Pith. sign in

REVIEW 4 major objections 5 minor 55 references

Scalable Out-of-distribution Robustness in the Presence of Unobserved Confounders

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A single additional variable — a proxy or multiple sources — approximately identifies the unobserved confounder's posterior $P(Z|X)$ under three structural assumptions, and the recovery powers a scalable, OOD-robust mixture-of-experts…

desk verdict Useful method with strong experiments, but the BBSE test-time step is under-derived and the regularizer theory only covers η=1; send to review, ask for a revision. read the letter →

arxiv 2411.19923 v2 pith:J2ERGENA submitted 2024-11-29 cs.LG stat.ML

Parjanya Prashant , Seyedeh Baharan Khatami , Bruno Ribeiro , Babak Salimi This is my paper · ORCID
classification cs.LGstat.ML
keywords out-of-distributiongeneralizationunobservedconfounderlatentshiftidentifiabilityproxyvariablemixtureofexpertsblackboxestimation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to show that out-of-distribution robustness under a shifting unobserved confounder does not need the multiple proxies, concept variables, or test-domain access that prior methods require: one extra variable, either a discrete proxy of the confounder or a collection of training sources with different confounder mixes, is enough to identify the latent posterior $P(Z|X)$ up to a small error. The error is $O((1-\eta)/(2\eta-1))$, controlled by how close each latent class comes to having an input where its posterior reaches one, and the paper argues this error tends to zero for high-dimensional informative features. On this foundation the authors build a two-stage predictor: an encoder-decoder recovers $P^{\mathrm{tr}}(Z|X)$, a mixture-of-experts classifier uses that posterior as its gating function, and at test time the gate is reweighted by a BBSE estimate of the confounder shift $P^{\mathrm{te}}(Z)/P^{\mathrm{tr}}(Z)$ computed from unlabeled test inputs alone. If these identifiability results hold, latent-shift OOD problems become solvable at roughly the computational cost of ordinary supervised learning, which is the concrete payoff the paper is after.

What carries the argument

The load-bearing object is the factorization $P(S|X=x)=\sum_z P(S|Z=z)\,P(Z=z|X=x)$ that the Markov assumption $S \perp X | Z$ makes valid: each observed proxy conditional is a convex combination of the rows of the proxy-confounder matrix $M=P(S|Z)$, weighted by the latent posterior. Because $M$ has full rank, the set of posteriors that could explain the observed data forms an $n_z$-dimensional subspace, so any two candidate posteriors are linked by a linear map $A$ with $P(Z|X=x)=A\,Q(Z|X=x)$; weak overlap forces $A$ close to a permutation matrix, which is what turns matching observed marginals into matching posteriors. A second mechanism is the row-variance regularizer $L_{\mathrm{var}}(M)=\max_z \frac{1}{n_s}\sum_s (M_{zs}-1/n_s)^2$, which selects the true confounder-proxy matrix among all minimizers of the reconstruction loss, and a third is the BBSE estimator of the density ratio $w=P^{\mathrm{te}}(Z)/P^{\mathrm{tr}}(Z)$, computed from unlabeled test features, which reweights the mixture gate while the expert distributions stay fixed.

What would settle it

Build a synthetic proxy-task dataset whose true posterior is engineered to satisfy $\max_x P(Z=i|X=x)=0.55$ for every class, and run the full encoder-decoder plus mixture-of-experts pipeline under the same style of train-to-test confounder shift used in Section 6; Theorem 1 then permits a constant-order identification error, so if the recovered posterior and test accuracy are essentially as strong as at $\eta\approx 1$, the theorem's bound is not the operative mechanism, while a collapse would indicate that the ACS results rely on near-pure inputs. A complementary observation is the trained encoder's per-class maximum posterior on the ACS training sets: a value at or below $1/2$ for any class would falsify Assumption 3 on real data.

