Pith. sign in

REVIEW 4 major objections 5 minor 72 references

Deep Adaptive Bayesian Screening (DABS) claims that an offline-trained neural policy can sequentially choose binary factorial experiments that maximize information about which factors are active, with effect sizes and noise integrated out a

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 19:32 UTC pith:47NWDGPR

load-bearing objection A genuinely new adaptive screening method with a clever training objective, but the deployment Gibbs sampler as written is not sampling the stated heredity-constrained posterior — so all interaction posterior probabilities and Case Study 3 results are in question. the 4 major comments →

arxiv 2607.16927 v1 pith:47NWDGPR submitted 2026-07-18 stat.ML cs.LG

Deep Adaptive Bayesian Screening

Jade Lejeune Herman , Arno Strouwen , Johan A.K. Suykens , Peter Goos This is my paper
classification stat.ML cs.LG MSC 62K0562F1562K15
keywords adaptive experimental designfactorial screeningspike-and-slab priorstrong heredityamortized Bayesian optimal experimental designPrior Contrastive EstimationGibbs samplingactive-set recovery
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper introduces DABS, a method for adaptive factorial screening that trains a policy network entirely offline so that, at deployment, each new experiment is selected in a single forward pass without per-step optimization. The key claim is that focusing the design objective on the active-set indicators — with effect sizes and noise variance marginalized out analytically — yields a screening-specific learning signal that outperforms both static designs and greedy adaptive baselines, but only when the experimental budget is smaller than the number of candidate factors. Below this identifiability boundary, adaptivity lets the policy steer subsequent experiments toward promising factors; above it, the paper predicts and observes that a well-chosen static design matches the adaptive policy. At deployment, a Gibbs sampler exploiting the same closed-form Student-t marginals recovers posterior probabilities of factor activity and credible intervals on effect sizes from a single pass through the data. If correct, DABS provides an amortized, ready-to-use screening pipeline for settings where experiments are expensive and budgets are tight.

Core claim

The paper's central claim is that sequential adaptive screening can be amortized by training a policy network offline to maximize a contrastive lower bound on the mutual information between the experimental history and the discrete active-set indicators, rather than on the full parameter vector. Because the spike-and-slab prior and Gaussian likelihood are conjugate, the continuous effect sizes and noise variance are integrated out analytically, leaving a closed-form multivariate Student-t marginal for the history given the active set. This makes the training objective tractable and lets the policy be learned with standard gradient methods on discrete {−1,+1} designs. At deployment, the same

What carries the argument

The load-bearing object is the closed-form multivariate Student-t marginal p(h_T | z) obtained by analytically integrating effect sizes and noise variance out of the conjugate spike-and-slab linear-Gaussian model. This single formula serves three roles: it defines the screening-specific sPCE training objective (contrastive samples are drawn only over the discrete activity indicators z), it makes the policy network's input the exactly sufficient statistics of the history, and it supplies the full conditional distributions of the deployment Gibbs sampler. The policy itself is a two-layer MLP mapping sufficient statistics to per-factor logits, with a straight-through Gumbel-Softmax relaxation t

Load-bearing premise

The whole training objective and the Gibbs decoder rely on the closed-form Student-t marginal p(h_T | z) being exact, but that formula is derived only for a main-effects-only linear model with a common slab variance; the paper applies it without proof to the full model with interactions, heredity, and a second slab, so if that marginal is inexact under the full model, both the learned policy and the posterior probabilities are aimed at the wrong likelihood.

What would settle it

Take a small problem with interactions and heredity (e.g., p=4 with one active interaction), generate random histories from the true generative model, and compare the closed-form Student-t p(h_T | z) against a numerically integrated or long-run MCMC estimate of the same marginal. A systematic mismatch would mean the sPCE training objective and the Gibbs full conditionals evaluate the wrong likelihood. Additionally, test the wedge prediction directly: for T just above p, the DABS-versus-static F1 gap should be statistically indistinguishable from zero; a persistent gap in that regime would cont

