Pith. sign in

REVIEW 4 major objections 6 minor 52 references

Progressive Generalization Risk Reduction for Data-Efficient Causal Effect Estimation

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read Greedily shrinking the causal-effect risk bound at each labeling step is provably optimal over the whole budget, and a simple pairing rule, MACAL, achieves it in practice.

desk verdict The theoretical centerpiece does not survive contact with the proof, but the heuristic and the open-sourced experiments are worth a look. read the letter →

arxiv 2411.11256 v1 pith:4W4WIS7K submitted 2024-11-18 cs.LG stat.ML

classification cs.LGstat.ML
keywords causaleffectestimationactivelearningPEHEgeneralizationriskboundpositivityassumptionbatch-modelabelacquisitiontreatment
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

Estimating individual treatment effects from observational data usually assumes a large labeled dataset, but in high-stakes domains labels are scarce and expensive. This paper studies which unlabeled samples to buy labels for, and claims that the right objective is to shrink the upper bound on the PEHE generalization risk at every acquisition step: the bound's per-step shrinkage decomposes into a model-variance term and a treatment-group distributional-discrepancy term, and Theorem 1 asserts that greedily maximizing each step's shrinkage maximizes the total risk reduction over the whole budget. To realize this, the paper proposes MACAL, a model-agnostic batch acquisition rule that iteratively pairs one treated and one control sample, rewarding distance from the already-labeled data (diversity and uncertainty) while penalizing the pair's mutual distance (preserving positivity). If the claim holds, a cheap $O(N^2)$ acquisition rule makes causal-effect models data-efficient under expensive labeling, and the experiments on IHDP, IBM, and CMNIST show lower $\sqrt{\epsilon_{\mathrm{PEHE}}}$ than general and causal-aware active learning baselines across two deep-kernel backbones.

What carries the argument

The engine of the paper is the Shalit et al. PEHE upper bound $B_i = \sum_{t\in\{0,1\}} \epsilon^t_{i,F} + C_\phi\,\mathrm{IPM}(\rho^{t=1}_\phi, \rho^{t=0}_\phi)$, rewritten as bias plus variance plus noise plus discrepancy, whose between-step difference $\Delta B_i = \sum_t \Delta^t_{\mathrm{Var},i} + C_\phi\,\Delta\mathrm{IPM}_i$ is the object Theorem 1 tells one to maximize at every query. Two lemmas carry the convergence claims: Lemma 2 uses Gaussian-process predictive variance, capped by the signal variance $\sigma^2_f$ and non-growing as the training set expands, to produce the $\Omega(\beta^i)$ variance rate; Lemma 3 uses the 1-Wasserstein distance and the existence of identical paired samples to produce the $O(1/(i+\gamma_0))$ discrepancy rate. The practical criterion is Eq. (7), the MACAL pair score, which combines group-wise nearest-neighbor distances with a pairing-penalty term $\alpha$.

What would settle it

Run a two-step acquisition on a small synthetic pool and compare MACAL's greedy per-step choice against a one-step-lookahead oracle that retrains the model on every candidate batch and picks the batch minimizing the final $\sqrt{\epsilon_{\mathrm{PEHE}}}$ after both steps; since $\Delta B_i$ depends on the model state left by the previous step, the oracle should sometimes beat the union of per-step greedy maximizers, which would falsify the equality in Theorem 1 as a statement about real models. A second check: compute Lemma 2's variance-reduction lower bound on a finite pool where the most uncertain points have predictive variance below the prior cap $\sigma^2_f$, and test whether the $b_0 \sigma^2_f / N_{\mathrm{pool}}$ step still holds.

