Pith. sign in

REVIEW 3 major objections 6 minor 51 references

Prioritizing Search Space Regions in the Low Autocorrelation Binary Sequences Problem

T0 review · 3 major / 6 minor · reviewed 2026-07-14 · grok-4.5

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

Pith's one-line read Thompson sampling that treats LABS partitions as bandit arms finds 35 new record binary sequences and the longest with merit factor above 8.

desk verdict Checkable new LABS records (including longest F>8 at L=451) on a solid two-stage GPU pipeline; the TS prioritization story is real but under-ablated and partly saturated by the F/7 clamp. read the letter →

arxiv 2607.09688 v1 pith:I3WR5JFK submitted 2026-06-17 cs.LG

Blaž Pšeničnik , Borko Bošković , Jan Popić , Janez Brest This is my paper
classification cs.LG
keywords LABSmeritfactorThompsonsamplingself-avoidingwalksrestrictionclassesbinarysequencesGPUparallelsearchmulti-armedbandit
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper attacks the low-autocorrelation binary sequence (LABS) problem: find a ±1 string of length L whose aperiodic autocorrelations are as small as possible, measured by the merit factor F. Exhaustive search is impossible for large L, so the space is cut into restriction classes (partitions). Instead of ranking those partitions by a fixed heuristic potential, the authors treat each partition as an arm of a multi-armed bandit and use fractional Thompson sampling to decide, online, where to spend the next self-avoiding walk. Parallel GPU walks share a common posterior, a Bloom filter prevents cycles, and the best candidates are refined without the original skew-symmetry constraints. On sequences of length 450–527 and 573 the method raises 35 previously best-known merit factors and produces the longest known sequence with F > 8 (L = 451, F = 8.0555). The practical claim is that data-driven reallocation of search effort beats static partition ordering for high-performance LABS optimization.

What carries the argument

TS-SAW: each partition is an arm whose Beta posterior is updated fractionally with the scaled merit factor r_t = clamp(F/7) obtained from a parallel self-avoiding walk; Thompson samples select the next walk, a shared global posterior coordinates many walks, and the top-m candidates are refined by unrestricted priority-queue search.

What would settle it

Re-run the identical compute budget on the same lengths with the original static normalized-potential ranking (no Thompson sampling) and check whether the same 35 records, or the L=451 sequence with F≥8, still appear; absence of the records would falsify the claim that online prioritization is responsible.

Watch

Extended reading notes

Core claim

Modeling LABS restriction classes as arms in a multi-armed bandit and allocating GPU self-avoiding walks by fractional Thompson sampling yields new best-known merit factors for 35 lengths in 450 ≤ L ≤ 527 and for L = 573, including the longest binary sequence yet reported with merit factor exceeding 8.0 (L = 451, F = 8.0555).

Load-bearing premise

The method assumes that the scaled merit factor from a single stochastic walk, after delayed batch updates, is a reliable enough signal of a partition’s true quality that the resulting sampling bias reflects genuine region superiority rather than noise or the untuned second-stage budget.

Editorial extensions

If this is right

  • New record binary sequences become available for communications, radar, and GNSS spreading-code design in the length range 450–573.
  • Static partition potentials can be replaced by online bandit feedback whenever the search space admits a natural partition into restriction classes.
  • The two-stage pattern (constrained parallel search followed by unrestricted refinement of the top-m candidates) is a reusable template for other hard binary combinatorial problems.
  • Empirical sampling frequencies supply a data-driven ranking of partitions that can be inspected and reused even without further Thompson sampling.

Reading between the lines

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

  • The same bandit-over-partitions idea could be tried on other Golay-type or aperiodic-correlation problems whose natural group actions already induce restriction classes.
  • Because the second-stage budget m=15 was chosen for hardware convenience rather than tuned, a modest increase in m or adaptive hand-off criteria might still raise several of the remaining unimproved lengths.
  • The observed mismatch between normalized-potential rank and sampling frequency suggests that classical potential heuristics systematically mis-order certain long partitions; re-deriving potentials from the learned posteriors could improve pure construction methods.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes TS-SAW, a hybrid LABS solver that treats restriction-class partitions as arms in a multi-armed bandit and allocates GPU-parallel self-avoiding walks via fractional Thompson sampling, followed by a two-stage refinement that lifts top candidates into the unrestricted space (with sequence operators for even lengths). Using shared Beta posteriors, Bloom-filter cycle prevention, and linear-time skew-symmetric neighborhood evaluation, the authors report improved best-known merit factors for 35 lengths in 450≤L≤527 and for L=573, including a new longest sequence with F>8.0 (L=451, F=8.0555). Hex encodings of the improved sequences are published (Table 3), and Table 2 / Figure 5 are offered as evidence that sampling shifts away from pure normalized-potential ranking toward empirically better partitions.

