Pith. sign in

REVIEW 2 major objections 5 minor 100 references

Agents That Certify Their Own Exploits: Confidence-Scheduled Restricted Responses for Safe Opponent Exploitation

T0 review · 2 major / 5 minor · reviewed 2026-07-31 · grok-4.5

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

Pith's one-line read An agent can safely exploit flawed opponents by certifying the strategy it actually plays, not the model that proposed it.

desk verdict Clean engineering move: safety lives on the played strategy via an exact BR certificate, not on the opponent model that proposed it. read the letter →

arxiv 2607.28520 v1 pith:O53STODN submitted 2026-07-30 cs.GT cs.AIcs.MA

Boning Li , Longbo Huang This is my paper
classification cs.GTcs.AIcs.MA
keywords opponentexploitationsaferestrictedNashresponseanytime-validconfidencesequencesimperfect-informationgamesexploitabilitycertificateLeduchold'emtwo-playerzero-sum
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

In two-player zero-sum imperfect-information games, a Nash strategy locks in the game value but leaves money on the table against a flawed opponent. The hard case is diffuse deviation: slight mistakes spread across many decisions, so binary “release when sure” rules never fire, while a full best response to a half-built opponent model can be extremely exploitable. This paper introduces CS-RNR, which watches pooled action frequencies with anytime-valid confidence sequences, builds a conservative opponent model only from confirmed excesses, and turns that model into candidate counter-strategies via restricted-response solves at several pin levels. Before any candidate is played, the agent runs a full-tree best response on the complete strategy itself and commits only if the resulting certificate stays inside a user budget. Model quality then decides how much extra value is captured; the certificate alone bounds reference-relative expected loss, even against adaptive opponents. On Leduc the method earns several times the steady-state gain of a money-verified binary gate while every deployed strategy stays inside budget, and tens of thousands of audited hands across three games show no certificate violation.

What carries the argument

Budget-constrained confidence-scheduled restricted responses (CS-RNR): anytime-valid confidence sequences flag pooled frequency excesses, a conservative model feeds restricted-response solves over a pin grid, and each complete candidate is admitted only after an original-game full-tree best-response certificate clears a user budget and is committed atomically with the strategy.

What would settle it

Find a hand, seed, or adaptive opponent in the reported Leduc, Liar’s Dice, or 5-rank Leduc audits where the played strategy’s true expected loss exceeds its reported certificate plus the stated solver tolerance, or show that the same schedule on a game where only approximate best responses exist still claims the Proposition 1 bound.

Watch

Extended reading notes

Core claim

CS-RNR is the first online opponent-exploitation method whose safety guarantee is a certificate computed on the strategy the agent actually deploys. Every committed candidate satisfies a full-tree best-response bound no larger than a chosen budget, so conditional expected value against any opponent—including adaptive ones—is at least the Nash floor minus that budget, while incomplete or wrong models can only reduce captured gain, never break the bound.

Load-bearing premise

The safety proof needs exact full-tree best-response certificates and restricted solves to be available online, which the paper restricts to solver-tractable games.

Editorial extensions

If this is right

  • Safety can be relocated from assumptions about the opponent model to a check on the played strategy, so model error costs gain rather than the loss bound.
  • A single budget dial interpolates continuously between pure Nash play and an unrestricted schedule while keeping every committed strategy inside the bound.
  • Trajectory mixtures that share the same estimator can match gain yet massively overshoot the same budget, so the behavioural restricted solve plus certificate is doing essential work.
  • Confirmation starvation—not certificate failure—is the binding limit on diffuse or rarely reached leaks, and longer horizons raise the number of confirmed opponents without breaking the budget.
  • In solver-tractable games the certificate pass is cheap enough to re-run at every checkpoint, making atomic certify-then-commit practical online.

Reading between the lines

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

  • Any poker-scale extension will stand or fall on whether a sound approximate or depth-limited certificate can replace the exact full-tree best response while preserving the runtime invariant.
  • The same certify-the-played-strategy pattern could apply to other online adaptation settings where a planner proposes risky policies from partial models.
  • Pool design and observation mechanisms become first-class engineering choices: better pools shrink confirmation starvation without touching the safety argument.
  • Independently trained near-equilibrium opponents already show the ordering the paper predicts; larger unscripted populations would test whether the gain gap survives when leaks are not hand-designed.
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

2 major / 5 minor

Summary. The paper proposes CS-RNR, an online opponent-exploitation method for two-player zero-sum imperfect-information games. Anytime-valid confidence sequences flag pooled action frequencies that separate from an equilibrium reference; confirmed excesses define a conservative opponent model; restricted-response solves produce candidate strategies over a pin grid; and each complete candidate is certified by an original-game full-tree best response before atomic commit under a user budget ε_max. Proposition 1 states a runtime invariant: if the executor plays the certified behavioural strategy without post-certification modification and only admits B_t ≤ ε_max, then conditional expected hero value is at least v*_hero − ε_max against any opponent, including adaptive ones. Detection and modeling affect captured gain only. Empirically, on Leduc CS-RNR obtains 6.2× the steady-state gain of a money-verified binary gate with worst certificate ≤ 0.15; Fixed-Mix with the same estimator overshoots the budget by large factors; and all 36,000 audited hands across Leduc, Liar’s Dice, and 5-rank Leduc satisfy the reported tolerance. Confirmation starvation is identified as a distinct data-dependent limit.

