Pith. sign in

REVIEW 2 major objections 4 minor 34 references

Stochastic Sequential Search in Very-High-Dimensional Feature Selection

T0 review · 2 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A fixed-size stochastic sample can replace the exhaustive sweep of sequential feature selection, keeping at least 97% of full floating-search quality at a quarter of its evaluations and extending set-level selection past 10,000 features.

desk verdict A well-executed, honest paper with a real structural advance; the abstract overstates retention and the online-statistic's contribution is smaller than advertised, but it deserves peer review. read the letter →

arxiv 2608.01502 v1 pith:EN4SRSHE submitted 2026-08-02 cs.LG cs.CVstat.ML

Petr Somol , Jiří Grim This is my paper · ORCID
classification cs.LGcs.CVstat.ML
keywords featureselectionstochasticsequentialsearchfloatingsSFFSsoftmaxsamplingonlinedependency-awarestatisticsveryhighdimensionalitywrappercriteria
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

Classical sequential feature selection — forward selection with floating backtracking — remains the quality reference for choosing feature subsets, but every step evaluates every remaining candidate feature, which prices the whole family out of problems with thousands to tens of thousands of features. This paper claims that the per-step full sweep is not what carries the search's quality: it can be replaced by a fixed-size stochastic sample whose candidates are drawn by a temperature-controlled softmax over dependency-aware per-feature statistics learned online, at no extra evaluation cost, from every criterion value the search itself produces, with a uniform exploration floor keeping every feature reachable. Substituting this operator pair into floating search yields sSFFS, which on the 500-dimensional madelon benchmark retains at least 97% of the full method's criterion value at every subset size at about a quarter of its evaluations — while uniform sampling at the same budget collapses on madelon's synergistic features. At 5,000 and 10,105 dimensions, far beyond the full method's reach, sSFFS beats dependency-aware and individual-feature ranking at matched budgets on the search objective and, on reuters, on holdout accuracy at every subset size, in about two minutes of single-core evaluation work. If correct, this makes set-level, interaction-aware feature selection routine in a regime previously reserved for interaction-blind rankers.

What carries the argument

The engine is the online dependency-aware contrast $s(f)=\mu_f-\bar\mu_f$: the difference between exponentially forgetting running means of batch-standardized criterion values of evaluated subsets containing $f$ and those lacking it, z-standardized within each step so the contrast aggregates across subset sizes. The operator sADD/sRMV draws $\lfloor\rho_u y\rfloor$ candidates uniformly from the least-evaluated features and the rest by softmax over $s(f)$ with temperature $\tau=\mathrm{IQR}/1.349$, evaluates exactly $y$, commits the best, then folds the batch into the statistics. Substituting sADD/sRMV for ADD/RMV defines the Stochastic Sequential Search family; the fixed budget $y$ makes per

What would settle it

Run sSFFS on a synthetic dataset at $D \approx 10^4$ with a known optimal subset and count how often the anytime frontier recovers the true informative features as the per-step budget $y$ shrinks toward 10. The mechanism predicts graceful degradation; if recovery drops to near chance while criterion values stay high — different spurious subsets on every seed — the contrast statistic is not informing the sampler at extreme $D$. A cheaper probe: on gisette at matched budget, correlate mid-search $s(f)$ with marginal contributions measured by a DAF probe campaign; near-zero correlation would mean

Watch

Extended reading notes

Core claim

Sequential search quality, the paper claims, lives in the search's own evaluation history, not in the exhaustive sweep. The budgeted operator pair sADD/sRMV evaluates exactly $y$ candidates per step — some uniform from the least-evaluated features, the rest by softmax over the contrast $s(f)=\mu_f-\bar\mu_f$, the batch-standardized mean criterion value with vs. without $f$, updated online. Substituting these operators makes per-step cost independent of the feature count $D$, yielding stochastic floating search sSFFS. Empirically sSFFS retains at least 97% of full-SFFS criterion value at every subset size on madelon at a quarter of the evaluations, and dominates BIF and DAF at matched budgets

Load-bearing premise

The method stands or falls on whether the online contrast $s(f)=\mu_f-\bar\mu_f$, accumulated from the search's own evaluations, reliably estimates a feature's usefulness in combination; the paper itself says the retention results are empirical and the uniform-case guarantee does not transfer. If the statistic becomes noise at very high dimensionality, the sampler could concentrate on spuriously good features and collapse the way uniform sampling does on madelon.