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Under the wedge prediction, adaptive screening is most valuable when the experimental budget is below the identifiability boundary; once T is at or above the number of candidate factors, a well-chosen static design achieves the same expected information gain.
  • DABS matches or outperforms the strongest non-adaptive Bayesian baseline across three published screening benchmarks, with large gains at tight budgets and no significant loss above the boundary.
  • The trained policy adds essentially no per-step optimization cost at deployment: each design is a single forward pass, and the Gibbs decoder returns activity probabilities and effect-size credible intervals in one pass through the data.
  • Targeting the screening-specific mutual information I(z; h_T) rather than total parameter information improves factor-recovery F1; the paper's ablation shows a measurable drop when the total-information objective is used.
  • The logit-L2 penalty is load-bearing: removing it collapses the policy into a nearly constant design and reduces main-effect F1 from 0.303 to 0.080.
  • The method assumes strong heredity for interactions; the paper reports that a moderate heredity mismatch at evaluation causes only a small F1 drop, suggesting some robustness to this assumption.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A practical corollary of the wedge prediction is a decision rule for practitioners: if the experimental budget is at or above the number of candidate factors, static designs are likely sufficient, and adaptive screening earns its extra training cost only in the under-identified regime.
  • The sufficient-statistics encoder's input dimension grows as O(p^4), which sets a practical ceiling around p≈26 in the paper's experiments; scaling to much larger factor sets would require a more compact representation of the design history.
  • Because the Student-t marginal rests on conjugacy, extending DABS to non-Gaussian responses or non-conjugate priors would require a likelihood-free contrastive estimator or auxiliary-variable augmentation; the paper identifies this as a limitation, so a natural next step is testing such extensions on binary or count responses.
  • The paper's ablation suggests that targeting I(z; h_T) rather than I(θ; h_T) is decisive; a similar recasting of other amortized design frameworks toward discrete support recovery could let them benefit from the same analytic marginalization whenever conjugacy holds.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes Deep Adaptive Bayesian Screening (DABS), an amortized policy-based method for sequential binary factorial screening under a spike-and-slab prior with strong heredity. The policy is trained offline by maximizing a prior-contrastive lower bound on the mutual information between the history and the active-set indicators, using a claimed closed-form Student-t marginal to integrate out effect sizes and noise. At deployment, a Gibbs sampler is used to recover posterior activity probabilities and credible intervals. The method is evaluated on three simulation benchmarks calibrated to published studies, with a 'wedge prediction' that adaptivity helps only below the identifiability boundary T=p.

Significance. The paper addresses a genuine gap in amortized Bayesian experimental design: discrete factorial screening with posterior uncertainty quantification. It has notable strengths: a clean model specification, pseudocode for training and deployment, a falsifiable wedge prediction, extensive ablations, and a stated intent to release code. If the closed-form marginal and Gibbs decoder were valid for the full model with interactions, the contribution would be useful. However, the analytic marginal is derived for a simpler main-effects-only model, the claimed complexity reduction under heredity is incorrect for pi_int<1, and the deployment Gibbs sampler is not a valid posterior sampler for the stated model. The reported interaction posteriors and credible intervals are therefore not supported. Because all benchmarks are simulated from the paper's own spike-and-slab model, the experiments also provide no external validation of the modeling assumptions.