Significance. If the published sequences check out, the empirical contribution is substantial for the LABS community: 35 new records in a hard length regime, a new longest F≥8 sequence (L=451), and an improved F>7 sequence at L=573. These results are independently falsifiable from Table 3 and do not depend on the bandit narrative. The engineering stack (CUDA SAW blocks, shared posteriors, Bloom filter, two-stage host refinement) is a credible scalable framework and builds usefully on the authors’ prior SAW and dual-step solvers. The methodological claim—that online Thompson sampling is a valuable, data-driven way to prioritize partitions—would matter for combinatorial search more broadly if better supported; as written, that claim is the weaker pillar relative to the checkable records.

major comments (3)
  1. [Method (reward scaling); Table 3] Reward design saturates exactly in the reported operating regime. The Method section scales rewards as r_t = clamp(F/7). Every sequence with F≥7 therefore yields identical reward 1. In Table 3 almost all “Our F” values exceed 7 (and the headline L=451 result is F=8.0555), so fractional TS updates cannot rank partitions by how high F goes above 7—only by how often they hit the ceiling. This undercuts the abstract and Results claim that TS “prioritizes partitions with better observed performance” / “higher merit factors.” A non-saturating map (e.g., soft normalization against a running high-water mark, rank-based reward, or a higher clamp informed by the F>8 target) is needed, or the claim must be restated as hit-rate learning above F=7.
  2. [Results (Table 2, Figure 5); Abstract] No equal-budget ablation isolates the contribution of Thompson sampling. The paper compares empirical sampling frequencies to normalized-potential rank (Table 2) and shows posterior concentration (Figure 5), but never runs the same two-stage pipeline, walk budgets, iteration count (100,000), and m=15 second-stage capacity with static normalized-potential ordering (or uniform/random partition selection). Without that control, the 35 record improvements cannot be attributed to online prioritization rather than to the underlying SAW engine, GPU budget, two-stage refinement, or sequence operators already present in prior work. An equal-compute ablation is load-bearing for the title claim and for the abstract sentence that TS “confirm[s] the value of online, data-driven resource allocation.”
  3. [Results (m=15 paragraph); Method (shared posterior updates)] Second-stage capacity m=15 is stated as chosen from node constraints “without performing any parameter tuning,” yet it gates which first-stage candidates ever see unrestricted refinement. Combined with batch-delayed shared (α,β) updates across CUDA blocks, this free parameter can confound the apparent value of partition prioritization: a partition that occasionally produces a top-m seed may dominate final records even if its mean walk quality is mediocre. Sensitivity of final F to m (and to update batching) should be reported, or m should be justified against a tuned baseline.
minor comments (6)
  1. [Algorithm 1] Algorithm 1 mixes ˆθ_k (line 3) and ˜θ_k (line 5) for the same Thompson sample; unify the notation.
  2. [Introduction] Introduction: “can be can be grouped into eight mutually equivalent classes” — duplicate wording.
  3. [Introduction] Introduction: “slightly greater than 2(L−3)” should be exponential notation 2^{L−3} for clarity.
  4. [Figure 4] Figure 4 x-axis is discontinuous by design; state explicitly in the caption that only improved lengths are plotted so readers do not infer a continuous interval.
  5. [Table 1; Results L=461 experiment] Table 1 reports partitions for p=81, g=7, while the TS case study uses p=67, g=5; a short note linking which (p,g) pairs were used for which L ranges would aid reproducibility.
  6. [Method (Bloom filter paragraph)] Clarify Bloom-filter false-positive rate used in practice and whether false positives ever truncated walks in the reported runs.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: new merit-factor records are independently checkable from published sequences; self-citations supply reusable algorithmic components, not definitional premises.