Significance. If the result holds as stated, the paper cleanly relocates safety from response construction to a certificate on the played strategy, separating model error (which can cost gain) from deployment risk (which cannot). That separation is conceptually useful and is supported by a nearly definitional runtime invariant, shared-estimator ablations that isolate the schedule, adversarial audits including every-hand best responders, pre-registered concentrated and independently trained opponent suites, and exact full-tree evaluation of every reported gain. The work is scoped to solver-tractable games where exact certificates are available—an honest and load-bearing limit the authors state in §2, §4.3, §7, and Future Work—but within that scope the contribution is solid, reproducible in design, and of clear interest to the safe-exploitation and extensive-form game-solving communities.

major comments (2)
  1. [§5, Finite-reference correction; Appendix A] §5 (Finite-reference correction) and Appendix A: the implementation certifies against a finite-iteration reference ẽv_opp, so the true-game guarantee is ε_max + η_v. The audit tolerance 5×10^{-3} covers CFR+ residue at 400 iterations, but the manuscript does not report a measured bound on |ẽv_opp − v*_opp| for the 4000-iteration Leduc reference (or the analogous references in the other games). A short quantification of η_v—or an explicit statement that the reported certificates already absorb the reference gap under the audit tolerance—would make the empirical claim “every deployed strategy within budget” line up tightly with Proposition 1’s exact-value statement.
  2. [§4.1 Detection; Assumption 1; §6.4–6.6] §4.1–4.2 and Assumption 1: Proposition 1 is independent of detection, but the gain claims (6.2× gate, cross-game tables) depend on the hand-chosen 12-pool taxonomy, margin δ, and the within-pool homogeneity idealization used to inherit excesses to infosets. The horizon sweeps and 5-rank concentrated pre-registration already show confirmation starvation; it would strengthen the paper to state more explicitly that safe gain is jointly limited by (pool design, δ, T) and that alternative poolings were not systematically ablated. This does not undermine the safety invariant, but it bounds how far the headline multipliers should be read as method-intrinsic rather than taxonomy-dependent.
minor comments (5)
  1. [Figure 2; §6.1] Figure 2 caption and §6.1: the offline frontier is central to motivating the schedule; stating the exact oracle-model protocol (same CFR+ iteration budget as online restricted solves?) in the caption would aid reproduction.
  2. [Table 1; §6.4; Appendix A] Table 1 vs. §6.4: the main table uses stitched sub-Gaussian boundaries (2/12 diffuse releases at T=800) while the horizon sweep uses empirical-Bernstein (1/12 at T=800). The appendix notes this; a one-sentence pointer in the main comparison would prevent readers from treating the two release counts as the same detector.
  3. [Algorithm 3; §4.3] Algorithm 3 and §4.3: the climb cap c=2 and checkpoint list are fixed without sensitivity analysis beyond the budget/pin sweeps in Table 2. A brief remark that the invariant does not require monotonicity in p or a globally maximal feasible pin (already implied by per-candidate certification) would clarify why the local search is sufficient.
  4. [§2 Related Work] Related Work §2: the distinction from safe nested subgame solving (Brown & Sandholm 2017) and opponent-limited online search is present but compressed. One additional sentence stating that those methods bound re-solve loss relative to a blueprint, whereas CS-RNR certifies a whole-game deployed policy between hands, would sharpen the novelty claim.
  5. [Abstract; §1] Minor typography: “diffusedeviations”, “budget-constrained confidence-scheduled”, and similar missing spaces appear in the abstract/introduction PDF text; clean these in production.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: safety is a definitional consequence of full-tree BR certificates on the played strategy, and gains are external exact evaluations.

full rationale

The load-bearing safety claim (Proposition 1) follows directly from the definition of exploitability and zero-sum payoffs once the runtime only commits complete behavioural strategies whose original-game full-tree best-response certificate satisfies B_t ≤ ε_max and plays them without post-certification modification. That is ordinary definition→theorem structure, not a self-definitional loop in which the certificate is fitted from or defined via the claimed bound. Detection, pools, δ, and the opponent model affect only which candidates are proposed (hence captured gain); the paper explicitly separates them from the deployment invariant and notes that model error can cost gain, never the certificate. Empirical gains are exact profile values against held-out scripted and independently trained opponents, not re-reports of a fitting objective. Restricted-response pins and ε_max are user control knobs. Self-citations (authors’ prior solver/abstraction work) are background implementation context, not uniqueness theorems or ansatze that force the central claim. No fitted-input-as-prediction, no renaming of a known pattern as a derived law, and no circular reduction of Eq. X to Eq. Y by construction beyond the honest definitional content of exploitability.