Watch

Extended reading notes

Core claim

Under Assumptions 1–3 (the proxy is conditionally independent of $X$ given $Z$; the proxy-confounder matrix $P(S|Z)$ is full rank; and each latent class has some input with posterior $P(Z=i|X=x) \ge \eta > 1/2$), Theorem 1 states that any two distributions matching the observed marginal $P(X,S)$ have latent posteriors that agree up to permutation, with $\sup_x |P(Z=i|X=x) - Q(Z=\pi(i)|X=x)| \le O((1-\eta)/(2\eta-1))$. At $\eta=1$ the bound vanishes and the posterior is fully identifiable (Corollary 1), and Proposition 1 argues that as the number of discriminative features grows the effective $\eta$ approaches one, so high-dimensional data approaches full identifiability. Theorem 2 resolves the residual factorization ambiguity: among all matrices that minimize the reconstruction loss, the one belonging to a confounder satisfying weak overlap at $\eta=1$ has the smallest maximum row variance, which is the justification for the regularizer used in training. The paper further claims that the resulting predictor, encoder-estimated gating plus class-specialized experts with BBSE gate reweighting, outperforms invariant, domain-adaptation, and proxy baselines on synthetic and ACS benchmark tasks and scales linearly in sample size.

Load-bearing premise

The load-bearing premise is weak overlap (Assumption 3): for every latent class there must exist some training input whose posterior for that class is above one half, and the variance regularizer is justified at the extreme where some input reaches posterior exactly one; the paper never measures this quantity on the real datasets, and without it the factorization of the observed distribution into confounder posterior and proxy mechanism is not uniquely recoverable.

Editorial extensions

If this is right

  • A single discrete proxy variable, or one labeled source plus unlabeled sources, is sufficient for OOD robustness under latent confounder shift; concept variables and multiple proxies are not needed.
  • The less the latent classes overlap, the tighter the recovered posterior: near-pure example inputs for every class make the approximation nearly exact, and high-dimensional discriminative features push toward full identifiability.
  • At test time the method needs only unlabeled inputs: BBSE estimates the shift ratio $P^{\mathrm{te}}(Z)/P^{\mathrm{tr}}(Z)$ from features of the test data, and adapting to a new confounder distribution reduces to reweighting the gating function.
  • The procedure scales linearly in training time and keeps memory constant as sample size grows, in contrast to kernel-based proxy methods with cubic time and growing memory.
  • On the reported synthetic and ACS benchmarks, the method achieves higher OOD accuracy than ERM, IRM, GroupDRO, VREx, DANN, DeepCORAL, and ProxyDA.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A deployment-ready diagnostic follows from the paper's own logic: estimate $\max_x P(Z=i|X=x)$ per class with the trained encoder; if any class never exceeds $1/2$, Assumption 3 is violated on that dataset and the theoretical support for the recovered gate lapses even if accuracy looks good.
  • The test-time reweighting inherits label-shift assumptions from BBSE: if the shift changes $P(X|Z)$ rather than only $P(Z)$, the reweighting formula is misspecified. A checkable consequence is that the encoder's posterior marginals on unlabeled test features should match the BBSE-estimated $P^{\mathrm{te}}(Z)$; large discrepancies would indicate a shift outside the assumed invariance.
  • The recovery problem is structurally an anchor-based decomposition of the observed matrix $P(S|X)$, so decomposition algorithms with anchor guarantees could give recovery conditions at or below the $\eta=1/2$ threshold and tighten the theorem's worst-case bound.
  • The paper selects $n_z$ by validation reconstruction loss; for a genuinely continuous confounder that is only binned, the chosen number of classes may overstate the true $n_z$ and the ordering information of the confounder is discarded, so extending the claims to continuous $Z$ would require a separate argument.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper studies out-of-distribution generalization when an unobserved discrete confounder Z affects both X and Y and its marginal shifts between train and test. The main theoretical contribution is an approximate identifiability result (Theorem 1): under Markovian structure, full rank of P(S|Z), and a weak-overlap condition (Assumption 3), any two distributions matching the observed P(X,S) have latent posteriors P(Z|X) that agree up to permutation within O((1-η)/(2η-1)). The method then learns P(Z|X) with an encoder-decoder regularized by maximum row variance (Theorem 2), trains a mixture-of-experts predictor, and at test time reweights the gating using BBSE to estimate the shift in P(Z). Experiments on synthetic and ACS datasets show consistent gains over baselines.