full rationale

The paper is an empirical combinatorial-optimization study. Merit factor F is the classical external definition F = L^{2}/(2E(S)) (Eq. 2); the sequences that achieve the claimed new records are published in hexadecimal form in Table 3 and can be decoded and re-evaluated by any third party without reference to Thompson sampling, the clamp(F/7) reward, or the authors’ prior solvers. The two-stage pipeline and self-avoiding-walk engine are taken from the authors’ earlier works [33,42], but those works supply concrete, independently executable algorithms whose outputs (binary sequences and their energies) are falsifiable outside the present paper; they do not define the quantity being optimized. Fractional Thompson sampling with shared Beta posteriors is used only as an online resource-allocation heuristic; the paper never claims that the observed sampling frequencies (Table 2, Figure 5) constitute a first-principles derivation or a uniqueness result. Consequently there is no self-definitional loop, no fitted parameter re-labeled as a prediction, and no load-bearing uniqueness theorem imported from the same authors. The single minor self-citation pattern is ordinary reuse of prior algorithmic infrastructure and does not raise the circularity score above 1.

Assumptions & free parameters 5 free parameters · 5 assumptions · 1 invented entities

The central claims rest on standard LABS definitions, prior partition/skew-symmetry reductions, and a small set of hand-chosen algorithmic knobs (reward scale, candidate count m, walk budgets). No new physical entities are postulated; the method is an algorithmic composition. Free parameters are the main non-derived ingredients that affect which partitions get compute and which candidates enter stage two.

free parameters (5)
  • reward_scale_clamp_F_over_7
    Merit factor is mapped to r_t = clamp(F/7) for fractional Beta updates; the constant 7 is chosen from expected MF range in prior work, not derived here, and directly shapes posterior updates.
  • second_stage_candidate_count_m
    m=15 top walks are passed to unrestricted refinement; authors state this was set from node capacity without parameter tuning, yet it gates which solutions can improve further.
  • walk_and_iteration_budgets_Ti_Tu_Tr_100000_iters
    Self-avoiding walk length, unrestricted depth, rotation radius, and 100k outer iterations are computational knobs inherited or set for the campaign; they determine search depth and reported records.
  • Beta_prior_alpha_beta_equals_1
    Uniform Beta(1,1) priors on every partition arm initialize the bandit; standard but still a modeling choice that affects early exploration.
  • partition_parameters_p_and_g
    Choice of prefix length p and number of summands g (e.g., p=67,g=5 for L=461; p=81,q=7 examples) defines the arm set and is taken from prior partition machinery rather than optimized end-to-end.
assumptions (5)
  • domain assumption Aperiodic autocorrelation energy and merit factor F=L^2/(2E) correctly score binary sequences for the LABS objective.
    Standard LABS formulation used throughout Introduction and Method; all reported improvements are with respect to this scalar.
  • domain assumption Skew-symmetry and restriction-class partitions reduce the search space while still containing high-merit sequences for long L.
    Invoked via Equations (4)–(5) and the two-stage pipeline; stage one only searches inside these constrained regions.
  • ad hoc to paper Fractional Thompson sampling with continuous rewards in [0,1] is a valid online allocation rule for partition selection under stochastic walk feedback.
    Method section adopts the fractional update of [49] without a LABS-specific regret analysis; correctness of prioritization is argued empirically.
  • ad hoc to paper Bloom-filter false positives for visited sequences are rare enough not to distort self-avoiding walks materially.
    Method states negligible error chance to justify longer walks in GPU shared memory; no measured false-positive rate is given.
  • domain assumption Sequence operators that append/remove ends preserve enough quality to seed even-length refinement from odd-length optima.
    Used at the end of Method to extend odd-length gains to even L; depends on prior operator literature [19].
invented entities (1)
  • TS-SAW algorithm (Thompson sampling + parallel SAW + two-stage refine)
    purpose: Name and package the hybrid procedure that allocates walks across partitions and refines top candidates unrestricted.
    Composition of known components; not a physical entity. independent_evidence is false because the entity is defined by this paper’s procedure rather than an external observable.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Prioritizing Search Space Regions in the Low Autocorrelation Binary Sequences Problem." pith.science (2026). https://pith.science/paper/I3WR5JFK

@misc{pith2026260709688,
  author       = {Pith},
  title        = {Pith review of: Prioritizing Search Space Regions in the Low Autocorrelation Binary Sequences Problem},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I3WR5JFK}},
  note         = {Machine review of arXiv:2607.09688}
}
abstract