Assumptions & free parameters 6 free parameters · 5 assumptions · 2 invented entities

Safety reduces to playing only strategies whose measured original-game BR gap is ≤ ε_max; that rests on standard zero-sum extensive-form structure plus the operational ability to compute exact BR and restricted equilibria. Detection and modeling add statistical and pooling assumptions that affect gain, not Prop. 1. Several numerical knobs (budget, margin, grids, pools) are chosen by the authors and shape what gets confirmed and deployed.

free parameters (6)
  • ε_max (exploitability budget) = 0.15 (main); 0.05 (Liar’s Dice)
    User/author-chosen deployment threshold (0.15 Leduc/5-rank, 0.05 Liar’s Dice). Directly defines the safety claim’s numerical content and the gain–safety operating point.
  • deviation margin δ = 0.10 / 0.05
    Hand-set separation margin before a cell is flagged (0.10 main; 0.05 long-horizon EB). Controls false flags vs confirmation starvation.
  • episode-wide α and Bonferroni split = α=0.05, m=30 (Leduc)
    α=0.05 split over m scanned cells; standard but chosen, and m depends on pool design.
  • pin grid, climb cap c=2, checkpoint schedule = p∈{0,0.1,0.2,0.3,0.5,0.7,0.9}, c=2, T-dependent checkpoints
    Discrete p levels and bounded local search determine which certified strategies can appear; not derived from theory.
  • pool taxonomy = 12 pools / 30 cells (Leduc)
    12 Leduc pools (and analogues) are a modeling choice that defines what frequencies are monitored and what can ever be confirmed.
  • CFR+ iteration budgets (reference, restricted solve) = 4000 / 400 iterations
    4000-iter reference and 400-iter restricted solves set residual error absorbed into η_v and audit tolerance 5e-3.
assumptions (5)
  • domain assumption Two-player zero-sum extensive-form game with perfect recall; exploitability equals max opponent gain over game value.
    Preliminaries §3; underpins Prop. 1 payoff identity u_hero ≥ v*_hero − B_t.
  • standard math Time-uniform confidence sequences (Howard et al. 2021) give simultaneous coverage under Bonferroni split for optional stopping/checking.
    Used in §4.1 and Prop. 2 for detection validity, not for the deployment invariant.
  • domain assumption Restricted-response (RNR/DBR) pin mixtures interpolate between Nash and model BR and can be solved by regret minimization.
    §3–4.2 citing Johanson et al.; response primitive, certified after the fact.
  • ad hoc to paper Assumption 1: monitored increments have conditional mean equal to reach-weighted pool frequency under fixed hero strategy; δ absorbs attribution and reach-mixture drift.
    §5; needed for Prop. 2 interpretation; Prop. 1 independent of it.
  • domain assumption Exact full-tree best response of any candidate behavioural strategy is computable at deploy time (solver-tractable games).
    §4.3, §7; load-bearing for the certificate as proved.
invented entities (2)
  • CS-RNR deployment loop (detect → conservative model → restricted solve grid → BR certificate → atomic commit) independent evidence
    purpose: Operational method that schedules restricted responses under a played-strategy loss budget.
    Composite of known pieces; the entity is the scheduled certified loop, not a new physical object.
  • Reference-relative certificate B̃(σ_H) evaluated on complete candidates before play independent evidence
    purpose: Make safety a measured property of the deployed strategy rather than of the model construction.
    Definitionally a BR gap; empirical content is that scheduling still yields large gain while respecting it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Agents That Certify Their Own Exploits: Confidence-Scheduled Restricted Responses for Safe Opponent Exploitation." pith.science (2026). https://pith.science/paper/O53STODN

@misc{pith2026260728520,
  author       = {Pith},
  title        = {Pith review of: Agents That Certify Their Own Exploits: Confidence-Scheduled Restricted Responses for Safe Opponent Exploitation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/O53STODN}},
  note         = {Machine review of arXiv:2607.28520}
}
abstract

An agent playing a Nash-equilibrium strategy in a two-player zero-sum imperfect-information game secures the game value but forfeits the additional value offered by a flawed opponent. Diffuse deviations pose a particular challenge: binary release rules may gather too little evidence to act, while a full best response to an incomplete opponent model can be highly exploitable. We introduce \emph{budget-constrained confidence-scheduled restricted responses} (CS-RNR), the first opponent-exploitation method whose safety guarantee is a certificate the agent computes on the strategy it actually deploys, so that every exploit it commits to is one it has audited itself. The method tracks pooled action frequencies with anytime-valid confidence sequences and treats a frequency as exploitable only once its interval separates from an equilibrium reference. The confirmed deviations define a conservative opponent model, which a restricted-response solve turns into candidate counter-strategies over a grid of pin levels. Before deployment, each complete candidate is evaluated by a full-tree best response. The resulting certificate is compared with a user-specified budget and committed atomically with the strategy. Because this check is performed on the played strategy, model quality determines the exploitation achieved while the certificate controls reference-relative expected loss. In Leduc hold'em, CS-RNR obtains $6.2\times$ the steady-state gain of a money-verified binary gate while keeping every deployed strategy within budget. A trajectory mixture using the same estimator reaches $13.6\times$ the budget. Across Leduc, Liar's Dice, and 5-rank Leduc, all $36{,}000$ audited hands satisfy the reported certificate tolerance.

