Pith. sign in

REVIEW 3 major objections 6 minor 52 references

Attribution Markets: A Fisher-Market Formulation for Fractional Credit Assignment Between Planned Tasks and Performed Actions

T0 review · 3 major / 6 minor · reviewed 2026-08-01 · deepseek-v4-flash

Pith's one-line read This paper claims the gap between planned tasks and logged actions is best bridged by a Fisher market — budgeted buyers (tasks) purchasing divisible goods (actions) — so that conservation, budget caps, and honest abstention become proven th

desk verdict A genuinely new and unusually honest Fisher-market application, but the central D2 guarantee is a theorem about b_i/ρ, not b_i—with ρ=0.25 the 'hard budget cap' permits 4× overshoot. read the letter →

arxiv 2607.20694 v1 pith:RGUUIIM5 submitted 2026-07-22 cs.LG cs.GT

classification cs.LGcs.GT MSC 91B5091B32
keywords Fishermarketfractionalcreditassignmenttaskattributionproportionalresponsedynamicsentropicregularizationoptimaltransportsatiationfixedpointbudgetconstraints
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 is trying to establish that the drift between what people plan and what they actually log — the gap that makes progress trackers report false stalls — is best bridged by a market, not a link. Concretely, it models each planned task as a budget-constrained buyer and each performed action as a divisible good, with a fused similarity signal as valuation, and claims that two instruments (a seller reserve price and a buyer cash option) convert three informal requirements — conservation, a hard budget cap, and honest abstention from weak matches — into proven theorems. A sympathetic reader would care because the all-or-nothing links used in today's systems strand genuinely related effort and fabricate stalls on active goals; if the market formulation is right, a progress report can carry hard guarantees and an auditable price explanation for every share. The paper also claims its own completion-seeking extension — progress worth less as a task nears its plan — breaks the standard convergence proof and is repaired by a satiation-threshold fixed point, and it surfaces and resolves a real weakness: the market's sharp equilibrium is noisier than entropy-smoothed optimal transport, unified by a one-parameter entropy dial.

What carries the argument

The central object is the attribution market: a quasi-linear Fisher market — the classical model where budgeted buyers spend on divisible goods until prices clear — in which each planned task i buys shares of each performed action j with budget bᵢ, and a fused text/structural/temporal affinity qᵢⱼ sets the valuation. The theorem-carrying instruments are the seller reserve rate ρ (the float's standing bid bounds every price below) and the cash utility u₀ (the outside option that sets the junk-filter threshold). Clearing is computed by proportional response dynamics, a multiplicative update that converges to the classical Fisher-market competitive equilibrium. The completion extension's load-b

What would settle it

Take one action shared by two near-equally-matched tasks, add independent mean-zero noise to the affinity estimates over hundreds of draws, and compare the total-variation shift in the market's equilibrium shares against the shift of an entropy-regularized transport solution under the same perturbations: if the market's shares are not systematically more sensitive, the paper's central noise-asymmetry claim (Section 6.4) fails to replicate.

Watch

Extended reading notes

Core claim

Attribution — dividing each logged action's hours among the planned tasks it served — is claimed to be a market-clearing problem, and two instruments turn informal wishes into theorems. A standing reserve bid ρdⱼ on every action from a 'float' account, plus letting tasks hold cash at utility u₀, yield conservation (each hour credited exactly once), the hard budget cap Pᵢ ≤ bᵢ/ρ, and the junk filter (affinity below u₀ρ gets zero share). The paper further claims a concave completion utility — diminishing returns near a task's plan — breaks the standard convergence proof, repaired by a satiation-threshold fixed point: an outer loop over satiation multipliers wrapping an unmodified inner market

Load-bearing premise

The completion market's convergence guarantee rests on a diagonal-dominance condition (Proposition 7) — each task's satiation must respond more to its own progress than to other tasks' — which the paper does not prove to hold in general and which Section 6.6 shows failing in one of ten adversarial near-duplicate instances.