Editorial extensions

If this is right

  • Set-level, interaction-aware feature selection becomes routinely feasible at $D \ge 10^4$: a full sSFFS frontier run on reuters (10,105 features) costs roughly two minutes of single-core evaluation work, where previously only interaction-blind ranking could operate.
  • The anytime and d-optimizing properties of floating search are preserved: one run reports the best subset found at every cardinality, tracing the criterion-vs-size curve and exposing the optimal dimensionality.
  • Per-step spend becomes predictable and independent of $D$: the non-floating form sSFS reaches size $d$ in exactly $y_f \cdot d$ evaluations, a hard a priori bound previously available only to ranking methods and probe-based ranking.
  • The operator substitution is generic: sSFS, sSBS, sSBFS, and sOS all follow from the same replacement rule, though only sSFFS is empirically studied here.
  • The gisette experiments locate the binding constraint beyond the sequential frontier in the criterion, not the search: at 500 training samples with 5,000 candidate features, more powerful optimization transfers the wrapper criterion's selection bias, making criterion design the practitioner's limiting decision.

Reading between the lines

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

  • The method's success is conditional on the contrast statistic $s(f)$ staying informative when each feature is observed rarely; the paper's ablations support this on madelon but offer no formal guarantee, so an open empirical question is whether the retention rate holds on datasets where the useful subset is large relative to $D$.
  • The near-total selection instability on gisette (a stability index near 0.036 with near-identical criterion values across runs) implies that standard stability indices cannot distinguish a stochastic optimizer sampling among many equally good subsets from a broken search — a caution that should transfer to any stochastic selector deployed in redundant high-dimensional spaces.
  • The occasional 'greedy-trap escapes' suggest a testable hybridization: using the budgeted search as a randomized perturbation and the full sweep as a local polish could improve even the exact sequential method at moderate dimensionality.
  • Because the statistics are learned from the criterion itself, the sampler inherits the criterion's biases; the gisette results imply the method should be paired with trustworthy criteria (as the reuters multinomial filter is) rather than raw cross-validation accuracy at small sample sizes.
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

2 major / 4 minor

Summary. The paper introduces Stochastic Sequential Search (SSS), a family of budgeted sequential feature-selection methods obtained by replacing the exhaustive per-step sweep of classical ADD/RMV operators with a fixed-size sampled step. The sampled candidates are drawn from a softmax proposal distribution driven by an online dependency-aware per-feature contrast statistic, with a uniform exploration floor; the resulting sSFFS variant is studied empirically. On madelon the method is compared with full floating search as retention evidence; on gisette and reuters it is compared with ranking baselines (BIF, DAF) at matched evaluation budgets. The central claims are that per-step cost becomes independent of dimensionality, that sSFFS retains at least 97% of full-SFFS criterion value at about a quarter of the evaluations on madelon, and that on reuters it dominates BIF and DAF on both the search objective and holdout accuracy at every subset size. The paper also reports an honest analysis of the gisette criterion-generalization gap and provides a verified standalone implementation.

