A DIRT-T Approach to Unsupervised Domain Adaptation

By Rui Shu, Hung Hai Bui, Hirokazu Narui, Stefano Ermon. 2018

665 citations recorded by the source corpus.

Domain adaptation refers to the problem of leveraging labeled data in a source domain to learn an accurate model in a target domain where labels are scarce or unavailable. A recent approach for finding a common representation of the two domains is via domain adversarial training (Ganin & Lempitsky, 2015), which attempts to induce a feature extractor that matches the source and target feature distributions in some feature space. However, domain adversarial training faces two critical limitations: 1) if the feature extraction function has high-capacity, then feature distribution matching is a weak constraint, 2) in non-conservative domain adaptation (where no single classifier can perform well in both the source and target domains), training the model to do well on the source domain hurts performance on the target domain. In this paper, we address these issues through the lens of the cluster assumption, i.e., decision boundaries should not cross high-density data regions. We propose two novel and related models: 1) the Virtual Adversarial Domain Adaptation (VADA) model, which combines domain adversarial training with a penalty term that punishes the violation the cluster assumption; 2) the Decision-boundary Iterative Refinement Training with a Teacher (DIRT-T) model, which takes the VADA model as initialization and employs natural gradient steps to further minimize the cluster assumption violation. Extensive empirical results demonstrate that the combination of these two models significantly improve the state-of-the-art performance on the digit, traffic sign, and Wi-Fi recognition domain adaptation benchmarks.

Subfields: Artificial Intelligence, 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 Radiology, Nuclear Medicine and Imaging

In Representation & Self-Supervised Learning

Citations in the corpus

S4L: Self-Supervised Semi-Supervised Learning (2019, methodology): “One way to resolve this is to encourage the model predictions to be locally-Lipschitz, which VAT does[36].”
Domain Adaptation with Asymmetrically-Relaxed Distribution Alignment (2019, methodology): “Concerning algorithms, several prior works demonstrate empirical success of domain-adversarial approaches, (Tzeng et al., 2014; Ganin et al., 2016; Bousmalis et al., 2016; Tzeng et al.; Hoﬀman et al., 2017; Shu et al., 2018).”
Constructing Unrestricted Adversarial Examples with Generative Models (2018, background): “Therefore, it is important to develop new training methods that can generalize to different input distributions, or new methods that can reliably detect covariate shift [52].”
Sim-To-Real via Sim-To-Sim: Data-Efficient Robotic Grasping via Randomized-To-Canonical Adaptation Networks (2018, methodology): “Prior methods can be split into: (1) feature-level adaptation, where domain-invariant features are learned between source and target domains [17, 15, 57, 7, 14, 36, 5, 55], or (2) pixellevel adaptation, which focuses on re-stylizing images from the source domain to make them look like images from th”
A Survey of Deep Learning for Scientific Discovery (2020, background): “Some popular ways to implement this include gradient reversal [62], minimizing a distance function on the hidden representations [137], and even adversarial training [63, 211].”
Model Adaptation: Unsupervised Domain Adaptation Without Source Data (2020, background): “Additionally, some works further explore using unlabeled target data to improve generalization by co-training [59], pseudo-labeling [48, 66], and entropy regularization [52].”
Domain Adaptive Ensemble Learning (2020, methodology): “, CR has also been applied to applications such as stabilizing GAN training [62]. As an unsupervised objective, it is natural to use CR for dealing with unlabeled data in UDA. Examples include Dirt-t [48] which extends VAT with entropy minimization [19] and self-ensembling [14] which is built on top”
A Fourier-based Framework for Domain Generalization (2021, background): “Some recent work [6, 40, 46] also use consistency regularization in UDA to improve the performance on target domain.”
Angular Visual Hardness (2019, background): “, 2010) presents an important transfer learning problem and deep self-training (Lee, 2013) recently emerged as a powerful framework to this problem (Saito et al., 2017a; Shu et al., 2018; Zou et al., 2018; 2019).”
Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision (2023, methodology): “It is also similar to self-training (Lee et al., 2013) and conditional entropy minimization (Grandvalet & Bengio, 2004), which have led to state-of-the-art results in semi-supervised learning (Xie et al., 2020) and domain adaptation (Shu et al., 2018).”
Adapting Convolutional Neural Networks for Geographical Domain Shift (2019, background): “Many recent works use GAN-based architectures for this kind of domain adaptation [10,11,20,28,32,33].”
Multiple Class Novelty Detection Under Data Distribution Shift (2020, background): “Some of these approaches include feature distribution alignment [52], [11], [48], [44], similarity learning [41], residual transfer [27], [28], and generative adversarial network-based methods [17], [31], [14], [45].”
Improving Domain-Specific Classification by Collaborative Learning with Adaptation Networks (2019, methodology): “On the other hand, it is difﬁcult to train a single classiﬁer having low generalization error in the target domain as well as the source domain on challenging adaptation tasks (Shu et al. 2018).”
An Overview of Deep Semi-Supervised Learning (2020, methodology): “As for unsupervised domain adaptation, semi-supervised methods, such as consistency regularization [140, 93, 45], co-regularization [89] or proxy labeling [132, 131] demonstrated their effectiveness in domain adaptation.”
Semi-Supervised Domain Adaptation via Minimax Entropy (2019, methodology): “Many UDA methods utilize a domain classifier to measure the distance [11, 37, 19, 20, 33].”
Uncertainty-aware multi-view co-training for semi-supervised medical image segmentation and domain adaptation (2020, background): “ptation (UDA) settings, networks are trained from a labeled source domain and an unlabeled target domain. Traditional approaches [20,44,52,58] align domains with statistical constraints. Recent works [14,17,18,24,25,55,61,72,73] utilizes adversarial training and self-training to adapt feature traini”
Sliced Wasserstein Discrepancy for Unsupervised Domain Adaptation (2019, methodology): “We note that the method in [63] obtained 99.”
Zero-Effort Cross-Domain Gesture Recognition with Wi-Fi (2019, background): “Tempts to adapt recognition schemes in various domains fall into two categories: virtually generating features for target domains [39, 40, 50, 53] and developing domain-independent features [9, 20, 37].”