Supplementary Material to A PAC-Bayesian Approach for Domain Adaptation with Specialization to Linear Classiﬁers

By Pascal Germain, Amaury Habrard, François Laviolette, Emilie Morvant. 2013

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

Domain-Adversarial Training of Neural Networks (2015, background): “Over the years, a large part of the literature has focused mainly on linear hypothesis (see for instance Blitzer et al., 2006; Bruzzone and Marconcini, 2010; Germain et al., 2013; Baktashmotlagh et al., 2013; Cortes and Mohri, 2014).”
Optimal Transport for Domain Adaptation (2014, neutral): “Several theoretical works [2], [36], [22] have emphasized the role played by the divergence between the data probability distribution functions of the domains.”
A New PAC-Bayesian Perspective on Domain Adaptation (2015, background): “R Here, � ( x ) can be seen as a smooth surrogate—sometimes called the probit loss ( e.g. , McAllester & Keshet, 2011)—of the zero-one loss function as k w k grows, which can provide very tight bounds (see the empirical analyses of Ambroladze et al., 2006; Germain et al., 2009a).”
Bridging Theory and Algorithm for Domain Adaptation (2019, background): “In addition, Germain et al. (2013) proposed a PAC-Bayesian theory for domain adaptation us-ing the domain disagreement pseudometric.”
Appendix for “Learning Overlapping Representations for the Estimation of Individualized Treatment Effects” (2020, methodology): “From Lemma 5, we have Using the importance weighting trick in [3,7], we can split L P 1 − t ( G ˆ ρ t ) into two parts, The second equality is achieved because the Overlap assumption in the Rubin-Neyman potential outcomes model [13] ensures that SUPP( P 1 − t ) = SUPP( P t ) .”
Adaptation Based on Generalized Discrepancy (2019, methodology): “…was successfully used by Mohri and Muñoz (2012) to provide tight learning guarantees for the related task of learning with drifting distributions, whereas a modified version of the discrepancy was used by Germain et al. (2013) to study the problem of domain adaptation in a PAC-Bayesian setting.”
TTT++: When Does Self-Supervised Test-Time Training Fail or Thrive? (2021, methodology): “We ﬁrst evaluate our method on the inter-twinning moons problem [4, 37], where the main task is to predict the moon class of a given data point and the SSL task is to predict on which side of the hyperplane ( i.e. , linear separator between the two moons) the data point lies on.”
On the Value of Target Data in Transfer Learning (2020, background): “The most common such measures are the so-called dA-divergence [3, 4, 5] and the Y-discrepancy [6, 7, 8].”
Risk bounds for the majority vote: from a PAC-Bayesian analysis to a learning algorithm (2015, background): “…this theory to many more complex paradigms: Transductive Learning (Derbeko et al., 2004; Catoni, 2007; Bégin et al., 2014), Domain Adaptation (Germain et al., 2013), Density Estimation (Seldin and Tishby, 2009; Higgs and Shawe-Taylor, 2010), Structured output Prediction (McAllester, 2007;…”
A Review of Domain Adaptation without Target Labels (2019, methodology): “The PAC-Bayesian framework is a combination of PAClearning and Bayesian inference [18], [190], [191].”
An introduction to domain adaptation and transfer learning (2018, background): “Nonetheless, it is possible to derive generalization error bounds if more is known on the relationship between S and T [4,17,29,46,44,43,141,18,16,212,74,15,75].”
On Localized Discrepancy for Domain Adaptation (2020, background): “06 24 2v 1 [ cs by a series of works Mansour et al. (2009a); Germain et al. (2013); Cortes et al. (2015); Zhang et al. (2019).”