Bregman Divergence-Based Regularization for Transfer Subspace Learning

By Si Si, Dacheng Tao, Bo Geng. 2010

481 citations recorded by the source corpus.

Subfields: Biomedical Engineering, Computer Vision and Pattern Recognition, Signal Processing

How researchers see this paper

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

In Biomedical Engineering

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

Domain Adaptation with Conditional Transferable Components (2016, methodology): “Other methods learn a low dimensional projection by explicitly minimizing the discrepancy between the distributions of projected features on source and target domains (Pan et al., 2011; Long et al., 2014; 2015; Baktashmotlagh et al., 2013; Si et al., 2010; 2011; Muandet et al., 2013).”
A Comprehensive Survey on Transfer Learning (2019, neutral): “Measurement Related Algorithms Maximum Mean Discrepancy [35] [36] [37] [38] [39]· · · Kullback-Leibler Divergence [40] [41] [42] [43] [44]· · · Jensen-Shannon Divergence [45] [46] [47] [48] [49]· · · Bregman Divergence [50] [51] [52] [53] [54]· · · Hilbert-Schmidt Independence Criterion [55] [36] [5”
Manifold Regularized Discriminative Nonnegative Matrix Factorization With Fast Gradient Descent (2011, background): “Other regularizations, e.g., the transfer subspace learning regularization [37], can be incorporated to (12), which we will discuss in future works.”
Transfer Learning with Graph Co-Regularization (2012, methodology): “Distribution Matching: Well-known methods are Maximum Mean Discrepancy Embedding (MMDE) [27], Transfer Component Analysis (TCA) [15], and Transfer Subspace Learning (TSL) [28].”
A Survey on Negative Transfer (2020, background): “HilbertSchmidt independence criterion (HSIC) [44], Bregman divergence [45], optimal transport and Wasserstein distance [46] etc.”
A Review of Domain Adaptation without Target Labels (2019, background): “which the reconstruction to both domains is optimal [145], [146].”
Distribution Calibration in Riemannian Symmetric Space (2011, background): “Basically, when all the samples from the source and target domains are projected into a lowdimensional subspace by some dimension reduction algorithms designed for cross-domain learning [28], their distribution bias can be calibrated.”
Joint cross-domain classification and subspace learning for unsupervised adaptation (2014, methodology): “We present an algorithm that learns jointly both a low dimensional representation and a reliable classifier by optimizing a trade-off between the sourcetarget similarity and the source training error.”
Joint distribution optimal transportation for domain adaptation (2017, methodology): “[6, 7, 8, 9, 5]), affine transform [4] or non-linear transformation as expressed by neural networks [10, 11] ii) The types of divergences used to compare Ps(T (X)) and Pt(T (X)), such as Kullback Leibler [12] or Maximum Mean Discrepancy [9, 5].”
Beyond Sharing Weights for Deep Domain Adaptation (2016, methodology): “method NA PCA SA [19] GFK [18] TCA [43] SSTCA [43] TSL [44] JCSL [42] DDC [7] Ours”
A Practical Transfer Learning Algorithm for Face Verification (2013, background): “The KL divergence, as well as alternative penalties based on Bregman divergences and maximum mean discrepancy, have been motivated for related transfer learning purposes [7, 23, 16, 20, 24, 9], although not in combination with a likelihood function as we have done here.”
Heterogeneous Few-Shot Model Rectification With Semantic Mapping (2020, background): “In this case, sufﬁcient source domain examples should be provided [34], [35], and sometimes even the instance alignment across domains is required [13].”
Support and Invertibility in Domain-Invariant Representations (2019, background): “…metrics such as the maximum mean discrepancy (MMD) (Pan et al. , 2011; Long et al. , 2015, 2016; Baktashmotlagh et al. , 2013) and the Wasserstein distance (Shalit et al. , 2016; Courty et al. , 2017), or other divergences (Berisha et al. , 2016; Si et al. , 2010; Muandet et al. , 2013).”
LTF: A Label Transformation Framework for Correcting Label Shift (2020, background): “More recent works learn domain-invariant representations X (cid:48) = h ( X ) that have similar marginal distributions across domains ( P TX ≈ P SX ) (Si et al., 2009; Pan et al., 2010; Baktashmotlagh et al., 2013; Tzeng et al., 2014; Ganin et al., 2016; Long et al., 2015).”
Overcoming Negative Transfer: A Survey (2020, methodology): “Before MMD, Hilbert-Schmidt independence criterion (HSIC) [57] and Bregman divergence [58] were introduced into transfer learning to measure domain discrepancy.”