On Universal Black-Box Domain Adaptation

Abstract

In this paper, we study an arguably least restrictive setting of domain adaptation in a sense of practical deployment, where only the interface of source model is available to the target domain, and where the label-space relations between the two domains are allowed to be different and unknown. We term such a setting as Universal Black-Box Domain Adaptation (UB${}^2$DA). The great promise that UB${}^2$DA makes, however, brings significant learning challenges, since domain adaptation can only rely on the predictions of unlabeled target data in a partially overlapped label space, by accessing the interface of source model. To tackle the challenges, we first note that the learning task can be converted as two subtasks of in-class\footnote{In this paper we use in-class (out-class) to describe the classes observed (not observed) in the source black-box model.} discrimination and out-class detection, which can be respectively learned by model distillation and entropy separation. We propose to unify them into a self-training framework, regularized by consistency of predictions in local neighborhoods of target samples. Our framework is simple, robust, and easy to be optimized. Experiments on domain adaptation benchmarks show its efficacy. Notably, by accessing the interface of source model only, our framework outperforms existing methods of universal domain adaptation that make use of source data and/or source models, with a newly proposed (and arguably more reasonable) metric of H-score, and performs on par with them with the metric of averaged class accuracy.

References in corpus (26)

• Distilling the Knowledge in a Neural Network
• Deep Domain Confusion: Maximizing for Domain Invariance
• MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks on Corrupted Labels
• Adversarial Discriminative Domain Adaptation
• Mutual Mean-Teaching: Pseudo Label Refinery for Unsupervised Domain Adaptation on Person Re-identification
• Wasserstein Distance Guided Representation Learning for Domain Adaptation
• How does Disagreement Help Generalization against Label Corruption?
• Do We Really Need to Access the Source Data? Source Hypothesis Transfer for Unsupervised Domain Adaptation
• Using Trusted Data to Train Deep Networks on Labels Corrupted by Severe Noise
• Making Deep Neural Networks Robust to Label Noise: a Loss Correction Approach
• Curriculum Loss: Robust Learning and Generalization against Label Corruption
• Unsupervised Multi-Class Domain Adaptation: Theory, Algorithms, and Practice
• Partial Transfer Learning with Selective Adversarial Networks
• Moment Matching for Multi-Source Domain Adaptation
• Unsupervised Intra-domain Adaptation for Semantic Segmentation through Self-Supervision
• Domain Adaptation for Semantic Segmentation with Maximum Squares Loss
• Unsupervised Domain Adaptation of Black-Box Source Models
• Iterative Learning with Open-set Noisy Labels
• Self-Training and Adversarial Background Regularization for Unsupervised Domain Adaptive One-Stage Object Detection
• Partial Adversarial Domain Adaptation
• Unsupervised Domain Adaptation via Structurally Regularized Deep Clustering
• Auxiliary Image Regularization for Deep CNNs with Noisy Labels
• Cross-domain Detection via Graph-induced Prototype Alignment
• Label Propagation with Augmented Anchors: A Simple Semi-Supervised Learning baseline for Unsupervised Domain Adaptation
• Robust Probabilistic Modeling with Bayesian Data Reweighting
• Hypothesis Disparity Regularized Mutual Information Maximization

Cited by in corpus (1)

• Domain Adaptation without Model Transferring