Significance. If the empirical claims hold, the paper makes a meaningful contribution: it extends set-level, interaction-aware floating search to regimes (D > 10^4) previously accessible only to ranking methods, while preserving anytime and d-optimizing behavior. The strengths of the work are its careful matched-budget comparisons, the explicit ablations separating uniform, top-k, frozen, and online-informed sampling, the honest reporting of the gisette holdout inversion, and the reproducible seed-pinned implementation with a standalone verification harness. The paper does not provide formal guarantees, and its headline retention number is sensitive to how the seed/split axis is read, but the central empirical edifice is credible and the presentation is largely transparent.

major comments (2)
  1. [Abstract; Section 5, madelon retention paragraph] The abstract states that sSFFS 'retains at least 97% of the full-SFFS criterion value at every subset size up to 165 at about a quarter of its evaluations.' In the body this is strictly true only on the shared-split seed (with the default y=100/y_b=50 configuration, 97.1% at every size). The ten-seed campaign is weaker: 'between 145 and 165 of the 165 sizes lie above the 97% mark per seed, and no single size falls below 93.5% anywhere in the campaign.' The unqualified abstract wording is thus an overstatement if read as a claim over seeds/splits. Please qualify the claim by stating that the 97%-at-every-size statement refers to the shared-split seed and give the across-seed range in the abstract or conclusion.
  2. [Section 3.1; Section 6; Table 3] The central mechanism is the online contrast s(f) of Eq. (3), and the paper's strongest regime is D=10,105 (Table 3). The ablation evidence for the online update being load-bearing is confined to madelon: Section 6 reports that online updating adds only +0.015 over the frozen warm-start ranking at y=100 (paired t=3.1). No frozen-statistics or warm-start-only ablation is reported for gisette or reuters. Since the proposal distribution is a function of s(f) and the evaluations fed back to s(f) come from proposals biased by s(f), it is important to show that the statistic remains informative when per-feature observation counts are low and the floor is small in absolute terms. I ask for at least one high-dimensional ablation (e.g., frozen vs. online statistics at the reuters setting) or diagnostics such as per-feature proposal counts and score stability; alternatively, the paper should expli
minor comments (4)
  1. [Section 3.2, stage 1] The exploration floor is defined as drawing u candidates 'uniformly without replacement from the 4u least-evaluated features of F.' The factor 4 is unexplained and appears inconsistent with the subsequent statement that the floor 'cycles through the least-observed features.' If the intended pool is the u least-evaluated features, please correct; otherwise justify the 4u factor.
  2. [Section 3.1, Eq. (3)] The notation z(f∈X) in the update equation is not defined. Please specify explicitly that it means z when f is in the evaluated subset and 0 when f is absent (or whichever convention is intended).
  3. [Section 5, madelon retention paragraph] The phrase 'per-seed mean retention spans 99.0–101.2%' is ambiguous: it should state whether this is the mean over subset sizes within each seed, the mean over seeds at each size, or another aggregation. The surrounding text gradually clarifies, but a precise definition would help.
  4. [Conclusion] The conclusion repeats the unqualified 'at least 97% of full floating-search quality at a quarter of its evaluations.' This should be aligned with the qualified statement recommended above so that the abstract, body, and conclusion are consistent.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: sSFFS is measured against external baselines and its central claims are empirical.

full rationale

The paper's derivation chain is self-contained and does not reduce any prediction to an input parameter or to a self-citation. The online statistic s(f) in Eq. (2) and its batch-standardized update in Eq. (3) are explicitly defined from the search's own criterion evaluations; they do not presuppose the claimed retention or dominance results. The headline retention figure (>=97% of full-SFFS value on madelon) is a measured comparison against an independently defined full-search baseline on a shared data split, not a fitted constant or a construction. The gisette and reuters dominance claims are likewise empirical comparisons against BIF and DAF at matched evaluation budgets, with DAF treated as a benchmark rather than as the source of the result. Self-citations to DAF [2], FST4, and stability measures [34] provide background, baselines, and tooling, but the load-bearing arguments do not rest on them: the DAF-style contrast is re-derived in Eqs. (2)-(3) and tested in ablations. No equation equates a 'prediction' to an input parameter, no uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled in via citation. The paper's acknowledged lack of a formal guarantee for the adaptive sampler (Section 6) is a limitation and a correctness risk, not a circularity.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a set of default hyperparameters (y, y_b, rho_u, h, warm-up) that are chosen by hand and partially fixed on madelon, and on the domain assumption that the criteria used (1-NN wrapper, multinomial Bhattacharyya) are meaningful. No new physical entities are introduced. The method's own statistic s(f) is an algorithmic construct, not an invented entity with independent evidence requirements.