Significance. If the theoretical results hold, the paper offers a valuable relaxation of prior proxy-based methods: only a single proxy or multiple unlabeled sources are needed, no test covariates during training, no concept variables, and no generative models. The approximate identifiability bound is a novel, clearly stated contribution, and the empirical gains are substantial (e.g., 0.709 vs 0.702 on ACS Employment, 0.883 vs 0.856 on ACS Income). However, the practical pipeline has gaps: the BBSE step is underspecified without Z labels, the weak-overlap assumption is unverified on real data, and the regularizer's justification covers only η=1. These gaps currently prevent the paper from fully delivering on its stated theoretical backing for the experiments.

major comments (4)
  1. [Section 5.2] The description of BBSE for estimating w = P_te(Z)/P_tr(Z) is incomplete. BBSE (Lipton et al., 2018) requires a classifier f(X) and a confusion matrix P(f(X)|Z) estimated from labeled examples. Since Z is unobserved, the paper never specifies how this confusion matrix is obtained; the only candidate is the inferred encoder posterior, whose calibration and consistency are not analyzed. Without this step, the test-time reweighting in Eq. (6) is not a well-defined algorithm. Please specify the estimator precisely (including the role of the encoder output and any consistency conditions) or state that the implementation uses an approximate confusion matrix and discuss the resulting bias.
  2. [Section 4.1 and Section 6] Assumption 3 (weak overlap) is load-bearing for Theorem 1, but the paper never verifies η on any dataset. For ACS Employment, P_tr(Z=1)=0.05; it is not evident that any region of the 54-dimensional feature space has P(Z=1|X=x) > 1/2, especially since the proxies (public insurance, independent living) are not part of X. The same concern applies to the other real datasets with extreme minority proportions (0.05). If Assumption 3 fails, the bound in Eq. (1) is inapplicable and the reported empirical advantage cannot be attributed to the paper's identifiability theory. Please report estimates of max_x P(Z=i|X=x) from the trained encoder on real and synthetic data, and discuss the implications for the theory.
  3. [Appendix A.4, proof of Theorem 2] The proof claims '∃x* such that φ*_i(x*)=1 for all i', which is impossible for n_z>1 under a valid probability assignment. The argument only needs, for each i, some x_i with φ*_i(x_i)=1; with that correction the variance comparison goes through. More importantly, Theorem 2 is proved only for η=1, while the algorithm uses the L_var regularizer for general η>1/2. The paper should clarify whether the regularizer is a heuristic for η<1 or provide a justification (e.g., a continuity argument) for why the η=1 result supports the practical choice.
  4. [Appendix A.1, proof of Theorem 1, Step 1] The proof asserts dim(span{P(S|X=x)}) = n_z, but this requires that the vectors P(Z|X=x) span R^{n_z} as x varies. Assumption 3 alone gives only coordinate-wise maxima; the proof should include an argument that these maxima imply the affine hull of {P(Z|X=x)} is the full simplex (e.g., by showing the points cannot lie in a proper affine subspace). This is likely true but is not established in the manuscript.
minor comments (5)
  1. [Appendix A.4] The notation in Eq. (4) uses M_{zs} while the proof writes Var(M_{z,:}); please harmonize the indexing.
  2. [Appendix A.3] Proposition 1 is proved only for binary Z; the extension to n_z > 2 should be stated or proven explicitly.
  3. [Figure 3] The left panel's y-axis is labeled 'Timing (seconds)' and the right panel's 'Memory Usage (MB)', but the two panels share a legend without a clear distinction of which curve corresponds to which metric; please make the axes self-contained.
  4. [Introduction] The quotation 'all models are wrong, but some are useful' is attributed only implicitly through 'Inspired by Box's principle'; please add a citation to Box (1976) or rephrase to avoid an unattributed quote.
  5. [Appendix B] The comparison table lists 'Existence of high-likelihood sample' as a difference from ProxyDA, but the practical implications of this assumption for the experiments are not discussed; a brief comment would help.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the identifiability bound, regularizer theorem, and BBSE reweighting are all derived from stated assumptions and external results, not from the target predictor.