Editorial extensions

If this is right

  • Any report produced by the market carries an exact, anytime budget guarantee: credited progress never exceeds planned hours (Pᵢ ≤ bᵢ/ρ), at convergence and at every intermediate iterate of the algorithm.
  • Evidence below the affinity floor is left genuinely unattributed (exact zero share), making the market the only fractional rule that is also sparse — no diffuse slivers that a human must mentally threshold.
  • A task nearing its plan discounts further progress through the concave completion utility, and the satiation-threshold fixed point keeps existence and local convergence where the naive fix (rescaling valuations by a constant) provably does nothing.
  • In the near-duplicate, high-contention regime — the same regime that stresses the convergence proof — the outer loop slows and can fail (1 of 10 adversarial instances), so the paper's own prescription is to monitor the residual and fall back to the unregularized market.
  • On raw share accuracy the paper's benchmark favors entropic optimal transport at every noise level, so the practical conclusion is a choice: the market where guarantees matter, entropic transport where accuracy does, with a one-parameter entropy dial interpolating between the two.

Reading between the lines

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

  • The reserve-and-cash instruments are not tied to task planning: the same pair could give hard over-credit guarantees in multi-touch advertising attribution, shared-infrastructure cost allocation, or co-authorship credit — domains the paper relates to but does not port the mechanism into.
  • The noise-adaptive entropy dial suggests a general principle — sharpen an equilibrium when inputs are trustworthy, smooth it when they are not — and a direct test: choose τ by cross-validation on held-out corrupted affinities and check that it lands between the two endpoints in total-variation error.
  • The exact-zero junk filter likely shifts where the model's value appears: not in aggregate error metrics but in whether a human can act on a sparse, thresholded report — a claim the paper gestures at but does not measure; a user study comparing legibility of sparse versus dense attributions would test it.
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 to bridge planned tasks and performed actions by formulating credit assignment as a quasi-linear Fisher market: tasks are budget-constrained buyers, actions are divisible goods, and a fused affinity signal sets valuations. A seller reserve price and a buyer cash option are claimed to deliver, as theorems, conservation (D1), a hard budget cap (D2), and a junk filter (D3). The paper then extends the market with a concave completion utility, resolved by a satiation-threshold fixed point (Proposition 6 for existence, Proposition 7 for local convergence), reports a de-circularized multi-seed benchmark against hard assignment, softmax, and entropic optimal transport, and proposes an entropy-regularized generalization to mitigate a documented noise-sensitivity weakness. Reproducibility parameters, a worked example, and candid limitations are included.

Significance. If the advertised guarantees held, the market would give a principled fractional-attribution mechanism with theorem-level budget control, which would be a meaningful contribution to task-attribution and decision-support systems. The paper has genuine strengths: the benchmark is deliberately de-circularized; the reproducibility table is unusually complete; the worked example supports independent implementation; and the limitations section candidly admits the conditional nature of the convergence result and the market's weaker raw accuracy. These strengths, however, do not repair the central problem: the 'hard budget cap' claimed in the abstract, Section 4.2, Table 4, and the conclusion is not what Proposition 2 proves. Because D2 is a load-bearing design requirement and the paper's own Section 6.7 explicitly writes the cap as P_i ≤ b_i/ρ, the mismatch is internal and substantive rather than a presentational slip.

major comments (3)
  1. [§3.3 (D2), §4.2 (Prop. 2), Table 4 caption] Same as above
  2. [§5.2, Proposition 6]
  3. [§5.2, Proposition 7 and §6.6]
minor comments (6)
  1. [§4.2] The sentence 'Proposition 2 and 3 give D2 and D3 as theorems' should be rephrased or corrected, because Proposition 2 does not imply the stated D2. At minimum, D2 should be restated as P_i ≤ b_i/ρ, or ρ should be required to be ≥ 1 for the original wording.
  2. [§6.7] The bullet 'Exact, anytime budget cap' correctly writes the cap as P_i ≤ b_i/ρ, which contradicts the earlier claim that the market never over-credits a task beyond its planned budget. Please reconcile the two statements.
  3. [Table 4 caption] The statement 'The market's zero-violation column is a consequence of Proposition 2' is not supportable, as explained in the major comment. The caption should instead say that zero violations are observed empirically in these instances, and that Proposition 2 only guarantees the weaker b_i/ρ bound.
  4. [§5.2 / Table 3] The domain of μ is stated as (0,1]^m in the text but the algorithm uses μ_floor = 0.02. Please make the domain explicit and consistent, e.g., [μ_floor, 1]^m, and check the proof of Proposition 6 against that domain.
  5. [§6.5] The 'noise-adaptive rule' for selecting τ is described only qualitatively ('select τ ... in proportion to the estimated affinity-observation noise'). No concrete estimator for σ-hat, no functional form, and no validation of the prescription are given. Since the paper acknowledges that the general-τ algorithm is not engineered, this part should be flagged as a heuristic proposal, not a validated result.
  6. [Abstract and §1.2] The abstract and contribution section say the market instruments yield a 'hard budget cap' as a theorem. Please qualify the cap as P_i ≤ b_i/ρ unless ρ ≥ 1 is assumed, and adjust the corresponding claims in Section 1.2 and the conclusion accordingly.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: central guarantees are derived from explicit market definitions; the only self-citation [2] is scoped out, and the benchmark is deliberately de-circularized. The D2-vs-P_i≤b_i/ρ gap is a logical overclaim, not a circular derivation.

