Test-Time Training for Zero-Shot SBIR (2022, result): “Similar notion has been used in few-shot learning [55], domain generalization [9], and unsupervised domain adaptation [27,57].”
Efficient Test-Time Model Adaptation without Forgetting (2022, methodology): “Moreover, we propose an anti-forgetting regularizer to prevent the model parameters Θ changing too much from Θ o . modeling (Li et al., 2020; Kundu et al., 2020; Qiu et al., 2021) or information maximization (Liang et al., 2020).”
ViDA: Homeostatic Visual Domain Adapter for Continual Test Time Adaptation (2023, background): “Test-time adaptation (TTA), also referred to as source-free domain adaptation [6, 35, 41, 62], aims to adapt a source model to an unknown target domain distribution without relying on any source domain data.”
A Comprehensive Survey on Test-Time Adaptation Under Distribution Shifts (2023, neutral): “Furthermore, SFDA-IT [104] employs entropy minimization over the fixed source model to promote semantic consistency.”
Test-Time Classifier Adjustment Module for Model-Agnostic Domain Generalization (2021, methodology): “(5) SHOT [34] updates feature extractor to minimize entropy, diversity regularizer, and pseudo-label loss.”
Cross-Domain Gradient Discrepancy Minimization for Unsupervised Domain Adaptation (2021, methodology): “Inspired by DeepCluster [2] and SHOT [22], in this paper, we use a weighted clustering strategy to obtain more reliable pseudo labels.”
Overcoming Negative Transfer: A Survey (2020, background): “Some existing works actually follow this protocol [103], [104].”
Test-Time Prompt Tuning for Zero-Shot Generalization in Vision-Language Models (2022, background): “Another choice is to update the feature extractor while freezing the prediction module [41, 48].”
Give Me Your Trained Model: Domain Adaptive Semantic Segmentation without Source Data (2021, methodology): “Furthermore, to ensure the global diversity of the target outputs, we adopt a diversity-promoting loss speciﬁed in [28] and propose a conﬁdence-weighted diversity objective as below, . λ is a hyper-parameter and we empirically set λ = 3 in all experiments.”
Exploiting the Intrinsic Neighborhood Structure for Source-free Domain Adaptation (2021, methodology): “SHOT [21] and 3C-GAN [20] are for closed-set DA where source and target domains have the same categories.”