Watch

Extended reading notes

Core claim

The central claim is that the PEHE risk of a causal effect estimation model can be reduced as fast as possible by treating every acquisition step as a maximization of the shrinkage of the Shalit et al. upper bound, $\Delta B_i = \sum_{t\in\{0,1\}} \Delta^t_{\mathrm{Var},i} + C_\phi\,\Delta\mathrm{IPM}_i$, with bias and noise terms cancelling between consecutive steps; Theorem 1 states that the union of the per-step optimal batches is the global optimum over the entire labeling budget. The convergence behaviors rest on Lemma 2, which gives an $\Omega(\beta^i)$ lower bound for variance reduction under Gaussian-process uncertainty sampling, and Lemma 3, which gives an $O(1/(i+\gamma_0))$ rate for the 1-Wasserstein discrepancy when matched pairs can be labeled. MACAL implements the combined objective through the pair-acquisition criterion $(\tilde{x}^{t=1}, \tilde{x}^{t=0})^* = \arg\max \sum_{t\in\{0,1\}} \min_{x'\in D^t_{\mathrm{train}}} d(\tilde{x}^t, x') - \alpha\, d(\tilde{x}^{t=1}, \tilde{x}^{t=0})$, acquiring $|\tilde{D}_i|/2$ pairs per query in $O(N_{\mathrm{pool}}^2)$ time. The paper reports that MACAL achieves lower empirical $\sqrt{\epsilon_{\mathrm{PEHE}}}$ than Random, BADGE, BAIT, LCMD, QHTE, and the $\mu$BALD, $\rho$BALD, and $\mu\rho$BALD variants on IHDP, IBM, and CMNIST with DUE backbones, and that its post-acquisition training sets show visibly improved treated-control overlap in t-SNE projections.

Load-bearing premise

The proof that maximizing each step's shrinkage is globally optimal assumes the per-step shrinkages are independent, so that choices made at step 1 cannot change what step 2 can achieve, yet every acquired batch retrains the model and changes the data, and the paper itself states in Appendix C that convergence is only established in two extreme regimes, not the realistic middle one where variance and discrepancy are comparably important.

Editorial extensions

If this is right

  • If Theorem 1 holds, any labeling budget is best spent by greedily acquiring the batch that maximizes the per-step bound shrinkage; no batch that looks worse at one step but better later can beat it.
  • In the variance-dominated regime the risk bound converges no slower than $\Omega(\beta^i)$, and in the discrepancy-dominated regime no faster than $O(1/(i+\gamma_0))$; the two rates justify combining uncertainty and pairing in one criterion.
  • MACAL turns the NP-hard combinatorial batch-selection problem into $O(N_{\mathrm{pool}}^2)$ distance computations per query step, so the acquisition rule remains usable with large pools.
  • Empirically, MACAL sets the lower envelope of $\sqrt{\epsilon_{\mathrm{PEHE}}}$ across IHDP, IBM, and CMNIST under two deep-kernel backbones, and its advantage on IHDP lasts until the treated pool is exhausted.
  • Because the criterion is model-agnostic, it can be dropped into any causal effect estimation backbone without re-deriving acquisition-specific uncertainties.

Reading between the lines

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

  • The paper does not prove the greedy step is globally optimal in the realistic case where $\Delta B_i$ depends on the model state; a lookahead or submodularity analysis of the acquisition objective would be needed to close that gap, and the empirical gains could survive even if the theorem's separability fails.
  • The paper's own ablation ($\alpha \in \{0, 1, 2.5, 5, 10\}$) shows the best regularization strength drifts downward as the pool is consumed, suggesting an annealed schedule for $\alpha$, stronger early and weaker late, is a natural extension the paper leaves implicit.
  • The pairing principle transfers beyond binary treatment: continuous treatments, multiple treatment arms, or unequal labeling costs per group would each need a modified pair score, but the variance-plus-discrepancy shrinkage view supplies the template.
  • The $\Omega(\beta^i)$ variance-rate argument is tied to Gaussian-process predictive variance; testing MACAL with deep ensembles or other uncertainty estimators would show whether the rate is a property of the acquisition rule or an artifact of the Gaussian-process assumption.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper studies active learning for causal effect estimation (CEE) under a limited labeling budget, proposing a batch-mode acquisition algorithm named MACAL. The central theoretical claim is Theorem 1, which asserts that the maximum reduction of the Shalit et al. PEHE upper bound over the entire labeling budget is achieved by greedily maximizing the per-query bound shrinkage at each step. MACAL operationalizes this idea by scoring candidate pairs with a distance-based criterion that combines within-group diversity/uncertainty and cross-group similarity, regularized by a hyperparameter α. Experiments on IHDP, IBM, and CMNIST compare MACAL with general and causal active learning baselines, reporting consistent empirical gains.