free parameters (6)
  • forward per-step budget y = 100 (default; swept 10-200)
    Number of candidate evaluations per forward step; fixed on madelon probes before gisette/reuters runs. Controls the cost-quality trade-off.
  • backward per-step budget y_b = 50 (default)
    Number of removal candidates per backward step; chosen to balance cost against floating-correction quality.
  • exploration floor fraction rho_u = 0.2
    Fraction of each batch drawn uniformly from least-evaluated features; kept constant, mildly asymmetric (not worse than 0.1 in sensitivity probes).
  • forgetting horizon h = 100 evaluations
    Exponential forgetting rate for online statistics; reported as never the binding constraint in sensitivity analysis.
  • warm-up probes m0@r = 200@10 (madelon), 1000@25 (gisette), 2000@25 (reuters)
    Random probes seeding the statistics; the only setting scaled with D, chosen by hand without a dedicated sensitivity study across D.
  • softmax temperature tau = IQR/1.349 (robust auto)
    Derived from the score distribution, so no manual tuning; but it depends on the scale of the accumulated scores and is not fully free.
assumptions (4)
  • domain assumption The feature selection criterion J can be evaluated at arbitrary subsets at a cost that depends only on subset size.
    The method's cost model and anytime claims assume J is a black-box oracle (Section 3.2 and Section 4).
  • domain assumption 1-NN wrapper accuracy with 3-fold CV is a meaningful criterion for madelon and gisette, and multinomial Bhattacharyya distance is a trustworthy criterion for reuters.
    The entire experimental evaluation rests on these criteria; the paper explicitly argues that the reuters filter criterion 'deserves trust' so that holdout transfer is meaningful (Sections 4-5).
  • domain assumption The online batch-standardized contrast s(f) = mu_f - mu_bar_f converges to a useful ordering of features as the search progresses.
    This is the core learning mechanism; no convergence proof is given, only empirical validation via ablations (Section 3.1 and Section 5).
  • domain assumption The uniform exploration floor keeps every feature reachable, so early mis-rankings are self-correcting.
    Design assumption used to argue against deterministic truncation of the candidate pool (Section 3.2); the paper provides no formal reachability guarantee.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Stochastic Sequential Search in Very-High-Dimensional Feature Selection." pith.science (2026). https://pith.science/paper/EN4SRSHE

@misc{pith2026260801502,
  author       = {Pith},
  title        = {Pith review of: Stochastic Sequential Search in Very-High-Dimensional Feature Selection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EN4SRSHE}},
  note         = {Machine review of arXiv:2608.01502}
}
read the original abstract

Sequential subset search -- forward selection with floating backtracking and its descendants -- remains the quality reference in feature selection, but every member of the family sweeps the full pool of remaining candidate features at each step, which excludes it from very-high-dimensional problems; there, only individual-feature ranking remains practical, and it models feature interplay weakly or not at all. We introduce a budgeted sampled step operator pair that replaces the full sweeps by a fixed number of candidate evaluations per step. Candidates are drawn by temperature-controlled softmax sampling from dependency-aware per-feature statistics learned online from every criterion evaluation the search performs, guarded by a uniform exploration floor; per-step cost becomes independent of dimensionality. Substituting the operators turns any sequential method into its stochastic counterpart, defining the Stochastic Sequential Search (SSS) family; we study the stochastic counterpart of floating search, sSFFS. On 500-dimensional madelon, sSFFS retains at least 97% of the full-SFFS criterion value at every subset size at about a quarter of its evaluations, while uniform sampling at the same budget collapses on madelon's synergistic features. On 5,000-dimensional gisette, far beyond full-SFFS reach, sSFFS exceeds the saturated criterion level of DAF and BIF ranking at matched budgets; holdout validation shows that at 500 training samples the binding constraint beyond the sequential frontier becomes the criterion, not the search. On 10,105-dimensional reuters, under a trustworthy multinomial filter criterion, sSFFS dominates BIF and DAF on the search objective and on holdout accuracy at every subset size, in about two minutes of single-core evaluation work. A verified standalone implementation accompanies the paper.