Low autocorrelation binary sequences problem (LABS) is a hard combinatorial optimization challenge with important applications in communications, signal processing, and satellite navigation. This paper proposes a hybrid search framework that combines Thompson sampling with parallel self-avoiding walks to adaptively allocate computational effort across restriction classes of the LABS search space. By modeling partitions as arms in a multi-armed bandit setting, the proposed method dynamically shifts search resources toward partitions that empirically produce higher merit factors while maintaining exploration of less-sampled regions. The approach is further accelerated through GPU-parallel execution, shared posterior updates, efficient neighborhood evaluation, and a Bloom filter for cycle prevention. In addition, we use a two-stage optimization strategy that first searches constrained partitioned skew-symmetric spaces and then refines the best candidates in the unrestricted space. Experiments on long binary sequences show that the proposed method improves the previously best-known results for 35 sequence lengths in the range $450 \le L \le 527$ and for $L=573$. In particular, we report a new longest sequence with merit factor exceeding $8.0$, obtained for $L=451$. The results also show that Thompson sampling effectively prioritizes partitions with better observed performance, confirming the value of online, data-driven resource allocation in LABS optimization. Overall, the proposed framework provides a scalable and effective strategy for high-performance merit factor maximization.

Figures

Figures reproduced from arXiv: 2607.09688 by the authors.

Figure 1
Figure 1. Two-dimensional projection of the search space for sequence lengths [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Flowchart illustrating the main steps of the proposed TS-SAW algorithm for generating binary sequences [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Two-step optimization pipeline combining parallel stochastic search and unconstrained refinement. The top [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Record merit factors achieved by the improved sequences, together with the previously best-known merit [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Probability density functions at different iterations for selected partitions for [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

51 extracted references · 3 linked inside Pith

  1. [1]

    M. Golay. A class of finite binary sequences with alternate auto-correlation values equal to zero (corresp.).IEEE Transactions on Information Theory, 18(3):449–450, 1972

    1972

  2. [2]

    J. E. Littlewood. On polynomialsPn ±zm,Pn eamizm, z=e 0i.Journal of the London Mathematical Society, s1-41(1):367–376, 1966

    1966

  3. [3]

    A survey of the merit factor problem for binary sequences

    Jonathan Jedwab. A survey of the merit factor problem for binary sequences. InInternational Conference on Sequences and Their Applications, pages 30–55. Springer, 2004

    2004

  4. [4]

    Moments of autocorrelation demerit factors of binary sequences.Designs, Codes and Cryptography, pages 1–45, 2024

    Daniel J Katz and Miriam E Ramirez. Moments of autocorrelation demerit factors of binary sequences.Designs, Codes and Cryptography, pages 1–45, 2024

    2024

  5. [5]

    Multi-criteria selection of a synchroni- sation word for low-power iot receivers based on the iqrf standard.Scientific Reports, 16(1):8777, 2026

    Milan Skula, Martin Pies, Radovan Hajovsky, Jan Velicka, and David Vala. Multi-criteria selection of a synchroni- sation word for low-power iot receivers based on the iqrf standard.Scientific Reports, 16(1):8777, 2026

    2026

  6. [6]

    Bernasconi

    J. Bernasconi. Low autocorrelation binary sequences: statistical mechanics and configuration space analysis.J. Physsique, 48:559–567, 4 1987

    1987

  7. [7]

    Shapiro, Gordon H

    Irwin I. Shapiro, Gordon H. Pettengill, Michael E. Ash, Melvin L. Stone, William B. Smith, Richard P. Ingalls, and Richard A. Brockelman. Fourth test of general relativity: Preliminary results.Phys. Rev. Lett., 20:1265–1269, 5 1968

    1968

  8. [8]

    Large-scale gnss spreading code optimization

    Alan Yang, Tara Mina, Stephen Boyd, and Grace Gao. Large-scale gnss spreading code optimization. In Proceedings of the 37th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2024), pages 948–957, 2024

    2024

Show all 51 references
  1. [9]

    Spreading code optimization for low-earth orbit satellites via mixed-integer convex programming.EURASIP Journal on Advances in Signal Processing, 2024(1):67, 2024

    Alan Yang, Tara Mina, and Grace Gao. Spreading code optimization for low-earth orbit satellites via mixed-integer convex programming.EURASIP Journal on Advances in Signal Processing, 2024(1):67, 2024

    2024

  2. [10]

    A competitive nisq and qubit-efficient solver for the labs problem.arXiv preprint arXiv:2506.17391, 2025

    Marco Sciorilli, Giancarlo Camilo, Thiago O Maciel, Askery Canabarro, Lucas Borges, and Leandro Aolita. A competitive nisq and qubit-efficient solver for the labs problem.arXiv preprint arXiv:2506.17391, 2025

    2025

  3. [11]

    Evidence of scaling advantage for the quantum approximate optimization algorithm on a classically intractable problem.Science Advances, 10(22):eadm6761, 2024

    Ruslan Shaydulin, Changhao Li, Shouvanik Chakrabarti, Matthew DeCross, Dylan Herman, Niraj Kumar, Jeffrey Larson, Danylo Lykov, Pierre Minssen, Yue Sun, et al. Evidence of scaling advantage for the quantum approximate optimization algorithm on a classically intractable problem...