full rationale

The paper's load-bearing derivation chain is self-contained. Propositions 1–3 follow algebraically from Definition 1 (market clearing, p_j≥ρd_j, and the equilibrium condition (3)); none of these theorems is assumed as an input or fit to the benchmark. Section 6.1 explicitly de-circularizes the evaluation by generating ground-truth shares independently of the observed affinity (Q_obs = clip(Q_clean + σN + confuser noise, 0,1)), so the comparative results are not self-validating. The only self-citation, the companion technical report [2], is explicitly non-load-bearing: Section 5.4 says the temporal and forward-looking extensions are 'summarized here only to situate the present paper's scope and do not evaluate them empirically in this work,' and Section 3.2 distinguishes the present concrete fitting procedure from [2]'s exploratory treatment. Proposition 7's diagonal-dominance condition is an openly stated sufficient condition whose failure is admitted and empirically probed, so it is not smuggled in as an assumption. One flagged passage is not a circularity: Section 4.2 says 'Proposition 2 and 3 give D2 and D3 as theorems,' and Table 4's caption calls the market's zero-violation column 'a consequence of Proposition 2, not an empirical observation,' but Proposition 2 proves only P_i≤b_i/ρ, not P_i≤b_i; with ρ=0.25 this leaves a 4× gap and makes the zero-violation count empirical rather than theorem-forced. This is an internal logical overclaim about the strength of a genuinely derived bound, not a reduction of an output to an input, so it does not raise the circularity score.

Assumptions & free parameters 4 free parameters · 5 assumptions · 2 invented entities

The central claims rest on classical market-equilibrium results and a conditional contraction argument. The free parameters (ρ, u0) are user-specified; τ is left open. The float and μ are invented constructs without independent empirical support.

free parameters (4)
  • reserve rate ρ = 0.25
    Set by hand in experiments; it sets the effective budget cap P_i ≤ b_i/ρ and the junk-filter threshold u0ρ. The paper does not justify this value.
  • cash utility u0 = 0.30
    Set by hand; controls the junk filter and the incentive to hold cash.
  • entropy-regularization strength τ
    The paper proposes a noise-adaptive rule but does not provide a functional form or fitted value.
  • affinity fusion parameters λ_sem, λ_link, λ_time = fitted by logistic regression (Eq. 2) on user corrections
    The paper says these are deployment-specific and not load-bearing; in the synthetic experiments affinities are generated directly.
assumptions (5)
  • standard math Eisenberg-Gale equilibrium existence and PRD convergence for linear Fisher markets
    Invoked for the base market guarantees and Algorithm 1 convergence (Sections 4.1, 4.3).
  • standard math Brouwer fixed-point theorem applied to a continuous map on a compact convex set
    Used to prove existence of the completion-market fixed point (Proposition 6).
  • standard math Banach fixed-point theorem for local contraction
    Used for the local uniqueness and convergence of Algorithm 2 under diagonal dominance (Proposition 7).
  • domain assumption Equilibrium utilities V(μ) of the inner linear Fisher market are unique and continuous in μ
    Required for the fixed-point map Φ to be continuous; asserted as a classical fact (Section 5.2).
  • ad hoc to paper The domain of μ is [μ_floor,1]^m with μ_floor=0.02 in the algorithm, not the stated (0,1]^m
    The written proof uses (0,1]^m which is non-compact; the implementation clamps μ to make Brouwer applicable (Table 3, Algorithm 2).
invented entities (2)
  • The float (task index 0)
    purpose: A synthetic buyer that bids a reserve ρd_j on every action, absorbing unattributed effort and enforcing the junk filter.
    It is a modeling device introduced by the paper; no external observable predicted.
  • Satiation multiplier μ_i
    purpose: A latent variable encoding the derivative of the completion utility, used in the fixed-point loop.
    It is an internal computational parameter, not an empirical entity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Attribution Markets: A Fisher-Market Formulation for Fractional Credit Assignment Between Planned Tasks and Performed Actions." pith.science (2026). https://pith.science/paper/RGUUIIM5