Figures

Figures reproduced from arXiv: 2608.01502 by the authors.

Figure 1
Figure 1. Anytime best-per-size criterion value on madelon ( [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. Where the budget goes under forward-only capping: with forward steps capped at [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Budget spent uniformly vs. informed (madelon, [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Budget-response of sSFFS on madelon (train-criterion axis: wrapper-1NN 3-fold CV value on the training half — the optimization target; validated holdout accuracies are quoted in the text). Each point: 10-seed mean (±std in the top row) of the best criterion value attai…
Figure 5
Figure 5. Figure 5: The fixed-d slices of [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 33 canonical work pages

  1. [1]

    Kohavi and G

    R. Kohavi and G. H. John. Wrappers for feat. subs. sel.Artif. Intell., 97(1-2):273–324, 1997

    work page 1997

  2. [2]

    Fast dependency-aware feature selection in very- high-dimensional pattern recognition

    Petr Somol, Jiˇ r ´ ı Grim, and Pavel Pudil. Fast dependency-aware feature selection in very- high-dimensional pattern recognition. InProc. IEEE Int. Conference on Systems, Man, and Cybernetics (SMC 2011), pages 502–509. IEEE, 2011

    work page 2011

  3. [3]

    L. I. Kuncheva. A stability index for FS. InProc. 25thIASTEDInt. Mul.-Conf. AIAP’07, pages 390–395. ACTA Pr., 2007

    work page 2007

  4. [4]

    Pudil, J

    P. Pudil, J. Novoviˇ cov´ a, and J. Kittler. Floating search methods in FS.Patt. Rec. Lett., 15(11):1119–1125, 1994

    work page 1994

  5. [5]

    Stockman

    George C. Stockman. A minimax algorithm better than alpha-beta?Artificial Intelligence, 12(2):179–196, 1979

    work page 1979

  6. [6]

    Lazier than lazy greedy

    Baharan Mirzasoleiman, Ashwinkumar Badanidiyuru, Amin Karbasi, Jan Vondr´ ak, and An- dreas Krause. Lazier than lazy greedy. InProc. 29th AAAI Conference on Artificial Intelligence (AAAI-15), 2015. arXiv:1409.7938

    arXiv 2015

  7. [7]

    Large Scale Attribute Selection Using Wrappers,

    Martin G¨ utlein, Eibe Frank, Mark A. Hall, and Andreas Karwath. Large-scale attribute se- lection using wrappers. InProc. IEEE Symposium on Computational Intelligence and Data Mining (CIDM 2009), pages 332–339, 2009. Method also documented in M. G¨ utlein, “Large Scale Attribute Selection Using Wrappers,” diploma thesis, Univ. Freiburg, 2006. 21

    work page 2009

  8. [8]

    Somol, J

    P. Somol, J. Novoviˇ cov´ a, and P. Pudil. Improving sequential feature selection methods perfor- mance by means of hybridization. InProc. 6th IASTED Int. Conf. on Advances in Computer Science and Engineering. ACTA Press, 2010

    work page 2010

Show all 34 references
  1. [9]

    Improving feature selection process resistance to failures caused by curse-of-dimensionality effects.Kybernetika, 47(3):401–425, 2011

    Petr Somol, Jiˇ r ´ ı Grim, Jana Novoviˇ cov´ a, and Pavel Pudil. Improving feature selection process resistance to failures caused by curse-of-dimensionality effects.Kybernetika, 47(3):401–425, 2011