Significance. If Theorem 1 were valid, the paper would contribute a provably near-optimal, model-agnostic acquisition rule with O(N^2) complexity for an important practical problem. The empirical study is a real strength: the implementation is open-sourced, the comparisons span three datasets and multiple baselines, and the paper includes visualizations and an ablation of the symmetry regularizer. However, the theoretical foundation is unsound: the key step in the proof of Theorem 1 is invalid, and the supporting lemmas rely on idealized assumptions that are neither stated as formal conditions nor satisfied in the paper's own experiments. The contribution therefore reduces to an empirical heuristic, which does not match the paper's advertised theoretical framing.

major comments (4)
  1. [Appendix A.1, Eq. (15)] The proof of Theorem 1 contains an invalid interchange of argmax and union. Equation (14d) is a telescoping identity for a fixed acquisition trajectory: ΔB_overall = Σ_i ΔB_i. This does not imply that arg max_{D_overall} Σ_i ΔB_i equals ∪_i arg max_{D_i} ΔB_i. Each ΔB_i depends on the accumulated dataset D_{i-1} and the model trained on D_i (Equations 13f–13h), so the objective at step i is a function of all previous acquisition choices. No monotonicity, submodularity, or exchange condition is stated to make greedy per-step maximization globally optimal, and the budget constraint |D_overall| ≤ M further couples the per-step feasible sets. This invalid step is the entirety of the theoretical justification for MACAL's design.
  2. [Appendix A.2, Eqs. (21h)–(21i)] The lower bound in Lemma 2 assumes that each of the b0 acquired most-uncertain samples has predictive variance equal to the prior signal variance σ^2_f and that, after acquisition, its posterior variance is zero. In a Gaussian process, the maximum prior variance is approached only by points far from all training data; there may be fewer than b0 such points in a finite pool, and the posterior variance of an observed point is zero only at that exact input. The manuscript does not establish that the necessary number of maximum-variance points exists, so the claimed Ω(β^i) rate is not guaranteed by the argument.
  3. [Appendix A.3, Eq. (27)] Lemma 3's proof assumes that identical cross-treatment samples can be found at every query step and that the optimal Wasserstein matching after adding these identical pairs leaves the matching of the previous points unchanged. The second assertion is not generally true: adding points can re-optimize the transport plan, so the equality in Eq. (27h)–(27i) is unsupported. The first assertion is violated in the paper's own IHDP experiment, where the t=1 pool is exhausted and MACAL switches to one-sided acquisition. Consequently, the O(1/(i+γ0)) convergence rate is not established.
  4. [Section 3.2, Eq. (7)] Even if Theorem 1 were valid, the paper does not show that the implemented criterion (7) maximizes the bound shrinkage ΔB_i or the combinatorial objective (4). The transition from the entropy-weighted objective (4) to the greedy distance-based pair selection is described as an approximation, but no error bound is given, and the hyperparameter α is not connected to the constant C_φ in the bound. The algorithmic claim is therefore a heuristic approximation of an already unproven optimality statement.
minor comments (6)
  1. [Assumption 1] The acronym is rendered as 'SUTV A' instead of 'SUTVA' in the assumption statement; please correct the typo.
  2. [Appendix A.5, Proposition 2 proof] The word 'fucntion' should be 'function'.
  3. [Algorithm 1, line 14] In the else branch for the exhausted t=1 group, the update writes to D^{t=1}_train; it should update D^{t=0}_train.
  4. [Section 3.2 and 5.3] The text refers to 'Appendix 5.3' when discussing the α ablation; this should be 'Section 5.3'.
  5. [Figure 3] The subfigure labels 'IHDP-CausalAL' and 'IHDP-GeneralAL' are not defined in the text; please clarify which baselines are causal-specific and which are general.
  6. [Section 5.1] Claimed performance gains are reported without confidence intervals or significance tests across the multiple outcome simulations; providing such statistics would strengthen the empirical claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the empirical claim is evaluated on held-out PEHE and the risk bound is imported from external work (Shalit et al.); the flagged Eq. (15) greedy step is a proof gap, not a constructional equivalence.