major comments (4)
  1. [Section 3.2 / Eq. (10) / Appendix C] Appendix C derives p(h_T|z) for a main-effects-only model: y|beta_z,sigma^2 ~ N(X_z beta_z, sigma^2 I), beta_z|sigma^2,z ~ N(0, sigma^2 tau_z^2 I_{p_z}). The full model in Section 3.1 has main effects, two-factor interactions, spike-and-slab with distinct tau_active and tau_inactive, and heredity. Conditional on z, the correct marginal is y|z,sigma^2 ~ N(0, sigma^2(I + X diag(tau_z^2) X^T)) with a vector tau_z^2, not the X_z X_z^T form. Eq. (10) claims the Appendix C form applies to the full model, but no derivation is given. Appendix G's Eq. (37) uses the diag form but calls it a 'tight approximation' of the exact Section 3.1 prior, which is inconsistent because the Section 3.1 prior is exactly the diag form. Since this marginal defines the sPCE training objective and the Gibbs full conditionals, the policy and decoder may be optimizing and evaluating the wrong likelihood.
  2. [Section 3.1, 'Under heredity ... reduces to at most {0,1}^p'] The claim that strong heredity reduces the admissible activity-pattern space to at most {0,1}^p is false for pi_int in (0,1). Under z_ij|z_i,z_j ~ Bernoulli(pi_int z_i z_j), a main-effect pattern with m active factors admits 2^{m(m-1)/2} possible interaction patterns. For p=15 and three active factors there are 2^3 possible interaction patterns, not 1. The space is at most {0,1}^p only in the deterministic limit pi_int=1. This matters because the tractability of contrastive sampling is argued from this reduction: drawing contrastive samples only from {0,1}^p omits interaction indicators, so the p(h_T|z_i) in Eq. (4) is not the marginal of the stated model.
  3. [Algorithm 2 / Appendix G, Eqs. (36)-(37)] Algorithm 2 updates only main-effect indicators z_k (line 12); interaction indicators z_ij are never updated. Eq. (36) defines P(z_k=1|z_-k,h_T) using p(h_T|z_k=1,z_-k) and p(h_T|z_k=0,z_-k) with the same fixed z_-k. Under strong heredity, z_ij are either free Bernoulli variables (pi_int<1) or deterministic functions z_i z_j (pi_int=1). In neither case is Eq. (36) the correct full conditional. With pi_int=1, toggling z_k changes the dependent z_ij, so the two likelihood terms must use different induced patterns. With pi_int<1, z_ij require their own Gibbs update, which is absent. The sampler therefore assigns positive posterior weight to heredity-forbidden configurations and never explores interaction states. Consequently, the interaction inclusion probabilities and credible intervals reported in Section 5.4 and Appendix H.3 are not posterior estimates under the stated model.
  4. [Section 3.1 / Algorithm 2: handling of rho] The model places a Beta(a,b) hyperprior on rho, but the Gibbs sampler in Algorithm 2 and Eq. (36) treats rho as a fixed constant. No Gibbs step for rho is provided. Posterior activity probabilities are therefore computed conditional on an arbitrary rho value rather than marginalizing over the Beta prior. This is a mismatch with the model and affects the reported posterior inclusion probabilities.
minor comments (5)
  1. [Appendix A, Proposition 1 proof] The proof concludes that 'any maximizer' of xi^T Sigma xi over [-1,1]^p lies in {-1,+1}^p, but the coordinate-wise argument only establishes that there exists a maximizer at a vertex. For example, f(xi)=xi_1^2 has non-vertex maximizers. The statement should be weakened to existence of a vertex maximizer.
  2. [Section 3.3 / Algorithm 1] Eq. (11) defines the design via sign(ell_t + g_t), while Algorithm 1 writes xi_t <- GumbelSoftmax(ell_t, tau). These are different operations; the pseudocode should clarify the straight-through relaxation used in the forward/backward pass.
  3. [Abstract] The abstract describes a 'single-pass Gibbs decoder', but Algorithm 2 runs G Gibbs iterations after burn-in. This is not a single pass; consider rephrasing to 'a closed-form Gibbs sampler'.
  4. [Eq. (10)] The text says the scale matrix is 'determined by ... pi_int', but p(h_T|z) conditional on z does not depend on pi_int; pi_int appears in the prior over z, not in the marginal likelihood. This should be clarified.
  5. [Section 5.1 / Code availability] The statement 'Code, model checkpoints and evaluation scripts are available from the author' lacks a URL or repository identifier, making the reproducibility claim unverifiable.

Circularity Check

1 steps flagged

Core sPCE/Gibbs derivations are not circular, but the empirical validation is: all benchmarks are simulated from the same spike-and-slab model whose information structure generates the wedge prediction, so the performance claims are self-confirming.

specific steps
  1. other [Section 5.1 (Experimental Setup), Section 3.6 (wedge prediction), Section 5.2 (confirmation)]
    "For each benchmark, we simulate responses under the model of Section 3 with hyperparameters calibrated to the published problem. ... This convergence matches the wedge prediction of Section 5.1 (formally derived in Section 3.6): once T clears the identifiability dimension of z, static and adaptive designs achieve the same expected information gain and DABS’ advantage collapses."