@misc{pith2026260720694,
  author       = {Pith},
  title        = {Pith review of: Attribution Markets: A Fisher-Market Formulation for Fractional Credit Assignment Between Planned Tasks and Performed Actions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RGUUIIM5}},
  note         = {Machine review of arXiv:2607.20694}
}
read the original abstract

Personal and organizational planning systems maintain two records that drift apart: what was planned (a task's effort budget) and what was done (a logged action's duration and description). Existing systems bridge them with an exclusive, all-or-nothing link that strands genuinely related but unlinked effort and reports false stalls on active goals. We formulate the bridge as a quasi-linear Fisher market: planned tasks are budget-constrained buyers, performed actions are divisible goods, and a fused text/structural/temporal signal sets each buyer's valuation. Two market instruments - a seller reserve price and a buyer cash option - yield conservation, a hard budget cap, and a provable junk filter as theorems. We extend the market with a concave completion utility discounting progress as a task nears its plan; standard convergence theory for the market's algorithm does not transfer here, resolved by a satiation-threshold fixed point with existence (Brouwer) and local uniqueness under an explicit diagonal-dominance condition, validated empirically on random and adversarial instances. A de-circularized, multi-seed benchmark - observed affinity corrupted independently of the scored ground truth - surfaces a genuine weak spot: the market's sharp, zero-entropy equilibrium is more sensitive to affinity noise than entropy-regularized optimal transport's permanently smoothed one. We resolve this with a one-parameter entropy-regularized generalization unifying the two, plus a noise-adaptive rule for its regularization strength. We report full reproducibility parameters, discuss limitations candidly, and relate the result to multi-touch attribution, optimal transport, and online Fisher-market algorithms.

Figures

Figures reproduced from arXiv: 2607.20694 by the authors.

Figure 1
Figure 1. The attribution pipeline, read left to right in four stages. [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. The attribution market. Two tasks (investors) hold shares of three actions [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Algorithm 2: an outer fixed-point loop over the satiation multipliers [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: (a) Total-variation share error rises with observation noise for every rule; [PITH_FULL_IMAGE:figures/full_fig_p020_4.png]
Figure 5
Figure 5. Figure 5: Outer-loop convergence: log-residual versus iteration. Random instances (blue) [PITH_FULL_IMAGE:figures/full_fig_p022_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

52 extracted references · 5 canonical work pages

  1. [1]

    Eisenberg, D

    E. Eisenberg, D. Gale, Consensus of subjective probabilities: The pari- mutuel method, The Annals of Mathematical Statistics 30 (1) (1959) 165–168

  2. [2]

    Ishbulatov, The attribution market: A theory of dynamic shared credit between planned tasks and performed actions, Technical report (2026)

    S. Ishbulatov, The attribution market: A theory of dynamic shared credit between planned tasks and performed actions, Technical report (2026). 29

  3. [3]

    Minsky, Steps toward artificial intelligence, Proceedings of the IRE 49 (1) (1961) 8–30.doi:10.1109/JRPROC.1961.287775

    M. Minsky, Steps toward artificial intelligence, Proceedings of the IRE 49 (1) (1961) 8–30.doi:10.1109/JRPROC.1961.287775

  4. [4]

    R. S. Sutton, Temporal credit assignment in reinforcement learning, Ph.D. thesis, University of Massachusetts Amherst (1984)