full rationale

The paper's central claim is Theorem 1, an approximate identifiability bound for P(Z|X) given P(X,S) under Assumptions 1-3. The bound is stated in terms of the assumed overlap parameter eta (Assumption 3), but it is not fitted from the quantity being predicted: it is a mathematical statement that any two distributions Q matching the observed marginals must have posteriors within O((1-eta)/(2eta-1)), with the proof in Appendix A.1 using only the Markov property, rank(M)=n_z, and the overlap condition. Theorem 2, which justifies the Lvar regularizer, is also non-circular: it proves that the matrix M* associated with a Z satisfying eta=1 has minimum max-row-variance among all reconstruction-loss minimizers. The proof uses the pure-region assumption to express each row of M* as a convex combination of rows of any other feasible M and then applies a variance inequality; it does not define Lvar in terms of the target posterior. The test-time procedure uses BBSE (Lipton et al., 2018) to estimate P_te(Z) from unlabeled test X through the already-trained encoder; this is a standard method-of-moments estimator and is not trained on test labels or on the final prediction. The paper contains no load-bearing self-citations: all cited identifiability, overlap, and shift-estimation results are from external authors. The identified weakness - that Assumption 3 (weak overlap) is never measured on real data and Theorem 2 is proved only for eta=1 - is a genuine robustness and validation gap, but it is not an instance of circularity because the theoretical derivation does not assume the empirical conclusion it is being used to support.

Assumptions & free parameters 2 free parameters · 6 assumptions · 0 invented entities

The central claim depends on discreteness of Z and S, support inclusion, Assumptions 1-3, and the asymptotic discriminative-feature condition in Proposition 1. The main fitted parameters are the regularizer λ and latent cardinality n_z. No new entities are postulated; Z is the problem's unobserved confounder, not an invention of this paper.

free parameters (2)
  • λ (max-variance regularizer weight) = tuned from {1e-3, 1e-2, 1e-1, 1e0, 1e1} on validation
    Controls the trade-off between reconstruction loss and the variance penalty that selects the M matrix; chosen per dataset, so the final predictor depends on it.
  • n_z (number of latent confounder categories) = chosen by validation reconstruction loss, e.g. 3 for ACS Income
    The paper explicitly does not assume n_z; it increments until validation reconstruction stops improving, making the method's complexity a fitted model-selection output.
assumptions (6)
  • domain assumption S ⊥ X | Z (Markov property of Figure 1a)
    Assumption 1 in Section 4.1; used to factor P(S|X=x)=Σ_z P(S|z)P(z|x).
  • domain assumption rank(P(S|Z)) = n_z, with n_s ≥ n_z
    Assumption 2 in Section 4.1; needed for the left-inverse and subspace-dimension argument in Theorem 1.
  • domain assumption Weak overlap: for each i, sup_x P(Z=i|X=x) ≥ η > 1/2
    Assumption 3 in Section 4.1; drives the error bound O((1-η)/(2η-1)) and is the regularizer's target.
  • domain assumption Z and S are discrete, or S can be discretized
    Definition 1 in Section 3; the matrix M and the dimension argument require a discrete proxy and a finite confounder.
  • domain assumption supp(P_te(Z)) ⊆ supp(P_tr(Z))
    Definition 1, condition 2; needed so the density ratio w is finite and BBSE reweighting is well-defined.
  • domain assumption Each feature is discriminative with KL information at least ε, conditioned on previous features
    Proposition 1 in Section 4.2; used to argue η→1 as dim(X)→∞, but only proved for binary Z.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Scalable Out-of-distribution Robustness in the Presence of Unobserved Confounders." pith.science (2026). https://pith.science/paper/J2ERGENA

@misc{pith2026241119923,
  author       = {Pith},
  title        = {Pith review of: Scalable Out-of-distribution Robustness in the Presence of Unobserved Confounders},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J2ERGENA}},
  note         = {Machine review of arXiv:2411.19923}
}
abstract

We consider the task of out-of-distribution (OOD) generalization, where the distribution shift is due to an unobserved confounder ($Z$) affecting both the covariates ($X$) and the labels ($Y$). This confounding introduces heterogeneity in the predictor, i.e., $P(Y | X) = E_{P(Z | X)}[P(Y | X,Z)]$, making traditional covariate and label shift assumptions unsuitable. OOD generalization differs from traditional domain adaptation in that it does not assume access to the covariate distribution ($X^\text{te}$) of the test samples during training. These conditions create a challenging scenario for OOD robustness: (a) $Z^\text{tr}$ is an unobserved confounder during training, (b) $P^\text{te}(Z) \neq P^\text{tr}(Z)$, (c) $X^\text{te}$ is unavailable during training, and (d) the predictive distribution depends on $P^\text{te}(Z)$. While prior work has developed complex predictors requiring multiple additional variables for identifiability of the latent distribution, we explore a set of identifiability assumptions that yield a surprisingly simple predictor using only a single additional variable. Our approach demonstrates superior empirical performance on several benchmark tasks.

