pith. sign in

arxiv: 2606.04779 · v1 · pith:NBJ7YWT4new · submitted 2026-06-03 · 💻 cs.AI · math.CO

Tree-Based Formalization of Multi-Agent Complementarity in Human-AI Interactions

Andrea Ferrario This is my paper

Pith reviewed 2026-06-28 05:59 UTC · model grok-4.3

classification 💻 cs.AI math.CO
keywords complementarityhuman-AI interactionmulti-agent protocolstree formalizationregressionclassificationloss functionsbarycentric coordinates
0
0 comments X

The pith

A tree-based protocol model shows multi-agent human-AI complementarity is attainable in regression but obstructed in classification under standard losses.

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

The paper introduces a formal representation of human-AI interaction protocols as ordered agent-role configurations paired with rooted planar binary trees whose leaves hold prediction vectors. Local binary composition rules are applied recursively down the tree to produce a complementarity measure relative to a pointwise-min oracle. Four theorems follow: selectors cannot reach complementarity, squared-loss regression reduces complementarity to Euclidean distance minimization from the ground truth, linear composition yields barycentric coordinates on the simplex with Tamari invariance, and endpoint-monotone losses block complementarity in binary and multiclass classification. A reader cares because the results separate tasks where combining agents can improve performance from those where it cannot under natural modeling choices.

Core claim

An HAI protocol is an ordered agent-role configuration together with a rooted planar binary tree whose leaves are decorated by prediction vectors; a local binary composition rule is evaluated recursively along the tree to yield a tree-relative complementarity functional relative to a pointwise-min oracle benchmark. Selector-based protocols cannot achieve complementarity. In squared-loss regression complementarity is equivalent to Euclidean distance minimization from the ground-truth vector, with a closed-form optimal weight for N=2. Under linear local composition every protocol tree defines a barycentric coordinate chart on the simplex of leaf weights, Tamari-cover reparameterizations preser

What carries the argument

Rooted planar binary tree with leaves as prediction vectors, composed recursively by a local binary rule to define a tree-relative complementarity functional relative to a pointwise-min oracle.

If this is right

  • Selector-based HAIs, including self-reliance or full AI-reliance, cannot achieve complementarity for any task or loss.
  • In regression under squared loss, complementarity is exactly Euclidean distance minimization from the ground-truth vector.
  • Under linear local composition, protocol trees induce barycentric coordinates on the simplex and remain invariant under Tamari covers.
  • No internal composition rule achieves complementarity in binary classification under endpoint-monotone losses or in multiclass under cross-entropy.

Where Pith is reading between the lines

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

  • The obstruction in classification suggests that HAI system designers should prefer regression-style tasks or redesign losses when the goal is complementarity.
  • The barycentric coordinate view may connect the framework to existing ensemble weighting methods in statistics, allowing transfer of known simplex results.
  • Testing the closed-form N=2 regression weight on real human-AI data would check whether the Euclidean-distance equivalence holds outside synthetic settings.
  • Extending the tree model to allow non-binary internal nodes could reveal whether the classification obstruction persists beyond binary trees.

Load-bearing premise

Every HAI protocol can be represented by an ordered agent-role configuration together with a rooted planar binary tree whose leaves are decorated by prediction vectors, with complementarity measured relative to a pointwise-min oracle benchmark.

What would settle it

An explicit local composition rule and loss function pair in binary classification that produces a combined prediction strictly better than the pointwise-min oracle under an endpoint-monotone loss would falsify the obstruction theorem.

Figures

Figures reproduced from arXiv: 2606.04779 by Andrea Ferrario.