  5. [5]

    Pignatelli, J

    E. Pignatelli, J. Ferret, M. Geist, T. Mesnard, H. van Hasselt, O. Pietquin, L. Toni, A survey of temporal credit assignment in deep reinforcement learning (2024).arXiv:2312.01072

  6. [6]

    X. Shao, L. Li, Data-drivenmulti-touchattributionmodels, in: Proceed- ings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’11, Association for Computing Ma- chinery, NewYork, NY,USA,2011, p.258–264.doi:10.1145/2020408. 2020453

  7. [7]

    Anderl, I

    E. Anderl, I. Becker, F. von Wangenheim, J. H. Schumann, Mapping the customer journey: Lessons learned from graph-based online attribution modeling, International Journal of Research in Marketing 33 (3) (2016) 457–474.doi:10.1016/j.ijresmar.2016.03.001

  8. [8]

    D. Yao, C. Gong, L. Zhang, S. Chen, J. Bi, Causalmta: Eliminating the userconfoundingbiasforcausalmulti-touchattribution, in: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD ’22, Association for Computing Machinery, New York, NY, USA, 2022, p. 4342–4352.doi:10.1145/3534678.3539108

Show all 52 references
  1. [9]

    A. P. Dempster, N. M. Laird, D. B. Rubin, Maximum likelihood from incomplete data via the em algorithm, Journal of the Royal Statistical Society: Series B (Methodological) 39 (1) (1977) 1–22.doi:10.1111/ j.2517-6161.1977.tb01600.x

  2. [10]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, I. Polosukhin, Attention is all you need, in: I. Guyon, U. V. Luxburg, S.Bengio, H.Wallach, R.Fergus, S.Vishwanathan, R.Garnett (Eds.), Advances in Neural Information Processing Systems, Vol. 30...

  3. [11]

    L. V. Kantorovich, On the translocation of masses, Doklady Akademii Nauk SSSR 37 (1942) 199–201. 30

  4. [12]

    Cuturi, Sinkhorn distances: Lightspeed computation of optimal transport, in: C

    M. Cuturi, Sinkhorn distances: Lightspeed computation of optimal transport, in: C. Burges, L. Bottou, M. Welling, Z. Ghahramani, K. Weinberger (Eds.), Advances in Neural Information Processing Sys- tems, Vol. 26, Curran Associates, Inc., 2013

  5. [13]

    G.Peyré, M.Cuturi, Computationaloptimaltransportwithapplications to data sciences, Foundations and Trends in Machine Learning 11 (5-6) (2019) 355–607.doi:10.1561/2200000073

  6. [14]

    Chizat, G

    L. Chizat, G. Peyré, B. Schmitzer, F.-X. Vialard, Scaling algorithms for unbalanced optimal transport problems, Mathematics of Computation 87 (2018) 2563–2609

  7. [15]

    Scetbon, M

    M. Scetbon, M. Klein, G. Palla, M. Cuturi, Unbalanced low-rank opti- mal transport solvers, in: A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, S. Levine (Eds.), Advances in Neural Information Processing Systems, Vol. 36, Curran Associates, Inc., 2023, pp. 52312–52325

  8. [16]

    Nisan, T

    N. Nisan, T. Roughgarden, É. Tardos, V. V. Vazirani (Eds.), Algorith- mic Game Theory, Cambridge University Press, 2007

  9. [17]

    L. Zhang, Proportional response dynamics in the fisher market, The- oretical Computer Science 412 (24) (2011) 2691–2698, selected Papers from 36th International Colloquium on Automata, Languages and Pro- gramming (ICALP 2009).doi:10.1016/j.tcs.2010.06.021

  10. [18]

    Birnbaum, N

    B. Birnbaum, N. R. Devanur, L. Xiao, Distributed algorithms via gra- dient descent for fisher markets, in: Proceedings of the 12th ACM Conference on Electronic Commerce, EC ’11, Association for Com- puting Machinery, New York, NY, USA, 2011, p. 127–136.doi: 10.1145/1993574.1993594

  11. [19]

    F. P. Kelly, A. K. Maulloo, D. K. H. Tan, Rate control for commu- nication networks: shadow prices, proportional fairness and stability, Journal of the Operational Research Society 49 (3) (1998) 237–252. doi:10.1057/palgrave.jors.2600523