    The wedge prediction is derived in Section 3.6 from the information decomposition of the spike-and-slab model. The benchmark responses are generated from that exact model (Section 5.1). Therefore the adaptive advantage below T=p and the convergence at T=p are logically entailed by the data-generating process rather than discovered empirically; the experiments are a self-consistency check. The abstract's claim of 'superior accuracy and scalability over classical and Bayesian baselines' is thus validated only under the model the method assumes, although the sPCE and Student-t derivation themselves are not circular.

full rationale

The mathematical core of DABS is not circular: the sPCE lower bound (Eq. 4), the Student-t marginal (Appendix C), and the Gibbs full conditionals (Appendix G) are derived from the stated prior and likelihood. No parameter is fitted and then renamed a prediction, and no load-bearing claim rests on a self-citation; the only self-citations (Strouwen & Micluta-Câmpeanu; Fanuel et al.; Schreurs et al.) are related-work or regularization choices. The central empirical evidence, however, is circularly structured: all case-study responses are simulated from the same spike-and-slab model of Section 3, while the wedge prediction is a direct consequence of that model. Consequently, the observed agreement in Sections 5.2-5.4 is a formal consistency check, not an independent benchmark test. This does not make the training-objective derivation circular, but it means the headline performance claims have not been tested against external data or a misspecified generative process. Separately, Algorithm 2 never updates interaction indicators and Eq. 36 conditions on fixed z_-k while toggling z_k, so under strong heredity the sampler's stationary distribution is not the stated posterior; this is a correctness risk and an omitted derivation step, though not itself a circularity of the training objective.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 0 invented entities

The central method rests on a specific Bayesian model (linear-Gaussian, spike-and-slab, strong heredity, normal-inverse-gamma) and a set of hand-chosen hyperparameters. None of these are fitted to external data; they are either user inputs or simulation pulls. The invented-entities list is empty because DABS does not introduce new physical or mathematical objects—only an algorithm.

free parameters (6)
  • logit-L2 penalty weight mu = 0.01
    Hand-set regularization in Eq. 12; Appendix B.3 shows removing it collapses main-effect F1 from 0.303 to 0.080, so performance depends on this ad hoc coefficient.
  • slab variances tau_active / tau_inactive = 1.0 / 0.01
    Hand-chosen in all case studies (Table 4); the active/inactive variance gap is what separates true effects from noise, but DABS does not estimate these from data.
  • interaction prior probability pi_int = not specified in Table 4
    Defined in Eq. 8; if pi_int < 1, the claimed 2^p support reduction fails, so this free hyperparameter controls whether the contrastive sampling scheme is valid.
  • inverse-gamma hyperparameters alpha0, beta0 = alpha0=3.0, beta0=1.0 or 0.25 per case study
    Set by hand in Table 4; they set the prior noise scale and the Student-t degrees of freedom nu=2*alpha0 used in training and Gibbs updates.
  • Beta sparsity prior parameters (a,b) = Beta(2,8), Beta(4,6), Beta(5,5)
    Chosen per case study; the prior mean determines the expected active count and is used both to generate data and as the DABS prior.
  • Gumbel temperature schedule (tau_start, tau_end, anneal fraction) = 1.0 -> 0.5 over 80%
    Hand-tuned for the straight-through relaxation; temperature annealing controls exploration/exploitation and affects whether policy collapse is avoided.
axioms (5)
  • domain assumption Strong heredity principle: an interaction is active only if both parent main effects are active.
    Stated in Eq. 8; used to prune the interaction space, define p(z_ij|z_i,z_j), and justify sampling contrastive designs only over main-effect indicators. The paper calls it a modeling assumption and abates it in Appendix B.
  • domain assumption Factor sparsity principle: only a small subset of factors is active.
    Encoded via rho~Beta(a,b) with small prior mean in case studies; underpins the spike-and-slab prior and the screening problem definition.
  • domain assumption Linear Gaussian response with additive noise and conjugate normal-inverse-gamma prior.
    Assumed in Eqs. 6-7 and used for the analytical Student-t marginal (Appendix C) and Gibbs full conditionals (Appendix G); non-conjugate extensions are acknowledged as a limitation in Section 5.6.
  • domain assumption Optimal designs on the continuous hypercube are attained at binary vertices for the screening objective.
    Proposition 1 (Appendix A) assumes the mutual information is a monotone function of a quadratic form xi^T Sigma xi; the proof is sketched only for that form and does not cover the full sequential Student-t screening objective.
  • ad hoc to paper The activity-pattern space under strong heredity reduces to at most {0,1}^p.
    Section 3.1 claims this reduction, but with pi_int in (0,1] interaction indicators remain stochastic given main effects; the statement holds only in the deterministic limit pi_int=1, so restricting contrastive samples to p(z) over main effects is not justified by the paper's own model.