    2024

  4. [12]

    Sequences with small correlation.Designs, Codes and Cryptography, 78, 01 2016

    Kai-Uwe Schmidt. Sequences with small correlation.Designs, Codes and Cryptography, 78, 01 2016

    2016

  5. [13]

    Code selection for CDMA systems.Department of Information Studies, University of Tampere, Finland, 1997

    Kimmo Kettunen. Code selection for CDMA systems.Department of Information Studies, University of Tampere, Finland, 1997

    1997

  6. [14]

    Mullen and Daniel Panario

    Gary L. Mullen and Daniel Panario. Other correlation measures. InHandbook of Finite Fields, chapter 10.3.5, pages 322–324. Chapman & Hall/CRC, 1st edition, 2013

    2013

  7. [15]

    Oddaljenost neperiodi ˇcnih binarnih zaporedij glede na dve meri avtokorelacijskih lastnosti.Elektrotehniski Vestnik, 92(3):97–103, 2025

    Janez Brest, Aljaž Brest, Blaž Pšeniˇcnik, Jan Popiˇc, and Boskovi´c Borko. Oddaljenost neperiodi ˇcnih binarnih zaporedij glede na dve meri avtokorelacijskih lastnosti.Elektrotehniski Vestnik, 92(3):97–103, 2025. (In Slovene)

    2025

  8. [16]

    Metastable states in short-ranged p-spin glasses

    Viviane M de Oliveira, José F Fontanari, and Peter F Stadler. Metastable states in short-ranged p-spin glasses. Journal of Physics A: Mathematical and General, 32(50):8793, 12 1999

    1999

  9. [17]

    Uniform manifold approximation and projection.Nature Reviews Methods Primers, 4(1):82, 2024

    John Healy and Leland McInnes. Uniform manifold approximation and projection.Nature Reviews Methods Primers, 4(1):82, 2024

    2024

  10. [18]

    Low autocorrelation binary sequences.Journal of Physics A: Mathematical and Theoretical, 49(16):165001, 2016

    Tom Packebusch and Stephan Mertens. Low autocorrelation binary sequences.Journal of Physics A: Mathematical and Theoretical, 49(16):165001, 2016

    2016

  11. [19]

    New classes of binary sequences with high merit factor.arXiv preprint arXiv:2206.12070, 2022

    Miroslav Dimitrov. New classes of binary sequences with high merit factor.arXiv preprint arXiv:2206.12070, 2022

    2022 arXiv

  12. [20]

    Exhaustive search for low-autocorrelation binary sequences.Journal of Physics A: Mathematical and General, 29(18):L473, 1996

    Stephan Mertens. Exhaustive search for low-autocorrelation binary sequences.Journal of Physics A: Mathematical and General, 29(18):L473, 1996

    1996

  13. [21]

    Exploiting relaxation in local search for labs.Annals of Operations Research, 156:129–141, 12 2007

    Steven Prestwich. Exploiting relaxation in local search for labs.Annals of Operations Research, 156:129–141, 12 2007

    2007

  14. [22]

    Improved branch-and-bound for low autocorrelation binary sequences.arXiv preprint arXiv:1305.6187, 2013

    Steven D Prestwich. Improved branch-and-bound for low autocorrelation binary sequences.arXiv preprint arXiv:1305.6187, 2013

    2013 arXiv

  15. [23]

    Parallele optimierungsstrategien des labs-problems in einem gpu-grid, 2010

    Jens Wiggenbrock. Parallele optimierungsstrategien des labs-problems in einem gpu-grid, 2010. (In German)

    2010

  16. [24]

    Hoholdt and H.E

    T. Hoholdt and H.E. Jensen. Determination of the merit factor of legendre sequences.IEEE Transactions on Information Theory, 34(1):161–164, 1988