Figures

Figures reproduced from arXiv: 2411.19923 by the authors.

Figure 1
Figure 1. Causal diagrams. Shaded circles denote unobserved variables and solid circles denote observed variables. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. This figure illustrates the methodology of our proposed model, which is structured into two primary [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Training time and memory usage for Our method and ProxyDA Scalability To evaluate the scalability of our ap￾proach, we measured memory consumption and run￾time during training on the Proxy task with synthetic data for different sample sizes [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Plot of L val S over different nz. Baselines We implement GroupDRO (Sagawa et al., 2020), IRM (Arjovsky et al., 2019), V-REx (Krueger et al., 2021), DeepCORAL (Sun and Saenko, 2016), and DANN (Ganin et al., 2016) using the DomainBed code (Gulrajani and Lopez-Paz, 2020)…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 31 canonical work pages

  1. [1]

    Kusner, Stephen R

    Ibrahim Alabdulmohsin, Nicholas Chiou, Alexander D'Amour, Arthur Gretton, Sanmi Koyejo, Matt J. Kusner, Stephen R. Pfohl, Olakunle Salaudeen, Jessica Schrouff, and Kai-Wei Tsai. Adapting to latent subgroup shifts via concepts and proxies. In International Conference on Artificial Intelligence and Statistics, pages 9637--9661. PMLR, 2023

    work page 2023

  2. [2]

    Invariant risk minimization

    Martin Arjovsky et al. Invariant risk minimization. arXiv preprint arXiv:1907.02893, 2019

    arXiv 1907

  3. [3]

    A causal framework for distribution generalization

    Rune Christiansen, Niklas Pfister, Martin Emil Jakobsen, Nicola Gnecco, and Jonas Peters. A causal framework for distribution generalization. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44 0 (10): 0 6614--6630, 2021

    work page 2021

  4. [4]

    Environment inference for invariant learning

    Elliot Creager, Jörn-Henrik Jacobsen, and Richard Zemel. Environment inference for invariant learning. In International Conference on Machine Learning. PMLR, 2021

    work page 2021

  5. [5]

    Retiring adult: New datasets for fair machine learning

    Frances Ding, Moritz Hardt, John Miller, and Ludwig Schmidt. Retiring adult: New datasets for fair machine learning. Advances in neural information processing systems, 34: 0 6478--6490, 2021

    work page 2021

  6. [6]

    Overlap in observational studies with high-dimensional covariates

    Alexander D’Amour, Peng Ding, Avi Feller, Lihua Lei, and Jasjeet Sekhon. Overlap in observational studies with high-dimensional covariates. Journal of Econometrics, 221 0 (2): 0 644--654, 2021

    2021

  7. [7]

    u gelgen, and Bernhard Sch \

    Cian Eastwood, Shashank Singh, Andrei L Nicolicioiu, Marin Vlastelica Pogan c i \'c , Julius von K \"u gelgen, and Bernhard Sch \"o lkopf. Spuriosity didn’t kill the classifier: Using invariant predictions to harness spurious features. Advances in Neural Information Processing Systems, 36, 2024

    work page 2024

  8. [8]

    Domain-adversarial training of neural networks

    Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, Fran c ois Laviolette, Mario March, and Victor Lempitsky. Domain-adversarial training of neural networks. Journal of machine learning research, 17 0 (59): 0 1--35, 2016

    2016

Show all 55 references
  1. [9]

    Robust causal graph representation learning against confounding effects

    Hang Gao, Jiangmeng Li, Wenwen Qiang, Lingyu Si, Bing Xu, Changwen Zheng, and Fuchun Sun. Robust causal graph representation learning against confounding effects. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 7624--7632, 2023

    2023

  2. [10]

    Rlsbench: Domain adaptation under relaxed label shift

    Saurabh Garg, Nick Erickson, James Sharpnack, Alex Smola, Sivaraman Balakrishnan, and Zachary Chase Lipton. Rlsbench: Domain adaptation under relaxed label shift. In International Conference on Machine Learning, pages 10879--10928. PMLR, 2023

    2023

  3. [11]

    In search of lost domain generalization

    Ishaan Gulrajani and David Lopez-Paz. In search of lost domain generalization. arXiv preprint arXiv:2007.01434, 2020

    2007 arXiv

  4. [12]

