Transferability vs. Discriminability: Batch Spectral Penalization for Adversarial Domain Adaptation

By Xinyang Chen, Sinan Wang, Mingsheng Long, Jianmin Wang. 2019

487 citations recorded by the source corpus.

How researchers see this paper

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

Citations in the corpus

Margin-Based Few-Shot Class-Incremental Learning with Class-Level Overfitting Mitigation (2022, background): “For example, [7] studied this problem for adversarial domain adaptation, [8] studied this problem from the aspect of information-bottleneck theory, and [9] investigated this problem under the label insufficient situations.”
A Comprehensive Survey on Transfer Learning (2019, background): “[165] investigated the transferability and the discriminability in the adversarial domain adaptation and proposed a spectral penalization approach to boost the existing adversarial transfer learning methods.”
Contrastive Test-Time Adaptation (2022, methodology): “For UDA methods we compare to DANN [12], CDAN [32], CDAN+BSP [8], CAN [22], SWD [25], and MCC [21].”
Towards Discriminability and Diversity: Batch Nuclear-Norm Maximization Under Label Insufficient Situations (2020, background): “Recently, BSP [7] penalizes the largest singular values of the batch feature matrix to boost the feature discriminability for domain adaptation, while we analyze the batch classification response matrix to increase both the prediction discriminability and diversity.”
Adversarial Domain Adaptation With Prototype-Based Normalized Output Conditioner (2020, result): “Besides, we also compare with the state-ofthe-art works including SAFN [46], SWD [49], ADR [47], DWC-MEC [45], BSP [48], CAT [66] and iCAN [67].”
Class-Incremental Learning via Dual Augmentation (2021, methodology): “Inspired by [48], we use corresponding angles, denoted by ψ , to explore the distance between two subspaces in incremental learning: Deﬁnition 1 (Corresponding Angle) Given two groups of eigenvectors: { u old, 1 , ..., u old,d } and { u new, 1 , ..., u new,d } , corresponding angle represents the…”
Source Data-Absent Unsupervised Domain Adaptation Through Hypothesis Transfer and Labeling Transfer (2020, result): “For vanilla unsupervisedDA in digit recognition, we compare SHOT with ADDA [5], ADR [56], CDAN [42], CyCADA [8], CAT [98], SWD [99] and STAR [100]; for object recognition, we compare ours with DANN [24], DAN [4], SAFN [82], BSP [101], MDD [102], TransNorm [55], DSBN [25], BNM [103] and GVB-GD [104].”
Do We Really Need to Access the Source Data? Source Hypothesis Transfer for Unsupervised Domain Adaptation (2020, methodology): “For object recognition, we compare with DANN (Ganin & Lempitsky, 2015), DAN (Long et al., 2015), ADR, CDAN, CAT, SAFN (Xu et al., 2019), BSP (Chen et al., 2019), and TransNorm (Wang et al., 2019) and SWD.”
Stochastic Classifiers for Unsupervised Domain Adaptation (2020, unknown): “0 BSP (CDAN) [5] 92.”
Minimum Class Confusion for Versatile Domain Adaptation (2019, methodology): “1 100.0 0.0 95.0 0.1 73.0 0.2 73.6 0.3 89.2 AFN [56] 88.8 0.5 98.4 0.3 99.8 0.1 87.7 0.6 69.8 0.4 69.7 0.4 85.7 AFN + MinEnt [13] 90.3 0.4 98.7 0.2 100.0 0.0 92.1 0.5 73.4 0.3 71.2 0.3 87.6 AFN + BSP [6] 89.7 0.4 98.0 0.2 99.8 0.1 91.0 0.4 71.4 0.3 71.4 0.2 86.9 AFN + MCC 95.4 0.3 98.6 0.2 100.0 0.0”
A Survey on Negative Transfer (2020, background): “To this end, Chen et al. [69] proposed to enhance the feature transferability with guaranteed acceptable discriminability by using batch spectral penalization regularization on the largest few singular values.”
Unsupervised Domain Adaptation via Structurally Regularized Deep Clustering (2020, background): “The recent work of [9] improves adversarial feature adaptation, where the discriminative structures of target data may be deteriorated [69].”
Harmonizing Transferability and Discriminability for Adapting Object Detectors (2020, result): “Some recent studies [6, 53] have also implied similar finding, but how to identify and calibrate the feature transferability still remains unclear.”
Deep Domain-Adversarial Image Generation for Domain Generalisation (2020, methodology): “A more economical solution is to use unsupervised domain adaptation (UDA) methods (Long et al. 2015; Ganin and Lempitsky 2015; Hoffman et al. 2018; Gong et al. 2019; Chen et al. 2019), which only use unlabelled target data.”
Catastrophic Forgetting Meets Negative Transfer: Batch Spectral Shrinkage for Safe Transfer Learning (2019, methodology): “Inspired by [3], we use corresponding angles, denoted by θ .”
Overcoming Negative Transfer: A Survey (2020, background): “Chen et al. [23] showed that in adversarial domain adaptation, features extracted by DANN present worse discriminability, and they enhanced the transferability while guaranteeing acceptable discriminability by introducing batch spectral pe-nalization (BSP) as a regularization term over these…”