Figures

Figures reproduced from arXiv: 2607.28520 by the authors.

Figure 1
Figure 1. The CS-RNR deployment loop. Online observations update anytime-valid confidence [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Restricted-response frontier with an oracle model ( [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Per-opponent schedule adaptation (median over [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Adversarial audit (T=800, εmax=0.15; hardest of 5 seeds). Solid: per-hand expected gain; dashed: −Bt . (a) Against the every-hand best responder the curves coincide. (b) Bait-and-punish costs −0.009/hand. Dash-dotted: measured mean loss with the same detector but no bu…
Figure 5
Figure 5. Figure 5: Diffuse suite vs. horizon under empirical-Bernstein boundaries ( [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

100 extracted references · 9 linked inside Pith

  1. [1]

    Bowling , title =

    Michael Johanson and Martin Zinkevich and Michael H. Bowling , title =

  2. [2]

    Artificial Intelligence and Statistics , pages=

    Data biased robust counter strategies , author=. Artificial Intelligence and Statistics , pages=. 2009 , organization=

    2009

  3. [3]

    ACM Transactions on Economics and Computation (TEAC) , volume=

    Safe opponent exploitation , author=. ACM Transactions on Economics and Computation (TEAC) , volume=. 2015 , publisher=

    2015

  4. [4]

    Advances in neural information processing systems , volume=

    Regret minimization in games with incomplete information , author=. Advances in neural information processing systems , volume=

  5. [6]

    Science , volume=

    Heads-up limit hold'em poker is solved , author=. Science , volume=. 2015 , publisher=

    2015

  6. [7]

    Advances in Neural Information Processing Systems , volume=

    Safe opponent-exploitation subgame refinement , author=. Advances in Neural Information Processing Systems , volume=

  7. [8]

    International Conference on Machine Learning , pages=

    Greedy when sure and conservative when uncertain about the opponents , author=. International Conference on Machine Learning , pages=. 2022 , organization=

    2022

  8. [9]

    Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems , pages=

    Adapting Beyond the Depth Limit: Counter Strategies in Large Imperfect Information Games , author=. Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems , pages=

Show all 100 references
  1. [10]

    Forty-first International Conference on Machine Learning , year=

    Safe and robust subgame exploitation in imperfect information games , author=. Forty-first International Conference on Machine Learning , year=

  2. [11]

    Science , volume=

    Deepstack: Expert-level artificial intelligence in heads-up no-limit poker , author=. Science , volume=. 2017 , publisher=

    2017

  3. [12]

    Science , volume=

    Superhuman AI for heads-up no-limit poker: Libratus beats top professionals , author=. Science , volume=. 2018 , publisher=

    2018

  4. [13]

    Science , volume=

    Superhuman AI for multiplayer poker , author=. Science , volume=. 2019 , publisher=

    2019

  5. [14]

    Advances in neural information processing systems , volume=

    Safe and nested subgame solving for imperfect-information games , author=. Advances in neural information processing systems , volume=

  6. [15]

    Advances in neural information processing systems , volume=

    Depth-limited solving for imperfect-information games , author=. Advances in neural information processing systems , volume=

  7. [16]

    Advances in neural information processing systems , volume=

    Regret-based pruning in extensive-form games , author=. Advances in neural information processing systems , volume=

  8. [17]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Dynamic thresholding and pruning for regret minimization , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  9. [18]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Solving imperfect-information games via discounted regret minimization , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  10. [19]

    International conference on machine learning , pages=

    Deep counterfactual regret minimization , author=. International conference on machine learning , pages=. 2019 , organization=

    2019

  11. [20]

    Advances in neural information processing systems , volume=

    Combining deep reinforcement learning and search for imperfect-information games , author=. Advances in neural information processing systems , volume=

  12. [21]

    Proceedings of the Twenty-First Conference on Uncertainty in Artificial Intelligence , pages=

    Bayes' bluff: opponent modelling in poker , author=. Proceedings of the Twenty-First Conference on Uncertainty in Artificial Intelligence , pages=

  13. [22]

    The Annals of Statistics , volume=

    Time-uniform, nonparametric, nonasymptotic confidence sequences , author=. The Annals of Statistics , volume=. 2021 , publisher=

    2021

  14. [23]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Proposer of the vote of thanks to Waudy-Smith and Ramdas and contribution to the Discussion of `Estimating means of bounded random variables by betting' , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2024 , publisher=

    2024

  15. [26]

    CoRR , volume =

    Sam Ganzfried , title =. CoRR , volume =

  16. [27]