    Harnessing out-of-distribution examples via augmenting content and style

    Zhuo Huang, Xiaobo Xia, Li Shen, Bo Han, Mingming Gong, Chen Gong, and Tongliang Liu. Harnessing out-of-distribution examples via augmenting content and style. arXiv preprint arXiv:2207.03162, 2022

    2022 arXiv

  5. [13]

    Adaptive mixtures of local experts

    Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. Adaptive mixtures of local experts. Neural computation, 3 0 (1): 0 79--87, 1991

    1991

  6. [14]

    Modeling the data-generating process is necessary for out-of-distribution generalization

    Jivat Neet Kaur, Emre Kiciman, and Amit Sharma. Modeling the data-generating process is necessary for out-of-distribution generalization. arXiv preprint arXiv:2206.07837, 2022

    2022 arXiv

  7. [15]

    Auto-encoding variational bayes

    Diederik P Kingma and Max Welling. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013

    2013 arXiv

  8. [16]

    Last layer re-training is sufficient for robustness to spurious correlations

    Polina Kirichenko, Pavel Izmailov, and Andrew Gordon Wilson. Last layer re-training is sufficient for robustness to spurious correlations. arXiv preprint arXiv:2204.02937, 2022

    2022 arXiv

  9. [17]

    Out-of-distribution generalization via risk extrapolation (rex)

    David Krueger, Ethan Caballero, Joern-Henrik Jacobsen, Amy Zhang, Jonathan Binas, Dinghuai Zhang, Remi Le Priol, and Aaron Courville. Out-of-distribution generalization via risk extrapolation (rex). In International Conference on Machine Learning, pages 5815--5826. PMLR, 2021

    2021

  10. [18]

    Measurement bias and effect restoration in causal inference

    Manabu Kuroki and Judea Pearl. Measurement bias and effect restoration in causal inference. Biometrika, 101 0 (2): 0 423--437, 2014

    2014

  11. [19]

    Zin: When and how to learn invariance without environment partition? Advances in Neural Information Processing Systems, 35: 0 24529--24542, 2022

    Yong Lin, Shengyu Zhu, Lu Tan, and Peng Cui. Zin: When and how to learn invariance without environment partition? Advances in Neural Information Processing Systems, 35: 0 24529--24542, 2022

    2022

  12. [20]

    Lipton, Yu-Xiang Wang, and Alex Smola

    Zachary C. Lipton, Yu-Xiang Wang, and Alex Smola. Detecting and correcting for label shift with black box predictors. In International Conference on Machine Learning, pages 3122--3130. PMLR, 2018

    2018

  13. [21]

    Just train twice: Improving group robustness without training group information

    Evan Z Liu, Behzad Haghgoo, Annie S Chen, Aditi Raghunathan, Pang Wei Koh, Shiori Sagawa, Percy Liang, and Chelsea Finn. Just train twice: Improving group robustness without training group information. In International Conference on Machine Learning, pages 6781--6792. PMLR, 2021 a

    2021

  14. [22]

    On the need for a language describing distribution shifts: Illustrations on tabular datasets

    Jiashuo Liu, Tianyu Wang, Peng Cui, and Hongseok Namkoong. On the need for a language describing distribution shifts: Illustrations on tabular datasets. Advances in Neural Information Processing Systems, 36, 2024 a

    2024

  15. [23]

    Enhancing distributional stability among sub-populations

    Jiashuo Liu, Jiayun Wu, Jie Peng, Xiaoyu Wu, Yang Zheng, Bo Li, and Peng Cui. Enhancing distributional stability among sub-populations. In International Conference on Artificial Intelligence and Statistics, pages 2125--2133. PMLR, 2024 b

    2024

  16. [24]

    Heterogeneous risk minimization

    Jiashuo Liu et al. Heterogeneous risk minimization. In International Conference on Machine Learning. PMLR, 2021 b

    2021

  17. [25]

    Invariant causal representation learning for out-of-distribution generalization

    Chaochao Lu et al. Invariant causal representation learning for out-of-distribution generalization. In International Conference on Learning Representations, 2021

    2021

  18. [26]

    How to scale up kernel methods to be as good as deep neural nets

    Zhiyun Lu, Avner May, Kuan Liu, Alireza Bagheri Garakani, Dong Guo, Aur \'e lien Bellet, Linxi Fan, Michael Collins, Brian Kingsbury, Michael Picheny, et al. How to scale up kernel methods to be as good as deep neural nets. arXiv preprint arXiv:1411.4000, 2014