full rationale

The paper's central empirical claim is not fitted to its target: PEHE is measured on held-out test sets (Section 5.1) and the risk upper bound used as the theoretical starting point is Shalit et al.'s external bound [33], not a quantity defined by MACAL. The convergence lemmas are explicitly conditional: Lemma 2 assumes the acquisition of maximum-predictive-variance points and Lemma 3 assumes that identical cross-treatment pairs can always be found, so they are consistency analyses of the intended acquisition behavior rather than independent predictions; Appendix C itself states that the mixed-C_phi regime is not covered ('due to technical difficulties, we do not obtain the risk convergence for the entire risk upper bound'). Theorem 1's Eq. (15) contains an unjustified identification of a global argmax with a union of per-step argmaxes, and each Delta B_i depends on the accumulated dataset and retrained model at step i, so the greedy theorem is not established; however, this is a mathematical validity gap, not circular reasoning, because the paper never defines the global objective in terms of the per-step solutions and no fitted parameter is relabeled as a prediction. Self-citations [42,43] appear only as generic CEE/positivity references and are not load-bearing; the load-bearing citations (Shalit et al. [33], Rasmussen & Williams [44], Williams & Vivarelli [45]) are external. The held-out benchmark comparisons therefore retain independent content.

Assumptions & free parameters 1 free parameters · 6 assumptions · 0 invented entities

The theoretical framework rests on the external Shalit et al. bound plus several idealizations: canceled bias and noise, exact cross-treatment identical pairs, and GP variance reaching its prior maximum. The most serious ledger entry is the separability axiom in Eq. (15), which is an unjustified step rather than a standard background result. The method itself adds one free hyperparameter, alpha.

free parameters (1)
  • symmetrical regularization coefficient alpha = tuned from {0, 1, 2.5, 5, 10}; no single value dominates
    Controls the penalty on treated-control pair distance in Eq. (7). The ablation in Section 5.3 shows the best alpha changes over the acquisition process, and no principled value is derived from the theory.
assumptions (6)
  • domain assumption SUTVA, unconfoundedness, and positivity (Assumptions 1-3)
    Required for identifying individual treatment effects and for the PEHE evaluation to be meaningful. These are standard causal inference assumptions stated in Section 2.1.
  • standard math Shalit et al. PEHE upper bound (Eq. 9)
    External bound from [33] used as the starting point for the risk decomposition in Theorem 1.
  • ad hoc to paper Bias and noise terms cancel between consecutive query steps
    Eq. (13e) and (13g) drop the bias difference and noise difference between steps i-1 and i. The paper justifies this by citing model class dependence or negligible bias, but provides no argument that this holds under active acquisition with deep models.
  • ad hoc to paper Separability of the overall argmax into per-step argmaxes
    Eq. (15) assumes the per-step bound shrinkages are independent, so the union of per-step maximizers equals the global maximizer. This is generally false because each ΔB_i depends on the accumulated dataset and model.
  • ad hoc to paper Identical cross-treatment samples exist at every query step
    Lemma 3's O(1/i) convergence requires that each acquired pair consists of identical treated and control feature vectors. Real pools rarely contain exact matches, and the algorithm only selects the most similar pair.
  • ad hoc to paper Acquired most-uncertain samples attain the GP prior variance and become zero-variance observations
    Lemma 2's lower bound assumes selected pool points have predictive variance exactly sigma_f^2, which is an asymptotic limit for RBF kernels and may not be attained in a finite pool.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Progressive Generalization Risk Reduction for Data-Efficient Causal Effect Estimation." pith.science (2026). https://pith.science/paper/4W4WIS7K