    The International FLAIRS Conference Proceedings , volume=

    Nonparametric Strategy Test , author=. The International FLAIRS Conference Proceedings , volume=

  17. [28]

    AAAI Technical Report (2) , pages=

    Safe Strategies for Agent Modelling in Games , author=. AAAI Technical Report (2) , pages=

  18. [29]

    AAAI , volume=

    Effective short-term opponent exploitation in simplified poker , author=. AAAI , volume=

  19. [30]

    Proceedings of the 3rd AAAI Conference on Interactive Decision Theory and Game Theory , pages=

    MCRNR: fast computing of restricted Nash responses by means of sampling , author=. Proceedings of the 3rd AAAI Conference on Interactive Decision Theory and Game Theory , pages=

  20. [31]

    Proceedings of the 41st International Conference on Machine Learning , pages=

    RL-CFR: improving action abstraction for imperfect information extensive-form games with reinforcement learning , author=. Proceedings of the 41st International Conference on Machine Learning , pages=

  21. [32]

    The Thirteenth International Conference on Learning Representations , year=

    Efficient online pruning and abstraction for imperfect information extensive-form games , author=. The Thirteenth International Conference on Learning Representations , year=

  22. [36]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Faster game solving via predictive blackwell approachability: Connecting regret matching and mirror descent , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  23. [37]

    Advances in Neural Information Processing Systems , volume=

    Exploiting opponents under utility constraints in sequential games , author=. Advances in Neural Information Processing Systems , volume=

  24. [38]

    IEEE Transactions on Cybernetics , volume=

    Modeling rationality: Toward better performance against unknown agents in sequential games , author=. IEEE Transactions on Cybernetics , volume=. 2022 , publisher=

    2022

  25. [39]

    Journal of Artificial Intelligence Research , volume=

    A survey of opponent modeling in adversarial domains , author=. Journal of Artificial Intelligence Research , volume=

  26. [40]

    Advances in Neural Information Processing Systems , volume=

    Regret matching+:(in) stability and fast convergence in games , author=. Advances in Neural Information Processing Systems , volume=

  27. [41]

    Advances in Neural Information Processing Systems , volume=

    Efficient subgame refinement for extensive-form games , author=. Advances in Neural Information Processing Systems , volume=

  28. [42]

    The Twelfth International Conference on Learning Representations , year=

    Towards offline opponent modeling with in-context learning , author=. The Twelfth International Conference on Learning Representations , year=

  29. [43]

    International Conference on Machine Learning , pages=

    Opponent-limited online search for imperfect information games , author=. International Conference on Machine Learning , pages=. 2023 , organization=

    2023

  30. [44]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Learning not to regret , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  31. [45]

    The Twelfth International Conference on Learning Representations , year=

    Dynamic discounted counterfactual regret minimization , author=. The Twelfth International Conference on Learning Representations , year=

  32. [46]

    Advances in Neural Information Processing Systems , volume=

    Opponent modeling with in-context search , author=. Advances in Neural Information Processing Systems , volume=

  33. [47]

    Expert Systems with Applications , volume=

    HORSE-CFR: Hierarchical opponent reasoning for safe exploitation counterfactual regret minimization , author=. Expert Systems with Applications , volume=. 2025 , publisher=

    2025

  34. [48]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    An open-ended learning framework for opponent modeling , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  35. [49]

    The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

    Efficient Last-Iterate Convergence in Solving Extensive-Form Games , author=. The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

  36. [50]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Faster Game Solving via Asymmetry of Step Sizes , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  37. [51]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Deep (Predictive) Discounted Counterfactual Regret Minimization , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  38. [54]

    2026 , eprint=

    Correlated Chance Sampling for Monte Carlo Counterfactual Regret Minimization , author=. 2026 , eprint=

    2026

  39. [55]

    Exploiting opponents under utility constraints in sequential games

    Martino Bernasconi-de Luca, Federico Cacciamani, Simone Fioravanti, Nicola Gatti, Alberto Marchesi, and Francesco Trov \`o . Exploiting opponents under utility constraints in sequential games. Advances in Neural Information Processing Systems, 34: 0 13177--13188, 2021

    2021

  40. [56]

    Heads-up limit hold'em poker is solved

    Michael Bowling, Neil Burch, Michael Johanson, and Oskari Tammelin. Heads-up limit hold'em poker is solved. Science, 347 0 (6218): 0 145--149, 2015

    2015

  41. [57]

    Regret-based pruning in extensive-form games

    Noam Brown and Tuomas Sandholm. Regret-based pruning in extensive-form games. Advances in neural information processing systems, 28, 2015

    2015

  42. [58]

    Safe and nested subgame solving for imperfect-information games

    Noam Brown and Tuomas Sandholm. Safe and nested subgame solving for imperfect-information games. Advances in neural information processing systems, 30, 2017

    2017

  43. [59]

    Superhuman ai for heads-up no-limit poker: Libratus beats top professionals

    Noam Brown and Tuomas Sandholm. Superhuman ai for heads-up no-limit poker: Libratus beats top professionals. Science, 359 0 (6374): 0 418--424, 2018