    1988

  17. [25]

    Borwein, K.-K.S

    P. Borwein, K.-K.S. Choi, and J. Jedwab. Binary sequences with merit factor greater than 6.34.IEEE Transactions on Information Theory, 50(12):3234–3249, 2004

    2004

  18. [26]

    Efficient Optimization of the Merit Factor of Long Binary Sequences.IEEE Transactions on Information Theory, 57(12):8084–8094, 2011

    John Michael Baden. Efficient Optimization of the Merit Factor of Long Binary Sequences.IEEE Transactions on Information Theory, 57(12):8084–8094, 2011

    2011

  19. [27]

    Katz, and Kai-Uwe Schmidt

    Jonathan Jedwab, Daniel J. Katz, and Kai-Uwe Schmidt. Advances in the merit factor problem for binary sequences.Journal of Combinatorial Theory, Series A, 120(4):882–906, 2013

    2013

  20. [28]

    Local search algorithm for low autocorrelation binary sequences

    Kaoutar Farnane, Khalid Minaoui, and Driss Aboutajdine. Local search algorithm for low autocorrelation binary sequences. In2018 4th International Conference on Optimization and Applications (ICOA), pages 1–5, 2018

    2018

  21. [29]

    On the skew-symmetric binary sequences and the merit factor problem.Digital Signal Processing, 156:104793, 2025

    Miroslav Dimitrov. On the skew-symmetric binary sequences and the merit factor problem.Digital Signal Processing, 156:104793, 2025

    2025

  22. [30]

    Steven Halim, Roland H. C. Yap, and Felix Halim. Engineering Stochastic Local Search for the Low Autocorrela- tion Binary Sequence Problem. In Peter J. Stuckey, editor,Principles and Practice of Constraint Programming, pages 640–645, Berlin, Heidelberg, 2008. Springer Berlin H...

    2008

  23. [31]

    Gallardo, Carlos Cotta, and Antonio J

    José E. Gallardo, Carlos Cotta, and Antonio J. Fernández. Finding low autocorrelation binary sequences with memetic algorithms.Applied Soft Computing, 9(4):1252–1262, 2009

    2009

  24. [32]

    Low-autocorrelation binary sequences: On improved merit factors and runtime predictions to achieve them.Applied Soft Computing, 56:262–285, 2017

    Borko Boškovi´c, Franc Brglez, and Janez Brest. Low-autocorrelation binary sequences: On improved merit factors and runtime predictions to achieve them.Applied Soft Computing, 56:262–285, 2017

    2017

  25. [33]

    Parallel self-avoiding walks for a low-autocorrelation binary sequences problem.Journal of Computational Science, 77:102260, 2024

    Borko Boškovi´c, Jana Herzog, and Janez Brest. Parallel self-avoiding walks for a low-autocorrelation binary sequences problem.Journal of Computational Science, 77:102260, 2024

    2024

  26. [34]

    A Heuristic Algorithm for a Low Autocorrelation Binary Sequence Problem With Odd Length and High Merit Factor.IEEE Access, 6:4127–4134, 2018

    Janez Brest and Borko Boškovi´c. A Heuristic Algorithm for a Low Autocorrelation Binary Sequence Problem With Odd Length and High Merit Factor.IEEE Access, 6:4127–4134, 2018

    2018

  27. [35]

    Computational Search of Long Skew-symmetric Binary Sequences with High Merit Factors.MENDEL, 28(2):17–24, 12 2022

    Janez Brest and Borko Boškovi´c. Computational Search of Long Skew-symmetric Binary Sequences with High Merit Factors.MENDEL, 28(2):17–24, 12 2022

    2022

  28. [36]

    Analysis based on statistical distributions: A practical approach for stochastic solvers using discrete and continuous problems.Information Sciences, 633:469–490, 2023

    Jana Herzog, Janez Brest, and Borko Boškovi ´c. Analysis based on statistical distributions: A practical approach for stochastic solvers using discrete and continuous problems.Information Sciences, 633:469–490, 2023

    2023

  29. [37]

    Toward hybrid platform for evolutionary computations of hard discrete problems.Procedia Computer Science, 108:877–886, 2017

    Dominik ˙Zurek, Kamil Pi˛ etak, Marcin Pietro´n, and Marek Kisiel-Dorohinicki. Toward hybrid platform for evolutionary computations of hard discrete problems.Procedia Computer Science, 108:877–886, 2017. 14 APREPRINT- JULY14, 2026