@misc{pith2026241111256,
  author       = {Pith},
  title        = {Pith review of: Progressive Generalization Risk Reduction for Data-Efficient Causal Effect Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4W4WIS7K}},
  note         = {Machine review of arXiv:2411.11256}
}
read the original abstract

Causal effect estimation (CEE) provides a crucial tool for predicting the unobserved counterfactual outcome for an entity. As CEE relaxes the requirement for ``perfect'' counterfactual samples (e.g., patients with identical attributes and only differ in treatments received) that are impractical to obtain and can instead operate on observational data, it is usually used in high-stake domains like medical treatment effect prediction. Nevertheless, in those high-stake domains, gathering a decently sized, fully labelled observational dataset remains challenging due to hurdles associated with costs, ethics, expertise and time needed, etc., of which medical treatment surveys are a typical example. Consequently, if the training dataset is small in scale, low generalization risks can hardly be achieved on any CEE algorithms. Unlike existing CEE methods that assume the constant availability of a dataset with abundant samples, in this paper, we study a more realistic CEE setting where the labelled data samples are scarce at the beginning, while more can be gradually acquired over the course of training -- assuredly under a limited budget considering their expensive nature. Then, the problem naturally comes down to actively selecting the best possible samples to be labelled, e.g., identifying the next subset of patients to conduct the treatment survey. However, acquiring quality data for reducing the CEE risk under limited labelling budgets remains under-explored until now. To fill the gap, we theoretically analyse the generalization risk from an intriguing perspective of progressively shrinking its upper bound, and develop a principled label acquisition pipeline exclusively for CEE tasks. With our analysis, we propose the Model Agnostic Causal Active Learning (MACAL) algorithm for batch-wise label acquisition, which aims to reduce both the CEE model's uncertainty and the post-acquisition ...

Figures

Figures reproduced from arXiv: 2411.11256 by the authors.

Figure 2
Figure 2. At 𝑖-th query step, the entire batch selection is divided into several pairs selection. Left: The most uncertain points are the candidates for selection, and the pair score is penalized by the distance between two points; Mid: The two most uncertain and closest points are selected (not labelled yet) and objectively bringing down other unlabelled points’ uncertainty in proximity; Right: Select the next pair recursive… view at source ↗
Figure 3
Figure 3. Benchmarking with the available SOTAs on three [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Visualization of the post-acquisition dataset (IHDP: First two columns, IBM: Middle two columns, and CMNIST: Last [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: Hyperparameter 𝛼 representing various levels of symmetrical regularization for label acquisition. The relative performance of MACAL_𝛼𝑖 is calculated as 𝜖PEHE,𝛼=0/𝜖PEHE,𝛼=𝛼𝑖 , the higher the better. on post-acquisition imbalance on raw features. Also, even though 𝜇𝜌BALD…
Figure 6
Figure 6. Figure 6: Visualization of the post-acquisition training set at query step 10, 30, and 50 via t-SNE on CMNIST dataset. [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: Visualization of the post-acquisition training set at query step 10, 30, and 50 via t-SNE on IBM dataset. [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: Visualization of the post-acquisition training set at query step 10, 15, and 35 via t-SNE on IHDP dataset. [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 31 canonical work pages

  1. [1]

    Raghavendra Addanki, David Arbour, Tung Mai, Cameron Musco, and Anup Rao. 2022. Sample constrained treatment effect estimation. Advances in Neural Information Processing Systems 35 (2022), 5417–5430

  2. [2]

    Jordan Ash, Surbhi Goel, Akshay Krishnamurthy, and Sham Kakade. 2021. Gone fishing: Neural active learning with fisher embeddings. Advances in Neural Information Processing Systems 34 (2021), 8927–8939

  3. [3]

    Jordan T Ash, Chicheng Zhang, Akshay Krishnamurthy, John Langford, and Alekh Agarwal. 2019. Deep Batch Active Learning by Diverse, Uncertain Gradient Lower Bounds. In International Conference on Learning Representations