  12. [20]

    J. F. Nash, The bargaining problem, Econometrica 18 (2) (1950) 155– 162. 31

  13. [21]

    Budish, The combinatorial assignment problem: Approximate com- petitive equilibrium from equal incomes, Journal of Political Economy 119 (6) (2011) 1061–1103.doi:10.1086/664613

    E. Budish, The combinatorial assignment problem: Approximate com- petitive equilibrium from equal incomes, Journal of Political Economy 119 (6) (2011) 1061–1103.doi:10.1086/664613

  14. [22]

    Conitzer, C

    V. Conitzer, C. Kroer, E. Sodomka, N. E. Stier-Moses, Multiplicative pacing equilibria in auction markets, Operations Research 70 (2) (2022) 963–989

  15. [23]

    Y. Gao, A. Peysakhovich, C. Kroer, Online market equilibrium with ap- plication to fair division, in: M. Ranzato, A. Beygelzimer, Y. Dauphin, P. Liang, J. W. Vaughan (Eds.), Advances in Neural Information Pro- cessing Systems, Vol. 34, Curran Associates, Inc., 2021, pp. 27305– 27318

  16. [24]

    L. Liao, Y. Gao, C. Kroer, Statistical inference for fisher market equi- librium (2025).arXiv:2209.15422

  17. [25]

    J. L. Kelly, Jr., A new interpretation of information rate, Bell System Technical Journal 35 (4) (1956) 917–926

  18. [26]

    T. M. Cover, Universal portfolios, Mathematical Finance 1 (1) (1991) 1–29.doi:10.1111/j.1467-9965.1991.tb00002.x

  19. [27]

    Markowitz, Portfolio selection, The Journal of Finance 7 (1) (1952) 77–91

    H. Markowitz, Portfolio selection, The Journal of Finance 7 (1) (1952) 77–91

  20. [28]

    Black, R

    F. Black, R. Litterman, Global portfolio optimization, Financial Ana- lysts Journal 48 (5) (1992) 28–43.doi:10.2469/faj.v48.n5.28

  21. [29]

    Ledoit, M

    O. Ledoit, M. Wolf, Honey, I shrunk the sample covariance matrix, Journal of Portfolio Management 30 (4) (2004) 110–119

  22. [30]

    A. N. Dragunov, T. G. Dietterich, K. Johnsrude, M. McLaughlin, L. Li, J. L. Herlocker, Tasktracer: a desktop environment to support multi-tasking knowledge workers, in: Proceedings of the 10th Inter- national Conference on Intelligent User Interfaces, IUI ’05, Associa- tion fo...

  23. [31]

    86–92.doi:10.1145/1111449.1111473

    J.Shen, L.Li, T.G.Dietterich, J.L.Herlocker, Ahybridlearningsystem for recognizing user tasks from desktop activities and email messages, 32 in: Proceedings of the 11th International Conference on Intelligent User Interfaces, IUI ’06, Association for Computing Machinery, New Y...

  24. [32]

    I. P. Fellegi, A. B. Sunter, A theory for record linkage, Journal of the American Statistical Association 64 (328) (1969) 1183–1210

  25. [33]

    Peeters, A

    R. Peeters, A. Steiner, C. Bizer, Entity matching using large language models (2024).arXiv:2310.11244

  26. [34]

    Muennighoff, N

    N. Muennighoff, N. Tazi, L. Magne, N. Reimers, MTEB: Massive text embedding benchmark, in: A. Vlachos, I. Augenstein (Eds.), Proceed- ings of the 17th Conference of the European Chapter of the Associa- tion for Computational Linguistics, Association for Computational Lin- guis...

  27. [35]

    Duchi, S

    J. Duchi, S. Shalev-Shwartz, Y. Singer, T. Chandra, Efficient projec- tions onto the l1-ball for learning in high dimensions, in: Proceedings of the 25th International Conference on Machine Learning, ICML ’08, Association for Computing Machinery, New York, NY, USA, 2008, p. 27...