    2014 arXiv

  19. [27]

    Understanding posterior collapse in generative latent variable models

    James Lucas, George Tucker, Roger Grosse, and Mohammad Norouzi. Understanding posterior collapse in generative latent variable models. In DeepGenStruct Workshop @ ICLR, 2019

    2019

  20. [28]

    Domain generalization using causal matching

    Divyat Mahajan, Shruti Tople, and Amit Sharma. Domain generalization using causal matching. In International conference on machine learning, pages 7313--7324. PMLR, 2021

    2021

  21. [29]

    Tchetgen Tchetgen

    Wang Miao, Zhi Geng, and Eric J. Tchetgen Tchetgen. Identifying causal effects with proxy variables of an unmeasured confounder. Biometrika, 105 0 (4): 0 987--993, 2018

    2018

  22. [30]

    Learning from failure: De-biasing classifier from biased classifier

    Junhyun Nam, Hyuntak Cha, Sungsoo Ahn, Jaeho Lee, and Jinwoo Shin. Learning from failure: De-biasing classifier from biased classifier. Advances in Neural Information Processing Systems, 33: 0 20673--20684, 2020

    2020

  23. [31]

    Spread spurious attribute: Improving worst-group accuracy with spurious attribute estimation

    Junhyun Nam, Jaehyung Kim, Jaeho Lee, and Jinwoo Shin. Spread spurious attribute: Improving worst-group accuracy with spurious attribute estimation. arXiv preprint arXiv:2204.02070, 2022

    2022 arXiv

  24. [32]

    Towards stable test-time adaptation in dynamic wild world

    Shuaicheng Niu, Jiaxiang Wu, Yifan Zhang, Zhiquan Wen, Yaofo Chen, Peilin Zhao, and Mingkui Tan. Towards stable test-time adaptation in dynamic wild world. arXiv preprint arXiv:2302.12400, 2023

    2023 arXiv

  25. [33]

    Rao-blackwellizing the straight-through gumbel-softmax gradient estimator

    Max B Paulus, Chris J Maddison, and Andreas Krause. Rao-blackwellizing the straight-through gumbel-softmax gradient estimator. arXiv preprint arXiv:2010.04838, 2020

    2010 arXiv

  26. [34]

    On measurement bias in causal inference

    Judea Pearl. On measurement bias in causal inference. In Proceedings of the Twenty-Sixth Conference on Uncertainty in Artificial Intelligence, pages 425--432, 2010

    2010

  27. [35]

    Dataset shift in machine learning

    Joaquin Qui \ n onero-Candela, Masashi Sugiyama, Anton Schwaighofer, and Neil D Lawrence. Dataset shift in machine learning. Mit Press, 2022

    2022

  28. [36]

    The risks of invariant risk minimization

    Elan Rosenfeld, Pradeep Ravikumar, and Andrej Risteski. The risks of invariant risk minimization. arXiv preprint arXiv:2010.05761, 2020

    2010 arXiv

  29. [37]

    Sagawa, P

    S. Sagawa, P. W. Koh, T. B. Hashimoto, and P. Liang. Distributionally robust neural networks for group shifts: On the importance of regularization for worst-case generalization. In International Conference on Learning Representations, 2020

    2020

  30. [38]

    On causal and anticausal learning

    Bernhard Sch \"o lkopf, Dominik Janzing, Jonas Peters, Eleni Sgouritsa, Kun Zhang, and Joris Mooij. On causal and anticausal learning. arXiv preprint arXiv:1206.6471, 2012

    2012 arXiv

  31. [39]

    Toward causal representation learning

    Bernhard Sch \"o lkopf, Francesco Locatello, Stefan Bauer, Nan Rosemary Ke, Nal Kalchbrenner, Anirudh Goyal, and Yoshua Bengio. Toward causal representation learning. Proceedings of the IEEE, 109 0 (5): 0 612--634, 2021

    2021

  32. [40]

    Alabdulmohsin, Elisabeth Schnider, Kaitlyn Opsahl-Ong, Anne Brown, Shalmali Roy, Daria Mincu, Chen Chen, et al

    Jessica Schrouff, Nathan Harris, Sanmi Koyejo, Ibrahim M. Alabdulmohsin, Elisabeth Schnider, Kaitlyn Opsahl-Ong, Anne Brown, Shalmali Roy, Daria Mincu, Chen Chen, et al. Diagnosing failures of fairness transfer across distribution shift in real-world medical settings. In Advan...