pith-pipeline@v1.3.0-alltime-deepseek · 23939 in / 21620 out tokens · 225599 ms · 2026-08-01T19:32:10.925260+00:00 · methodology

0 comments
read the original abstract

We introduce Deep Adaptive Bayesian Screening (DABS), a method for performing adaptive factorial screening in high-dimensional discrete design spaces. DABS learns a policy network offline to sequentially select informative experiments, amortizing Bayesian Optimal Experimental Design. It handles binary designs, incorporates sparsity and interactions via a spike-and-slab prior with strong heredity. The model is trained using a contrastive lower bound on information about factor activity with nuisance effect sizes and noise variance analytically integrated out. Unlike prior amortized Bayesian design approaches, DABS also integrates Gibbs posterior inference at deployment, yielding posterior probabilities of factor activity and credible intervals on effect sizes. We demonstrate DABS on screening problems calibrated to real-world benchmarks and show it achieves superior accuracy and scalability over classical and Bayesian baselines under tight experimental budgets.

Figures

Figures reproduced from arXiv: 2607.16927 by Arno Strouwen, Jade Lejeune Herman, Johan A.K. Suykens, Peter Goos.

Figure 1
Figure 1. Figure 1: (a) DABS policy network πϕ and (b) deployment with Gibbs posterior inference. 3.6 When Does Adaptivity Help The screening model of Sections 3.1–3.2 decomposes θ = (z, β, σ2 ) into discrete activity indicators, continuous effect sizes, and the noise variance. This decomposition determines which component of θ carries the adaptive gain and which does not. By the chain rule of mutual information, the total in… view at source ↗
Figure 2
Figure 2. Figure 2: The wedge prediction: below the boundary T = p, DABS opens an advantage over the optimal static design, with arrows marking each case study’s sweep. The adaptive term I(z; hT ) is bounded above by the identifiability dimension of z. When the bud￾get T is large relative to this dimension, the de￾sign has enough capacity to span the coordinate directions required to resolve z, and no fixed￾versus-adaptive ga… view at source ↗
Figure 3
Figure 3. Figure 3: Main-effect F1 (left) and TPR (right) versus budget [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Main-effect F1 (left) and FDR (right) versus [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Main-effect F1 versus budget T across problem sizes p ∈ {4, 6, 8}. Bands are 95% CIs over 5 seeds [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Two-factor interaction F1 versus budget T across problem sizes p ∈ {4, 6, 8}. Bands are 95% CIs over 5 seeds. tested cover the identifiable regime for main effects, so DABS and the static design achieve nearly the same expected information gain on z and their F1 curves converge. What separates the two amortised policies from the myopic baseline is that both allocate budget with a global view of the sPCE ob… view at source ↗
Figure 7
Figure 7. Figure 7: Main-effect FDR versus budget T at p = 15 for the biotechnology benchmark [PITH_FULL_IMAGE:figures/full_fig_p027_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Main-effect TPR versus number of candidate factors [PITH_FULL_IMAGE:figures/full_fig_p028_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Main-effect TPR versus budget T across p ∈ {4, 6, 8} [PITH_FULL_IMAGE:figures/full_fig_p028_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Two-factor interaction TPR versus budget [PITH_FULL_IMAGE:figures/full_fig_p028_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Main-effect FDR versus budget T across p ∈ {4, 6, 8} [PITH_FULL_IMAGE:figures/full_fig_p029_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Two-factor interaction FDR versus budget [PITH_FULL_IMAGE:figures/full_fig_p029_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Illustration of adaptive behavior in a single rollout [PITH_FULL_IMAGE:figures/full_fig_p033_13.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

72 extracted references

  1. [1]

    Foster, Adam and Ivanova, Desi R and Malik, Ilyas and Rainforth, Tom , booktitle =