    2011

  2. [10]

    P. A. Devijver and J. Kittler.Pattern Recognition: A Statistical Approach. Prentice Hall, 1982

    1982

  3. [11]

    Somol and P

    P. Somol and P. Pudil. Oscillating search algorithms for FS. InICPR 2000, volume 02, pages 406–409. IEEE Comp. Sc., 2000

    2000

  4. [12]

    Somol, J

    P. Somol, J. Novoviˇ cov´ a, J. Grim, and P. Pudil. Dynamic oscillating search algorithms for FS. InICPR 2008. IEEE Comp. Soc., 2008

    2008

  5. [13]

    A. W. Whitney. A direct method of nonparametric measurement selection.IEEE Trans. Comput., 20(9):1100–1103, 1971

    1971

  6. [14]

    Somol, P

    P. Somol, P. Pudil, J. Novoviˇ cov´ a, and P. Pacl ´ ık. Adaptive floating search methods in FS. Pattern Recogn. Lett., 20(11-13):1157–1163, 1999

    1999

  7. [15]

    Stephen D. Stearns. On selecting features for pattern classifiers. InProc. 3rd Int. Joint Conference on Pattern Recognition, pages 71–75, Coronado, CA, 1976

    1976

  8. [16]

    Gan, Bashar Awwad Shiekh Hasan, and Chun S

    John Q. Gan, Bashar Awwad Shiekh Hasan, and Chun S. L. Tsui. A filter-dominating hybrid sequential forward floating search method for feature subset selection in high-dimensional space. International Journal of Machine Learning and Cybernetics, 5(3):413–423, 2014

    2014

  9. [17]

    Somol, Bart Baesens, P

    P. Somol, Bart Baesens, P. Pudil, and Jan Vanthienen. Filter- versus wrapper-based feature selection for credit scoring.Int. J. of Intelligent Systems, 20(10):985–1000, 2006

    2006

  10. [18]

    A hybrid feature selection scheme for high- dimensional data.Engineering Applications of Artificial Intelligence, 113:104894, 2022

    Mohammad Ahmadi Ganjei and Reza Boostani. A hybrid feature selection scheme for high- dimensional data.Engineering Applications of Artificial Intelligence, 113:104894, 2022

    2022

  11. [19]

    High-dimensional hybrid feature selection using interaction information- guided search.Knowledge-Based Systems, 145:59–66, 2018

    Songyot Nakariyakul. High-dimensional hybrid feature selection using interaction information- guided search.Knowledge-Based Systems, 145:59–66, 2018

    2018

  12. [20]

    A greedy feature selection algorithm for Big Data of high dimensionality

    Ioannis Tsamardinos, Giorgos Borboudakis, Pavlos Katsogridakis, Polyvios Pratikakis, and Vassilis Christophides. A greedy feature selection algorithm for Big Data of high dimensionality. Machine Learning, 108(2):149–202, 2018

    2018

  13. [21]

    Dimakis, Sahand Negahban, and Joydeep Ghosh

    Rajiv Khanna, Ethan Elenberg, Alexandros G. Dimakis, Sahand Negahban, and Joydeep Ghosh. Scalable greedy feature selection via weak submodularity. InProc. 20th Int. Con- ference on Artificial Intelligence and Statistics (AISTATS 2017), volume 54 ofPMLR, pages 1560–1568, 2017

    2017

  14. [22]

    Approximation guarantees of stochastic greedy algorithms for subset selection

    Chao Qian, Yang Yu, and Ke Tang. Approximation guarantees of stochastic greedy algorithms for subset selection. InProc. 27th Int. Joint Conference on Artificial Intelligence (IJCAI-18), pages 1478–1484, 2018

    2018

  15. [23]

    Submodular optimization under noise

    Avinatan Hassidim and Yaron Singer. Submodular optimization under noise. InProc. 2017 Conference on Learning Theory (COLT), volume 65 ofPMLR, pages 1069–1122, 2017. 22

    2017

  16. [24]

    Quinn, and Mohamed-Slim Alouini

    Fares Fourati, Vaneet Aggarwal, Christopher J. Quinn, and Mohamed-Slim Alouini. Random- ized greedy learning for non-monotone stochastic submodular maximization under full-bandit feedback. InProc. 26th Int. Conference on Artificial Intelligence and Statistics (AISTATS 2023), 2...