    2018

  44. [60]

    Solving imperfect-information games via discounted regret minimization

    Noam Brown and Tuomas Sandholm. Solving imperfect-information games via discounted regret minimization. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33, pages 1829--1836, 2019 a

    2019

  45. [61]

    Superhuman ai for multiplayer poker

    Noam Brown and Tuomas Sandholm. Superhuman ai for multiplayer poker. Science, 365 0 (6456): 0 885--890, 2019 b

    2019

  46. [62]

    Dynamic thresholding and pruning for regret minimization

    Noam Brown, Christian Kroer, and Tuomas Sandholm. Dynamic thresholding and pruning for regret minimization. In Proceedings of the AAAI conference on artificial intelligence, volume 31, 2017

    2017

  47. [63]

    Depth-limited solving for imperfect-information games

    Noam Brown, Tuomas Sandholm, and Brandon Amos. Depth-limited solving for imperfect-information games. Advances in neural information processing systems, 31, 2018

    2018

  48. [64]

    Deep counterfactual regret minimization

    Noam Brown, Adam Lerer, Sam Gross, and Tuomas Sandholm. Deep counterfactual regret minimization. In International conference on machine learning, pages 793--802. PMLR, 2019

    2019

  49. [65]

    Combining deep reinforcement learning and search for imperfect-information games

    Noam Brown, Anton Bakhtin, Adam Lerer, and Qucheng Gong. Combining deep reinforcement learning and search for imperfect-information games. Advances in neural information processing systems, 33: 0 17057--17069, 2020

    2020

  50. [66]

    Stratformer: Adaptive opponent modeling and exploitation in imperfect-information games

    Andy Caen, Mark HM Winands, and Dennis JNJ Soemers. Stratformer: Adaptive opponent modeling and exploitation in imperfect-information games. arXiv preprint arXiv:2604.25796, 2026

    2026 arXiv

  51. [67]

    Test-then-punish: A statistical approach to repeated games

    Aymeric Capitaine, Antoine Scheid, Etienne Boursier, Alain Durmus, and Michael I Jordan. Test-then-punish: A statistical approach to repeated games. arXiv preprint arXiv:2603.05619, 2026

    2026

  52. [68]

    Faster game solving via predictive blackwell approachability: Connecting regret matching and mirror descent

    Gabriele Farina, Christian Kroer, and Tuomas Sandholm. Faster game solving via predictive blackwell approachability: Connecting regret matching and mirror descent. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 5363--5371, 2021

    2021

  53. [69]

    Regret matching+:(in) stability and fast convergence in games

    Gabriele Farina, Julien Grand-Cl \'e ment, Christian Kroer, Chung-Wei Lee, and Haipeng Luo. Regret matching+:(in) stability and fast convergence in games. Advances in Neural Information Processing Systems, 36: 0 61546--61572, 2023

    2023

  54. [70]

    Greedy when sure and conservative when uncertain about the opponents

    Haobo Fu, Ye Tian, Hongxiang Yu, Weiming Liu, Shuang Wu, Jiechao Xiong, Ying Wen, Kai Li, Junliang Xing, Qiang Fu, et al. Greedy when sure and conservative when uncertain about the opponents. In International Conference on Machine Learning, pages 6829--6848. PMLR, 2022

    2022

  55. [71]

    Consistent opponent modeling of static opponents in imperfect-information games

    Sam Ganzfried. Consistent opponent modeling of static opponents in imperfect-information games. CoRR, abs/2508.17671, 2025 a

    2025 arXiv

  56. [72]

    Nonparametric strategy test

    Sam Ganzfried. Nonparametric strategy test. In The International FLAIRS Conference Proceedings, volume 38, 2025 b

    2025

  57. [73]

    Safe opponent exploitation

    Sam Ganzfried and Tuomas Sandholm. Safe opponent exploitation. ACM Transactions on Economics and Computation (TEAC), 3 0 (2): 0 1--28, 2015

    2015

  58. [74]

    Betting on equilibrium: Monitoring strategic behavior in multi-agent systems

    Etienne Gauthier, Francis Bach, and Michael I Jordan. Betting on equilibrium: Monitoring strategic behavior in multi-agent systems. arXiv preprint arXiv:2601.05427, 2026

    2026 arXiv

  59. [75]

    Modeling rationality: Toward better performance against unknown agents in sequential games

    Zhenxing Ge, Shangdong Yang, Pinzhuo Tian, Zixuan Chen, and Yang Gao. Modeling rationality: Toward better performance against unknown agents in sequential games. IEEE Transactions on Cybernetics, 54 0 (5): 0 2966--2977, 2022

    2022

  60. [76]

    Efficient subgame refinement for extensive-form games

    Zhenxing Ge, Zheng Xu, Tianyu Ding, Wenbin Li, and Yang Gao. Efficient subgame refinement for extensive-form games. Advances in Neural Information Processing Systems, 36: 0 8280--8291, 2023