    2017

  30. [38]

    Striving for performance of discrete optimisation via memetic agent-based systems in a hybrid CPU/GPU environment

    Kamil Pi˛ etak, Dominik˙Zurek, Marcin Pietro ´n, Andrzej Dymara, and Marek Kisiel-Dorohinicki. Striving for performance of discrete optimisation via memetic agent-based systems in a hybrid CPU/GPU environment. Journal of Computational Science, 31:151–162, 2019

    2019

  31. [39]

    A deep neural network as a tabu support in solving labs problem

    Dominik ˙Zurek, Marcin Pietro´n, Kamil Pi˛ etak, and Marek Kisiel-Dorohinicki. A deep neural network as a tabu support in solving labs problem. InInternational Conference on Computational Science, pages 237–243. Springer, 2022

    2022

  32. [40]

    New improvements in solving large labs instances using massively parallelizable memetic tabu search.arXiv preprint arXiv:2504.00987, 2025

    Zhiwei Zhang, Jiayu Shen, Niraj Kumar, and Marco Pistoia. New improvements in solving large labs instances using massively parallelizable memetic tabu search.arXiv preprint arXiv:2504.00987, 2025

    2025 arXiv

  33. [41]

    Scaling advantage with quantum-enhanced memetic tabu search for labs.arXiv preprint arXiv:2511.04553, 2025

    Alejandro Gomez Cadavid, Pranav Chandarana, Sebastián V Romero, Jan Trautmann, Enrique Solano, Taylor Lee Patti, and Narendra N Hegade. Scaling advantage with quantum-enhanced memetic tabu search for labs.arXiv preprint arXiv:2511.04553, 2025

    2025

  34. [42]

    Dual-step optimization for binary sequences with high merit factors.Digital Signal Processing, 165:105316, 2025

    Blaž Pšeniˇcnik, Rene Mlinariˇc, Janez Brest, and Borko Boškovi´c. Dual-step optimization for binary sequences with high merit factors.Digital Signal Processing, 165:105316, 2025

    2025

  35. [43]

    A tutorial on thompson sampling.Foundations and Trends (R) in Machine Learning, 11(1):1–96, 2018

    Daniel J Russo, Benjamin Van Roy, Abbas Kazerouni, Ian Osband, Zheng Wen, et al. A tutorial on thompson sampling.Foundations and Trends (R) in Machine Learning, 11(1):1–96, 2018

    2018

  36. [44]

    Katehakis and Arthur F

    Michael N. Katehakis and Arthur F. Veinott Jr. The multi-armed bandit problem: Decomposition and computation. Mathematics of Operations Research, 12(2):262–268, 1987

    1987

  37. [45]

    Thompson

    William R. Thompson. On the likelihood that one unknown probability exceeds another in view of the evidence of two samples.Biometrika, 25(3/4):285–294, 1933

    1933

  38. [46]

    On the theory of apportionment.American Journal of Mathematics, 57(2):450–456, 1935

    William R Thompson. On the theory of apportionment.American Journal of Mathematics, 57(2):450–456, 1935

    1935

  39. [47]

    ’A modern Bayesian look at the multi-armed bandit’ by Steven L

    Deepak K Agarwal. ’A modern Bayesian look at the multi-armed bandit’ by Steven L. Scott: Discussion.Applied Stochastic Models in Business & Industry, 26(6), 2010

    2010

  40. [48]

    An empirical evaluation of thompson sampling.Advances in neural information processing systems, 24, 2011

    Olivier Chapelle and Lihong Li. An empirical evaluation of thompson sampling.Advances in neural information processing systems, 24, 2011

    2011

  41. [49]

    Analysis of thompson sampling for the multi-armed bandit problem

    Shipra Agrawal and Navin Goyal. Analysis of thompson sampling for the multi-armed bandit problem. In Conference on learning theory, pages 39–1. JMLR Workshop and Conference Proceedings, 2012

    2012

  42. [50]

    Space/time trade-offs in hash coding with allowable errors.Communications of the ACM, 13(7):422–426, 1970

    Burton H Bloom. Space/time trade-offs in hash coding with allowable errors.Communications of the ACM, 13(7):422–426, 1970

    1970

  43. [51]

    Pads: Python algorithms and data structures, 2015

    David Eppstein. Pads: Python algorithms and data structures, 2015. MIT licensed software library. 15

    2015

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.