  28. [36]

    Nesterov, Introductory Lectures on Convex Optimization: A Basic Course, Springer, 2004

    Y. Nesterov, Introductory Lectures on Convex Optimization: A Basic Course, Springer, 2004

  29. [37]

    Bubeck, Convex optimization: Algorithms and complexity, Founda- tions and Trends in Machine Learning 8 (3-4) (2015) 231–357.doi: 10.1561/2200000050

    S. Bubeck, Convex optimization: Algorithms and complexity, Founda- tions and Trends in Machine Learning 8 (3-4) (2015) 231–357.doi: 10.1561/2200000050

  30. [38]

    F.Facchinei, J.-S.Pang, Finite-DimensionalVariationalInequalitiesand Complementarity Problems, Springer, 2003

  31. [39]

    M. H. A. Davis, A. R. Norman, Portfolio selection with transaction costs, Mathematics of Operations Research 15 (4) (1990) 676–713.doi: 10.1287/moor.15.4.676

  32. [40]

    Freund, R

    Y. Freund, R. E. Schapire, A decision-theoretic generalization of on- line learning and an application to boosting, Journal of Computer and System Sciences 55 (1) (1997) 119–139.doi:10.1006/jcss.1997.1504. 33

  33. [41]

    P. Auer, N. Cesa-Bianchi, Y. Freund, R. E. Schapire, The nonstochastic multiarmed bandit problem, SIAM Journal on Computing 32 (1) (2002) 48–77.doi:10.1137/S0097539701398375

  34. [42]

    T. Lai, H. Robbins, Asymptotically efficient adaptive allocation rules, Adv. Appl. Math. 6 (1) (1985) 4–22.doi:10.1016/0196-8858(85) 90002-8

  35. [43]

    Amershi, M

    S. Amershi, M. Cakmak, W. B. Knox, T. Kulesza, Power to the people: The role of humans in interactive machine learning, AI Magazine 35 (4) (2014) 105–120.doi:10.1609/aimag.v35i4.2513

  36. [44]

    Zheng, W.-L

    L. Zheng, W.-L. Chiang, Y. Sheng, S. Zhuang, Z. Wu, Y. Zhuang, Z. Lin, Z. Li, D. Li, E. Xing, H. Zhang, J. Gonzalez, I. Stoica, Judging llm-as- a-judge with mt-bench and chatbot arena, in: A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, S. Levine (Eds.), Advances in Neu-...

  37. [45]

    Sinkhorn, P

    R. Sinkhorn, P. Knopp, Concerning nonnegative matrices and doubly stochastic matrices, Pacific Journal of Mathematics 21 (2) (1967) 343– 348

  38. [46]

    L. S. Shapley, A value forn-person games, in: Contributions to the Theory of Games II, Princeton University Press, 1953, pp. 307–317

  39. [47]

    S. M. Lundberg, S.-I. Lee, A unified approach to interpreting model predictions, in: Advances in Neural Information Processing Systems (NeurIPS), 2017

  40. [48]

    Harkin, T

    B. Harkin, T. L. Webb, B. P. I. Chang, A. Prestwich, M. Conner, I. Kel- lar, Y. Benn, P. Sheeran, Does monitoring goal progress promote goal attainment? a meta-analysis of the experimental evidence., Psychologi- cal Bulletin 142 (2) (2016) 198–229.doi:10.1037/bul0000025

  41. [49]

    D. Xiu, G. E. Karniadakis, The wiener–askey polynomial chaos for stochastic differential equations, SIAM Journal on Scientific Comput- ing 24 (2) (2002) 619–644.doi:10.1137/S1064827501387826

  42. [50]

    Lasry, P.-L

    J.-M. Lasry, P.-L. Lions, Mean field games, Japanese Journal of Math- ematics 2 (1) (2007) 229–260.doi:10.1007/s11537-007-0657-8. 34

  43. [51]

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stu- art, A. Anandkumar, Fourier neural operator for parametric partial dif- ferential equations, 2021.arXiv:2010.08895

  44. [52]

    W. E, J. Han, A. Jentzen, Deep learning-based numerical methods for high-dimensional parabolic partial differential equations and backward stochastic differential equations, Communications in Mathematics and Statistics 5 (4) (2017) 349–380.doi:10.1007/s40304-017-0117-6. 35

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.