  4. [4]

    Vinod Kumar Chauhan, Jiandong Zhou, Ghadeer Ghosheh, Soheila Molaei, and David A Clifton. 2024. Dynamic inter-treatment information sharing for indi- vidualized treatment effects estimation. In International Conference on Artificial Intelligence and Statistics. PMLR, 3529–3537

  5. [5]

    Philippe Clement and Wolfgang Desch. 2008. An elementary proof of the triangle inequality for the Wasserstein metric. Proc. Amer. Math. Soc. 136, 1 (2008), 333– 339

  6. [6]

    David A Cohn, Zoubin Ghahramani, and Michael I Jordan. 1996. Active learning with statistical models. Journal of artificial intelligence research 4 (1996), 129–145

  7. [7]

    Kun Deng, Joelle Pineau, and Susan Murphy. 2011. Active learning for personal- izing treatment. In 2011 IEEE Symposium on Adaptive Dynamic Programming and Reinforcement Learning (ADPRL). IEEE, 32–39

  8. [8]

    Kaito Fujii and Hisashi Kashima. 2016. Budgeted stream-based active learning via adaptive submodular maximization. Advances in Neural Information Processing Systems 29 (2016)

Show all 52 references
  1. [9]

    Yarin Gal, Riashat Islam, and Zoubin Ghahramani. 2017. Deep bayesian active learning with image data. In International conference on machine learning . PMLR, 1183–1192

  2. [10]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition . 770–778

  3. [11]

    Jennifer L Hill. 2011. Bayesian nonparametric modeling for causal inference. Journal of Computational and Graphical Statistics 20, 1 (2011), 217–240

  4. [12]

    David Holzmüller, Viktor Zaverkin, Johannes Kästner, and Ingo Steinwart. 2023. A framework and benchmark for deep batch active learning for regression. Journal of Machine Learning Research 24, 164 (2023), 1–81

  5. [13]

    Nguyen Quoc Viet Hung, Huynh Huu Viet, Nguyen Thanh Tam, Matthias Wei- dlich, Hongzhi Yin, and Xiaofang Zhou. 2017. Computing crowd consensus with partial agreement. IEEE Transactions on Knowledge and Data Engineering 30, 1 (2017), 1–14

  6. [14]

    Guido W Imbens and Donald B Rubin. 2015. Causal inference in statistics, social, and biomedical sciences. Cambridge University Press

  7. [15]

    Andrew Jesson, Sören Mindermann, Yarin Gal, and Uri Shalit. 2021. Quantifying ignorance in individual-level causal-effect estimates under hidden confounding. In International Conference on Machine Learning . PMLR, 4829–4838

  8. [16]

    Andrew Jesson, Sören Mindermann, Uri Shalit, and Yarin Gal. 2020. Identifying causal-effect inference failure with uncertainty-aware models.Advances in Neural Information Processing Systems 33 (2020), 11637–11649

  9. [17]

    Andrew Jesson, Panagiotis Tigas, Joost van Amersfoort, Andreas Kirsch, Uri Shalit, and Yarin Gal. 2021. Causal-bald: Deep bayesian active learning of out- comes to infer treatment-effects from observational data. Advances in Neural Information Processing Systems 34 (2021), 30465–30478

  10. [18]

    Fredrik Johansson, Uri Shalit, and David Sontag. 2016. Learning representations for counterfactual inference. In International conference on machine learning . PMLR, 3020–3029

  11. [19]

    Andreas Kirsch, Sebastian Farquhar, Parmida Atighehchian, Andrew Jesson, Frederic Branchaud-Charron, and Yarin Gal. 2021. Stochastic Batch Acquisition: A Simple Baseline for Deep Active Learning. arXiv preprint arXiv:2106.12059 (2021)

  12. [20]

    Andreas Kirsch, Joost Van Amersfoort, and Yarin Gal. 2019. Batchbald: Efficient and diverse batch acquisition for deep bayesian active learning. Advances in neural information processing systems 32 (2019)