  2. [2]

    Jang, Eric and Gu, Shixiang and Poole, Ben , booktitle=

  3. [3]

    , title =

    Williams, Ronald J. , title =. Machine Learning , volume =

  4. [4]

    , year =

    Ishwaran, Hemant and Rao, J. , year =

  5. [5]

    Statistical Science , volume =

    Chaloner, Kathryn and Verdinelli, Isabella , title =. Statistical Science , volume =

  6. [6]

    and Rainforth, Tom , booktitle =

    Ivanova, Desi R and Foster, Adam and Kleinegesse, Steven and Gutmann, Michael U. and Rainforth, Tom , booktitle =

  7. [7]

    and Meyer, R

    Box, George E.P. and Meyer, R. Daniel , title =. Technometrics , volume =

  8. [8]

    and McCulloch, Robert E

    George, Edward I. and McCulloch, Robert E. , title =. Journal of the American Statistical Association , volume =

  9. [9]

    Journal of Machine Learning Research , volume =

    Mohamed, Shakir and Rosca, Mihaela and Figurnov, Michael and Mnih, Andriy , title =. Journal of Machine Learning Research , volume =

  10. [10]

    and Mnih, Andriy and Yee, Whye Teh , title =

    Maddison, Chris J. and Mnih, Andriy and Yee, Whye Teh , title =. International Conference on Learning Representations , year =

  11. [11]

    , title =

    Zaheer, Manzil and Kottur, Satwik and Ravanbhakhsh, Siamak and Poczos, Barnabas and Salakhutdinov, Ruslan and Smola, Alexander J. , title =. Advances in Neural Information Processing Systems , year =

  12. [12]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =

    Geman, Stuart and Geman, Donald , title =. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =

  13. [13]

    Casella, George and Robert, Christian , title =

  14. [14]

    and Wu, C

    Hamada, M. and Wu, C. F. Jeff , title =. Journal of Quality Technology , volume =

  15. [15]

    , title =

    Montgomery, Douglas C. , title =

  16. [16]

    and Hunter, J

    Box, George E.P. and Hunter, J. Stuart and Hunter, William G. , title =

  17. [17]

    and Drovandi, Christopher C

    Ryan, Elizabeth G. and Drovandi, Christopher C. and McGree, James M. and Pettitt, Anthony N. , title =. International Statistical Review , volume =

  18. [18]

    , title =

    Lindley, D.V. , title =. The Annals of Mathematical Statistics , volume =

  19. [19]

    2018 , pages =

    Rainforth, Tom and Cornish, Robert and Yang, Hongseok and Warrington, Andrew and Wood, Frank , booktitle =. 2018 , pages =

    2018

  20. [20]

    and Teh, Yee Whye and Rainforth, Tom , booktitle =

    Foster, Adam and Jankowiak, Martin and O'Meara, Matthew J. and Teh, Yee Whye and Rainforth, Tom , booktitle =. A. 2019 , pages =

    2019

  21. [21]

    International Conference on Machine Learning , year =

    On Variational Bounds of Mutual Information , author =. International Conference on Machine Learning , year =

  22. [22]

    , journal =

    Mitchell, Toby J. , journal =. An Algorithm for the Construction of

  23. [23]

    Jeff and Hamada, Michael S

    Wu, C.F. Jeff and Hamada, Michael S. , title =

  24. [24]

    The Canadian Journal of Statistics , volume =

    Bayesian Variable Selection with Related Predictors , author =. The Canadian Journal of Statistics , volume =

  25. [25]

    and Barto, Andrew G

    Sutton, Richard S. and Barto, Andrew G. , title =

  26. [26]

    and Lawson, Dietrich and Sohl-Dickstein, Jascha , booktitle =

    Tucker, George and Mnih, Andriy and Maddison, Chris J. and Lawson, Dietrich and Sohl-Dickstein, Jascha , booktitle =. 2017 , pages =

    2017

  27. [27]

    Advances in Neural Information Processing Systems , year =

    Attention Is All You Need , author =. Advances in Neural Information Processing Systems , year =

  28. [28]

    International Conference on Machine Learning , year =

    Set Transformer: A Framework for Attention-based Permutation-Invariant Neural Networks , author =. International Conference on Machine Learning , year =