    2022

  33. [41]

    Improving predictive inference under covariate shift by weighting the log-likelihood function

    Hidetoshi Shimodaira. Improving predictive inference under covariate shift by weighting the log-likelihood function. Journal of Statistical Planning and Inference, 90 0 (2): 0 227--244, 2000

    2000

  34. [42]

    Glymour, and Richard Scheines

    Peter Spirtes, Clark N. Glymour, and Richard Scheines. Causation, Prediction, and Search. MIT Press, Cambridge, MA, 2000

    2000

  35. [43]

    Deep coral: Correlation alignment for deep domain adaptation

    Baochen Sun and Kate Saenko. Deep coral: Correlation alignment for deep domain adaptation. In Computer Vision--ECCV 2016 Workshops: Amsterdam, The Netherlands, October 8-10 and 15-16, 2016, Proceedings, Part III 14, pages 443--450. Springer, 2016

    2016

  36. [44]

    spurious

    Qingyao Sun, Kevin P Murphy, Sayna Ebrahimi, and Alexander D'Amour. Beyond invariance: test-time label-shift adaptation for addressing" spurious" correlations. Advances in Neural Information Processing Systems, 36: 0 23789--23812, 2023

    2023

  37. [45]

    Recovering latent causal factor for generalization to distributional shifts

    Xinwei Sun et al. Recovering latent causal factor for generalization to distributional shifts. In Advances in Neural Information Processing Systems, volume 34, pages 16846--16859, 2021

    2021

  38. [46]

    Test-time training with self-supervision for generalization under distribution shifts

    Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei Efros, and Moritz Hardt. Test-time training with self-supervision for generalization under distribution shifts. In International conference on machine learning, pages 9229--9248. PMLR, 2020

    2020

  39. [47]

    An introduction to proximal causal learning

    Eric J Tchetgen Tchetgen, Andrew Ying, Yifan Cui, Xu Shi, and Wang Miao. An introduction to proximal causal learning. arXiv preprint arXiv:2009.10982, 2020

    2009 arXiv

  40. [48]

    Proxy methods for domain adaptation

    Katherine Tsai, Stephen R Pfohl, Olawale Salaudeen, Nicole Chiou, Matt Kusner, Alexander D’Amour, Sanmi Koyejo, and Arthur Gretton. Proxy methods for domain adaptation. In International Conference on Artificial Intelligence and Statistics, pages 3961--3969. PMLR, 2024

    2024

  41. [49]

    van der Laan and Sherri Rose

    Mark J. van der Laan and Sherri Rose. Targeted Learning. Springer, New York, NY, 2011

    2011

  42. [50]

    Tent: Fully test-time adaptation by entropy minimization

    Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Olshausen, and Trevor Darrell. Tent: Fully test-time adaptation by entropy minimization. arXiv preprint arXiv:2006.10726, 2020

    2006 arXiv

  43. [51]

    Out-of-distribution generalization with causal invariant transformations

    Ruoyu Wang, Mingyang Yi, Zhitang Chen, and Shengyu Zhu. Out-of-distribution generalization with causal invariant transformations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 375--385, 2022 a

    2022

  44. [52]

    Causal balancing for domain generalization

    Xinyi Wang et al. Causal balancing for domain generalization. arXiv preprint arXiv:2206.05263, 2022 b

    2022 arXiv

  45. [53]

    Kernel single proxy control for deterministic confounding

    Liyuan Xu and Arthur Gretton. Kernel single proxy control for deterministic confounding. arXiv preprint arXiv:2308.04585, 2023

    2023 arXiv

  46. [54]

    Multi-source domain adaptation: A causal view

    Kun Zhang, Mingming Gong, and Bernhard Schölkopf. Multi-source domain adaptation: A causal view. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 29, 2015

    2015

  47. [55]

    Adaptive risk minimization: Learning to adapt to domain shift

    Marvin Zhang, Henrik Marklund, Nikita Dhawan, Abhishek Gupta, Sergey Levine, and Chelsea Finn. Adaptive risk minimization: Learning to adapt to domain shift. Advances in Neural Information Processing Systems, 34: 0 23664--23678, 2021

    2021

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.