    2023 arXiv

  17. [25]

    Performance-complexity tradeoffs in greedy weak submodular maximization with random sampling, 2019

    Abolfazl Hashemi, Haris Vikalo, and Gustavo de Veciana. Performance-complexity tradeoffs in greedy weak submodular maximization with random sampling, 2019. arXiv:1907.09064 (rev. 2021); partial version presented at IEEE ICASSP 2021

    2019 arXiv

  18. [26]

    Feature selection as a one-player game

    Romaric Gaudel and Mich` ele Sebag. Feature selection as a one-player game. InProc. 27th Int. Conference on Machine Learning (ICML 2010), pages 359–366, Haifa, Israel, 2010

    2010

  19. [27]

    Multi-armed bandit based feature selection

    Kunpeng Liu, Haibo Huang, Wei Zhang, Ahmad Hariri, Yanjie Fu, and Kien Hua. Multi-armed bandit based feature selection. InProc. 2021 SIAM Int. Conference on Data Mining (SDM 2021), pages 316–323, 2021

    2021

  20. [28]

    Thompson sampling for combinatorial bandits and its application to online feature selection

    Audrey Durand and Christian Gagn´ e. Thompson sampling for combinatorial bandits and its application to online feature selection. InSequential Decision-Making with Big Data: Papers from the AAAI-14 Workshop, pages 6–9, 2014

    2014

  21. [29]

    Variable selection via Thompson sampling.Journal of the American Statistical Association, 118(541):287–304, 2023

    Yunfan Liu and Veronika Roˇ ckov´ a. Variable selection via Thompson sampling.Journal of the American Statistical Association, 118(541):287–304, 2023

    2023

  22. [30]

    Automating feature subspace exploration via multi-agent reinforcement learning

    Kunpeng Liu, Yanjie Fu, Pengfei Wang, Le Wu, Rui Bo, and Xiaolin Li. Automating feature subspace exploration via multi-agent reinforcement learning. InProc. 25th ACM SIGKDD Int. Conference on Knowledge Discovery & Data Mining (KDD ’19), 2019

    2019

  23. [31]

    Automated feature selection: A reinforcement learning perspective.IEEE Transactions on Knowledge and Data Engineering, 35(3):2272–2284, 2023

    Kunpeng Liu, Yanjie Fu, Le Wu, Xiaolin Li, Charu Aggarwal, and Hui Xiong. Automated feature selection: A reinforcement learning perspective.IEEE Transactions on Knowledge and Data Engineering, 35(3):2272–2284, 2023

    2023

  24. [32]

    ˇS. J. Raudys. Feature over-selection. InProc. S+SSPR, volume LNCS 4109, pages 622–631. Springer, 2006

    2006

  25. [33]

    Result analysis of the NIPS 2003 feature selection challenge

    Isabelle Guyon, Steve Gunn, Asa Ben-Hur, and Gideon Dror. Result analysis of the NIPS 2003 feature selection challenge. InAdvances in Neural Information Processing Systems 17 (NIPS 2004), pages 545–552. MIT Press, 2005

    2003

  26. [34]

    Evaluating stability and comparing output of feature se- lectors that optimize feature subset cardinality.IEEE Trans

    Petr Somol and Jana Novoviˇ cov´ a. Evaluating stability and comparing output of feature se- lectors that optimize feature subset cardinality.IEEE Trans. Pattern Anal. Mach. Intell., 32(11):1921–1939, 2010. 23

    1921

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.