  29. [29]

    and McCulloch, Robert E

    George, Edward I. and McCulloch, Robert E. , journal =. Approaches for

  30. [30]

    and McCulloch, Robert E

    Chipman, Hugh and George, Edward I. and McCulloch, Robert E. , journal =. The Practical Implementation of. 1997 , publisher =

    1997

  31. [31]

    and Pham, Tung-Dinh and Vuong, Mai Phuong

    Nguyen, Nam-Ky and Kenett, Ron S. and Pham, Tung-Dinh and Vuong, Mai Phuong. D-Efficient Mixed-Level Foldover Designs for Screening Experiments. Springer Handbook of Engineering Statistics. 2023

    2023

  32. [32]

    Smucker , title =

    Jacob Abugre Akubire and Byran J. Smucker , title =. Journal of Statistical Computation and Simulation , volume =. 2025 , publisher =

    2025

  33. [33]

    Casting a Spotlight on Factorial Design: Exploring the Power of

    Prathap Madeswara Guptha and Vijayaraj Surendran and Raghavendra Kumar Gunda , journal =. Casting a Spotlight on Factorial Design: Exploring the Power of. 2025 , publisher =

    2025

  34. [34]

    Biometrika , volume =

    A comparison of several variance component estimators , author=. Biometrika , volume =

  35. [35]

    Journal of Quality Technology , volume =

    A class of Three-Level Designs for Definitive Screening in the Presence of Second-order Effects , author=. Journal of Quality Technology , volume =

  36. [36]

    Journal of the American Statistical Association , volume =

    Bayesian Variable Selection in Linear Regression , author=. Journal of the American Statistical Association , volume =

  37. [37]

    A review of Bayesian variable selection methods: what, how and which , author=

  38. [38]

    Bayesian Analysis Advance Publication , pages =

    Bayesian iterative screening in ultra-high dimensional linear regressions , author=. Bayesian Analysis Advance Publication , pages =

  39. [39]

    Double-robust

    Linhan, Ouyang and Ling, Yan and Minghe, Sun and Min, Wang , journal =. Double-robust

  40. [40]

    Becerra, Mario and Goos, Peter , journal =

  41. [41]

    and Guan, Cong and Rainforth, Tom , booktitle =

    Hedman, Marcel and Ivanova, Desi R. and Guan, Cong and Rainforth, Tom , booktitle =. Step-. 2025 , volume =

    2025

  42. [42]

    2019 , volume =

    Poole, Ben and van den Oord, Aäron and Alemi, Alexander A., Tucker George , booktitle =. 2019 , volume =

    2019

  43. [43]

    Advances in Neural Information Processing Systems , year =

    Mutual information estimation via normalizing flows , author =. Advances in Neural Information Processing Systems , year =

  44. [44]

    Advances in Neural Information Processing Systems , year =

    Mutual Information Estimation via f-Divergence and Data Derangements , author =. Advances in Neural Information Processing Systems , year =

  45. [45]

    Gaussian process based expected information gain computation for Bayesian optimal design

    Xu, Zhihang and Liao, Qifeng. Gaussian process based expected information gain computation for Bayesian optimal design. Entropy (Basel)

  46. [46]

    Efficient estimation of expected information gain in Bayesian experimental design with multi-index Monte Carlo

    Du, Xinting and Wang, Hejin. Efficient estimation of expected information gain in Bayesian experimental design with multi-index Monte Carlo. Statistics and Computing

  47. [47]

    Pre-trained Gaussian processes for Bayesian optimization

    Wang, Zi and Dahl, George E and Swersky, Kevin and Lee, Chansoo and Nado, Zachary and Gilmer, Justin and Snoek, Jasper and Ghahramani, Zoubin. Pre-trained Gaussian processes for Bayesian optimization. Journal of Machine Learning Research

  48. [48]

    and Chai, Tianyou , year =

    Guo, Dan and Wang, Xilu and Gao, Kailai and Jin, Yaochu and Ding, J.L. and Chai, Tianyou , year =

  49. [49]

    Science and Technology of Advanced Materials: Methods , volume =

    Tomoki Yamashita and Hiori Kino and Koji Tsuda and Takashi Miyake and Tamio Oguchi , title =. Science and Technology of Advanced Materials: Methods , volume =