    2023

  61. [77]

    Safe and robust subgame exploitation in imperfect information games

    Zhenxing Ge, Zheng Xu, Tianyu Ding, Linjian Meng, Bo An, Wenbin Li, and Yang Gao. Safe and robust subgame exploitation in imperfect information games. In Forty-first International Conference on Machine Learning, 2024

    2024

  62. [78]

    Proposer of the vote of thanks to waudy-smith and ramdas and contribution to the discussion of `estimating means of bounded random variables by betting'

    Peter Gr \"u nwald. Proposer of the vote of thanks to waudy-smith and ramdas and contribution to the discussion of `estimating means of bounded random variables by betting'. Journal of the Royal Statistical Society Series B: Statistical Methodology, 86 0 (1): 0 28--30, 2024

    2024

  63. [79]

    Effective short-term opponent exploitation in simplified poker

    Bret Hoehn, Finnegan Southey, Robert C Holte, and Valeriy Bulitko. Effective short-term opponent exploitation in simplified poker. In AAAI, volume 5, pages 783--788, 2005

    2005

  64. [80]

    Time-uniform, nonparametric, nonasymptotic confidence sequences

    Steven R Howard, Aaditya Ramdas, Jon McAuliffe, and Jasjeet Sekhon. Time-uniform, nonparametric, nonasymptotic confidence sequences. The Annals of Statistics, 49 0 (2): 0 1055--1080, 2021

    2021

  65. [81]

    Towards offline opponent modeling with in-context learning

    Yuheng Jing, Kai Li, Bingyun Liu, Yifan Zang, Haobo Fu, Qiang Fu, Junliang Xing, and Jian Cheng. Towards offline opponent modeling with in-context learning. In The Twelfth International Conference on Learning Representations, 2023

    2023

  66. [82]

    Opponent modeling with in-context search

    Yuheng Jing, Bingyun Liu, Kai Li, Yifan Zang, Haobo Fu, Qiang Fu, Junliang Xing, and Jian Cheng. Opponent modeling with in-context search. Advances in Neural Information Processing Systems, 37: 0 61549--61591, 2024

    2024

  67. [83]

    An open-ended learning framework for opponent modeling

    Yuheng Jing, Kai Li, Bingyun Liu, Haobo Fu, Qiang Fu, Junliang Xing, and Jian Cheng. An open-ended learning framework for opponent modeling. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 23222--23230, 2025

    2025

  68. [84]

    Data biased robust counter strategies

    Michael Johanson and Michael Bowling. Data biased robust counter strategies. In Artificial Intelligence and Statistics, pages 264--271. PMLR, 2009

    2009

  69. [85]

    Michael Johanson, Martin Zinkevich, and Michael H. Bowling. Computing robust counter-strategies. In NIPS , pages 721--728. Curran Associates, Inc., 2007

    2007

  70. [86]

    Efficient online pruning and abstraction for imperfect information extensive-form games

    Boning Li and Longbo Huang. Efficient online pruning and abstraction for imperfect information extensive-form games. In The Thirteenth International Conference on Learning Representations, 2025

    2025

  71. [87]

    Effective, efficient, and general information abstraction for imperfect-information extensive-form games

    Boning Li and Longbo Huang. Effective, efficient, and general information abstraction for imperfect-information extensive-form games. arXiv preprint arXiv:2605.10900, 2026 a

    2026 arXiv

  72. [88]

    Real-time parallel counterfactual regret minimization

    Boning Li and Longbo Huang. Real-time parallel counterfactual regret minimization. arXiv preprint arXiv:2605.19928, 2026 b

    2026 arXiv

  73. [89]

    Rl-cfr: improving action abstraction for imperfect information extensive-form games with reinforcement learning

    Boning Li, Zhixuan Fang, and Longbo Huang. Rl-cfr: improving action abstraction for imperfect information extensive-form games with reinforcement learning. In Proceedings of the 41st International Conference on Machine Learning, pages 27752--27770, 2024

    2024

  74. [90]

    Correlated chance sampling for monte carlo counterfactual regret minimization, 2026 a

    Boning Li, Yu Chen, and Longbo Huang. Correlated chance sampling for monte carlo counterfactual regret minimization, 2026 a . URL https://arxiv.org/abs/2607.27035

    2026 arXiv

  75. [91]

    Pokerskill: Llms can play expert-level poker without training or solvers

    Boning Li, Baoxiang Wang, and Longbo Huang. Pokerskill: Llms can play expert-level poker without training or solvers. arXiv preprint arXiv:2605.30094, 2026 b

    2026 arXiv

  76. [92]

    Safe opponent-exploitation subgame refinement

    Mingyang Liu, Chengjie Wu, Qihan Liu, Yansen Jing, Jun Yang, Pingzhong Tang, and Chongjie Zhang. Safe opponent-exploitation subgame refinement. Advances in Neural Information Processing Systems, 35: 0 27610--27622, 2022

    2022

  77. [93]