  13. [21]

    Ron Kohavi and Roger Longbotham. 2015. Online controlled experiments and A/B tests. Encyclopedia of machine learning and data mining (2015), 1–11

  14. [22]

    Yann LeCun. 1998. The MNIST database of handwritten digits. http://yann. lecun. com/exdb/mnist/ (1998)

  15. [23]

    Christos Louizos, Uri Shalit, Joris M Mooij, David Sontag, Richard Zemel, and Max Welling. 2017. Causal effect inference with deep latent-variable models. Advances in neural information processing systems 30 (2017)

  16. [24]

    Quoc Viet Hung Nguyen, Chi Thang Duong, Thanh Tam Nguyen, Matthias Wei- dlich, Karl Aberer, Hongzhi Yin, and Xiaofang Zhou. 2017. Argument discovery via crowdsourcing. The VLDB Journal 26 (2017), 511–535

  17. [25]

    Thanh Tam Nguyen, Chi Thang Duong, Matthias Weidlich, Hongzhi Yin, and Quoc Viet Hung Nguyen. 2017. Retaining data from streams of social platforms with minimal regret. In Twenty-sixth International Joint Conference on Artificial Intelligence. KDD ’25, August 3–7, 2025, Toront...

  18. [26]

    C Pilat, T Frech, A Wagner, K Krüger, A Hillebrecht, J Pons-Kühnemann, C Scheibelhut, R-H Bödeker, and F-C Mooren. 2015. Exploring effects of a natural combination medicine on exercise-induced inflammatory immune response: A double-blind RCT. Scandinavian Journal of Medicine &...

  19. [27]

    Robert Pinsler, Jonathan Gordon, Eric Nalisnick, and José Miguel Hernández- Lobato. 2019. Bayesian batch active learning as sparse subset approximation. Advances in neural information processing systems 32 (2019)

  20. [28]

    Tian Qin, Tian-Zuo Wang, and Zhi-Hua Zhou. 2021. Budgeted heterogeneous treatment effect estimation. In International Conference on Machine Learning . PMLR, 8693–8702

  21. [29]

    Pengzhen Ren, Yun Xiao, Xiaojun Chang, Po-Yao Huang, Zhihui Li, Brij B Gupta, Xiaojiang Chen, and Xin Wang. 2021. A survey of deep active learning. ACM computing surveys (CSUR) 54, 9 (2021), 1–40

  22. [30]

    Bent Rosenbaum, Kristian Valbak, Susanne Harder, Per Knudsen, Anne Køster, Matilde Lajer, Anne Lindhardt, Gerda Winther, Lone Petersen, Per Jørgensen, et al. 2005. The Danish National Schizophrenia Project: prospective, comparative longitudinal treatment study of first-episode...

  23. [31]

    Ozan Sener and Silvio Savarese. 2018. Active Learning for Convolutional Neu- ral Networks: A Core-Set Approach. In International Conference on Learning Representations

  24. [32]

    Burr Settles. 2009. Active learning literature survey. Computer Sciences Technical Report (2009)

  25. [33]

    Uri Shalit, Fredrik D Johansson, and David Sontag. 2017. Estimating individual treatment effect: generalization bounds and algorithms. In International Confer- ence on Machine Learning . PMLR, 3076–3085

  26. [34]

    Yishai Shimoni, Chen Yanover, Ehud Karavani, and Yaara Goldschmnidt. 2018. Benchmarking framework for performance-evaluation of causal inference analy- sis. arXiv preprint arXiv:1802.05046 (2018)

  27. [35]

    Kirstine Smith. 1918. On the standard deviations of adjusted and interpolated values of an observed polynomial function and its constants and the guidance they give towards a proper choice of the distribution of observations. Biometrika 12, 1/2 (1918), 1–85

  28. [36]

    Iiris Sundin, Peter Schulam, Eero Siivola, Aki Vehtari, Suchi Saria, and Samuel Kaski. 2019. Active learning for decision-making from imbalanced observational data. In International conference on machine learning . PMLR, 6046–6055