  50. [50]

    Zaballa, Vincent D and Hui, Elliot E , title =

  51. [51]

    2024 , eprint =

    Huang, Daolang and Guo, Yujia and Acerbi, Luigi and Kaski, Samuel , title =. 2024 , eprint =

    2024

  52. [52]

    2026 , eprint =

    Strouwen, Arno and Micluta-C. 2026 , eprint =

    2026

  53. [53]

    , title =

    Fanuel, Michaël and Schreurs, Joachim and Suykens, Johan A.K. , title =. SIAM Journal on Mathematics of Data Science , volume =

  54. [54]

    Machine Learning, Optimization, and Data Science , year=

    Schreurs, Joachim and De Meulemeester, Hannes and Fanuel, Micha. Machine Learning, Optimization, and Data Science , year=

  55. [55]

    and Wu, C

    Hamada, M. and Wu, C. F. J. , title =. Journal of Quality Technology , volume =. 1992 , publisher =

    1992

  56. [56]

    Bioresource Technology , volume =

    A study of inulinase production in. Bioresource Technology , volume =. 1995 , author =

    1995

  57. [57]

    2004 , author =

    A systematic strategy to optimize ex vivo expansion medium for human hematopoietic stem cells derived from umbilical cord blood mononuclear cells , journal =. 2004 , author =

    2004

  58. [58]

    2023 , address =

    Novel Aspects of Gas Chromatography and Chemometrics , publisher =. 2023 , address =

    2023

  59. [59]

    and Meyer, R

    Box, G.E.P. and Meyer, R. Daniel , title =. Journal of Quality Technology , volume =. 1993 , publisher =

    1993

  60. [60]

    Materials , VOLUME =

    Liu, Guiqun and Fang, Xinyu and Zhang, Xiaoli and Lv, Guanglei , TITLE =. Materials , VOLUME =. 2025 , NUMBER =

    2025

  61. [61]

    2023 , volume =

    Bickford Smith, Freddie and Kirsch, Andreas and Farquhar, Sebastian and Gal, Yarin and Foster, Adam and Rainforth, Tom , booktitle =. 2023 , volume =

    2023

  62. [62]

    and Jennings, Joel and Rainforth, Tom and Zhang, Cheng and Foster, Adam , booktitle =

    Ivanova, Desi R. and Jennings, Joel and Rainforth, Tom and Zhang, Cheng and Foster, Adam , booktitle =. 2023 , volume =

    2023

  63. [63]

    Journal of Global Optimization , volume =

    Efficient global optimization of expensive black-box functions , author =. Journal of Global Optimization , volume =

  64. [64]

    Variational Bayesian Optimal Experimental Design

    Foster, Adam and Jankowiak, Martin and Bingham, Eli and Horsfall, Paul and Teh, Yee Whye and Rainforth, Tom and Goodman, Noah. Variational Bayesian Optimal Experimental Design. Advances in Neural Information Processing Systems , volume =

  65. [65]

    Pukelsheim, Friedrich , title =

  66. [66]

    General equivalence theory for optimum designs (approximate theory)

    Kiefer, J. General equivalence theory for optimum designs (approximate theory). Annals of Statistics

  67. [67]

    and Donev, A.N

    Atkinson, A.C. and Donev, A.N. and Tobias, R.D. , title =

  68. [68]

    Goos, Peter and Jones, Bradley , title =

  69. [69]

    European Journal of Operational Research

    Efficient space-filling and non-collapsing sequential design strategies for simulation-based modeling , author =. European Journal of Operational Research

  70. [70]

    Experimental Design on a Budget for Sparse Linear Models and Applications

    Ravi, Sathya Narayanan and Ithapu, Vamsi and Johnson, Sterling and Singh, Vikas , booktitle =. Experimental Design on a Budget for Sparse Linear Models and Applications. 2016 , volume =

    2016

  71. [71]

    IEEE Transactions on Information Theory

    Distilled Sensing: Adaptive Sampling for Sparse Detection and Estimation , author =. IEEE Transactions on Information Theory

  72. [72]

    Biometrika

    The Design of Optimum Multifactorial Experiments , author =. Biometrika