    Opponent-limited online search for imperfect information games

    Weiming Liu, Haobo Fu, Qiang Fu, and Yang Wei. Opponent-limited online search for imperfect information games. In International Conference on Machine Learning, pages 21567--21585. PMLR, 2023

    2023

  78. [94]

    Safe strategies for agent modelling in games

    Peter McCracken and Michael Bowling. Safe strategies for agent modelling in games. In AAAI Technical Report (2), pages 103--110, 2004

    2004

  79. [95]

    Efficient last-iterate convergence in solving extensive-form games

    Linjian Meng, Tianpei Yang, Youzhi Zhang, Zhenxing Ge, Shangdong Yang, Tianyu Ding, Wenbin Li, Bo An, and Yang Gao. Efficient last-iterate convergence in solving extensive-form games. In The Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025. URL htt...

    2025

  80. [96]

    Faster game solving via asymmetry of step sizes

    Linjian Meng, Tianpei Yang, Youzhi Zhang, Zhenxing Ge, and Yang Gao. Faster game solving via asymmetry of step sizes. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 17161--17169, 2026

    2026

  81. [97]

    Adapting beyond the depth limit: Counter strategies in large imperfect information games

    David Milec, Vojt e ch Kova r \' k, and Viliam Lis \`y . Adapting beyond the depth limit: Counter strategies in large imperfect information games. In Proceedings of the 24th International Conference on Autonomous Agents and Multiagent Systems, pages 2675--2677, 2025

    2025

  82. [98]

    Deepstack: Expert-level artificial intelligence in heads-up no-limit poker

    Matej Morav c \' k, Martin Schmid, Neil Burch, Viliam Lis \`y , Dustin Morrill, Nolan Bard, Trevor Davis, Kevin Waugh, Michael Johanson, and Michael Bowling. Deepstack: Expert-level artificial intelligence in heads-up no-limit poker. Science, 356 0 (6337): 0 508--513, 2017

    2017

  83. [99]

    Alphaexploitem: Going beyond the nash equilibrium in poker by learning to exploit suboptimal play

    Vlad Murgoci, Matthijs Spaan, and Yaniv Oren. Alphaexploitem: Going beyond the nash equilibrium in poker by learning to exploit suboptimal play. arXiv preprint arXiv:2605.09150, 2026

    2026 arXiv

  84. [100]

    A survey of opponent modeling in adversarial domains

    Samer Nashed and Shlomo Zilberstein. A survey of opponent modeling in adversarial domains. Journal of Artificial Intelligence Research, 73: 0 277--327, 2022

    2022

  85. [101]

    Mcrnr: fast computing of restricted nash responses by means of sampling

    Marc Ponsen, Marc Lanctot, and Steven De Jong. Mcrnr: fast computing of restricted nash responses by means of sampling. In Proceedings of the 3rd AAAI Conference on Interactive Decision Theory and Game Theory, pages 43--49, 2010

    2010

  86. [102]

    Bayes' bluff: opponent modelling in poker

    Finnegan Southey, Michael Bowling, Bryce Larson, Carmelo Piccione, Neil Burch, Darse Billings, and Chris Rayner. Bayes' bluff: opponent modelling in poker. In Proceedings of the Twenty-First Conference on Uncertainty in Artificial Intelligence, pages 550--558, 2005

    2005

  87. [103]

    Learning not to regret

    David Sychrovsk \`y , Michal S ustr, Elnaz Davoodi, Michael Bowling, Marc Lanctot, and Martin Schmid. Learning not to regret. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 15202--15210, 2024

    2024

  88. [104]

    Solving large imperfect information games using cfr+

    Oskari Tammelin. Solving large imperfect information games using cfr+. arXiv preprint arXiv:1407.5042, 2014

    2014 arXiv

  89. [105]

    Horse-cfr: Hierarchical opponent reasoning for safe exploitation counterfactual regret minimization

    Shijia Wang, Jiao Wang, and Bangyan Song. Horse-cfr: Hierarchical opponent reasoning for safe exploitation counterfactual regret minimization. Expert Systems with Applications, 263: 0 125697, 2025

    2025

  90. [106]

    Dynamic discounted counterfactual regret minimization

    Hang Xu, Kai Li, Haobo Fu, Qiang Fu, Junliang Xing, and Jian Cheng. Dynamic discounted counterfactual regret minimization. In The Twelfth International Conference on Learning Representations, 2024

    2024

  91. [107]

    Deep (predictive) discounted counterfactual regret minimization

    Hang Xu, Kai Li, Haobo Fu, Qiang Fu, Junliang Xing, and Jian Cheng. Deep (predictive) discounted counterfactual regret minimization. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 17284--17292, 2026

    2026

  92. [108]

    Regret minimization in games with incomplete information

    Martin Zinkevich, Michael Johanson, Michael Bowling, and Carmelo Piccione. Regret minimization in games with incomplete information. Advances in neural information processing systems, 20, 2007

    2007

Pith tools

Reviewed July 31, 2026 · model on record in the stance chip above.