  29. [37]

    Ivor W Tsang, James T Kwok, Pak-Ming Cheung, and Nello Cristianini. 2005. Core vector machines: Fast SVM training on very large data sets. Journal of Machine Learning Research 6, 4 (2005)

  30. [38]

    Joost Van Amersfoort, Lewis Smith, Andrew Jesson, Oscar Key, and Yarin Gal

  31. [39]

    Laurens Van der Maaten and Geoffrey Hinton. 2008. Visualizing data using t-SNE. JMLR (2008)

  32. [40]

    Hao Wang, Jiajun Fan, Zhichao Chen, Haoxuan Li, Weiming Liu, Tianqiao Liu, Quanyu Dai, Yichao Wang, Zhenhua Dong, and Ruiming Tang. 2024. Opti- mal transport for treatment effect estimation. Advances in Neural Information Processing Systems 36 (2024)

  33. [41]

    Liantao Wang, Xuelei Hu, Bo Yuan, and Jianfeng Lu. 2015. Active learning via query synthesis and nearest neighbour search. Neurocomputing 147 (2015), 426–434

  34. [42]

    Hechuan Wen, Tong Chen, Li Kheng Chai, Shazia Sadiq, Junbin Gao, and Hongzhi Yin. 2023. Variational Counterfactual Prediction under Runtime Domain Corrup- tion. IEEE Transactions on Knowledge and Data Engineering (2023)

  35. [43]

    Hechuan Wen, Tong Chen, Li Kheng Chai, Shazia Sadiq, Kai Zheng, and Hongzhi Yin. 2023. To Predict or to Reject: Causal Effect Estimation with Uncertainty on Networked Data. In 2023 IEEE International Conference on Data Mining (ICDM) . IEEE, 1415–1420

  36. [44]

    Christopher KI Williams and Carl Edward Rasmussen. 2006. Gaussian processes for machine learning. MIT press Cambridge, MA

  37. [45]

    Christopher KI Williams and Francesco Vivarelli. 2000. Upper and lower bounds on the learning curve for Gaussian processes.Machine Learning 40 (2000), 77–102

  38. [46]

    Andrew Gordon Wilson, Zhiting Hu, Ruslan Salakhutdinov, and Eric P Xing. 2016. Deep kernel learning. In Artificial intelligence and statistics. PMLR, 370–378

  39. [47]

    Dongrui Wu. 2018. Pool-based sequential active learning for regression. IEEE transactions on neural networks and learning systems 30, 5 (2018), 1348–1359

  40. [48]

    Liuyi Yao, Zhixuan Chu, Sheng Li, Yaliang Li, Jing Gao, and Aidong Zhang. 2021. A survey on causal inference. ACM TKDD 15, 5 (2021), 1–46

  41. [49]

    Liuyi Yao, Sheng Li, Yaliang Li, Mengdi Huai, Jing Gao, and Aidong Zhang. 2018. Representation learning for treatment effect estimation from observational data. Advances in Neural Information Processing Systems 31 (2018)

  42. [50]

    Hongzhi Yin, Qinyong Wang, Kai Zheng, Zhixu Li, and Xiaofang Zhou. 2020. Overcoming data sparsity in group recommendation.IEEE Transactions on Knowl- edge and Data Engineering 34, 7 (2020), 3447–3460

  43. [51]

    Xueying Zhan, Qingzhong Wang, Kuan-hao Huang, Haoyi Xiong, Dejing Dou, and Antoni B Chan. 2022. A comparative survey of deep active learning. arXiv preprint arXiv:2203.13450 (2022). A Theory A.1 Convergence Behaviour of Risk Upper Bound Theorem 1. With budgetM, the maximum ris...

  44. [2021]

    arXiv preprint arXiv:2102.11409 (2021)

    On feature collapse and deep kernel learning for single forward pass uncertainty. arXiv preprint arXiv:2102.11409 (2021)

Pith tools

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