Figure 1
Figure 1. Figure 1: Geometric interpretation of complementarity in regression under squared loss in [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The rooted planar binary tree for N = 2, with leaves decorated by human and AI roles, the corresponding prediction vectors yˆ (1) and yˆ (2) . yˆT is the tree output. If An > 0, the maximizing weight is α ∗ = Π[0,1] − Bn An  . If An = 0, the aggregate is constant in α and yields no complementarity. Proof. Appendix A. Proposition 4 admits an explicit geometric interpretation. Let us rewrite Bn as Bn = ∥yˆ… view at source ↗
Figure 3
Figure 3. Figure 3: Geometry for the N = 2 regression case with linear pooling. The segment of the line through yˆ AI and yˆ H that lies between the two vectors is the locus of feasible human–AI team predictions. The line through the origin 0 is the human–AI disagreement direction. The vector yˆ AI − y is projected onto this line and θ is the angle between yˆ AI − y and yˆ H − yˆ AI . The interior case inequalities (16) are b… view at source ↗
Figure 4
Figure 4. Figure 4: N = 2 regression complementarity under linear pooling using the California housing dataset. Each curve shows the quadratic value of nΨ mid 2,α T [($100,000)2 ] as a function of the aggregation weight α, for synthetic human predictions constructed by controlling the angle θ between yˆ H − yˆ AI and yˆ AI − y, and the relative displacement q = ∥yˆ H − yˆ AI ∥2/∥yˆ AI − y∥2. Vertical markers indicate the cons… view at source ↗
Figure 5
Figure 5. Figure 5: The two rooted planar binary trees for N = 3 and the ordered configuration (expert, assistant, AI), with corresponding prediction vectors yˆ (1) , yˆ (2), and yˆ (3). The trees represent two distinct protocols of the same prediction￾task HAI: TL = ((12)3) first combines expert and assistant predictions, whereas TR = (1(23)) first combines assistant and AI predictions. The vectors yˆTL and yˆTR denote the c… view at source ↗
Figure 6
Figure 6. Figure 6: N = 3 regression complementarity under linear pooling using the California housing dataset. Each panel shows the values of P(α1, α2)/n [($100,000)2 ] and the protocol-indifference locus P(α1, α2) = 0 for the two protocol trees TL = ((12)3) and TR = (1(23)). Blue regions, where P(α1, α2) < 0, indicate higher complementarity for TL; red regions, where P(α1, α2) > 0, indicate higher complementarity for TR, ac… view at source ↗
Figure 7
Figure 7. Figure 7: The five rooted planar binary trees for N = 4 under the ordered configuration (user, AI, AI, user), with corresponding prediction vectors yˆ (1) , yˆ (2) , yˆ (3) , yˆ (4) and tree outputs yˆT1 , . . . , yˆT5 . 6 Trees in Regression Under Linear Pooling: Barycentric Coordinates We turn now to the question whether in regression problems distinct protocols can lead to the same HAI output for the same paramet… view at source ↗
Figure 8
Figure 8. Figure 8: The pentagon identity satisfied by the Tamari-cover reparameterizations for [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: N = 2 binary classification under cross-entropy with amplified logit pooling and fixed α = 0.5. Each point is one simulated pair of probabilistic predictors, plotted by the class-wise rates (k0, k1) of canonical local complementarity: for yi = 0, the pooled prediction lies below both input probabilities; for yi = 1, it lies above both input probabilities. Color indicates global complementarity: blue points… view at source ↗
read the original abstract

Complementarity is the case in which a human--AI interaction (HAI) outperforms the best prediction benchmark available among its members. Although this idea is central in HAI research, formal work on complementarity remains limited. Existing frameworks do not model how agents' predictions compose into workflow-sensitive multi-agent protocols. We close this gap by introducing a tree-based formalization of complementarity in multi-agent HAI. An HAI protocol is represented by an ordered agent-role configuration together with a rooted planar binary tree whose leaves are decorated by prediction vectors. A local binary composition rule is evaluated recursively along the tree, yielding a tree-relative complementarity functional relative to a pointwise-min oracle benchmark. We prove four results. First, selector-based HAIs, including self- or AI-reliance, cannot achieve complementarity regardless of task, loss, or prediction quality. Second, in regression under squared loss, complementarity is equivalent to Euclidean distance minimization from the ground-truth vector; for $N=2$, the optimal linear-pooling weight has a closed form and a residual-correction interpretation. Third, under linear local composition, every protocol tree defines a barycentric coordinate chart on the simplex of leaf weights; Tamari-cover reparameterizations of protocol trees preserve complementarity, and for $N=4$, they satisfy the pentagon identity. Fourth, in binary classification, no internal local composition can achieve complementarity under endpoint-monotone losses, including standard Bregman and many finite Bernoulli $f$-divergence losses; an analogous obstruction holds for multiclass aggregation under cross-entropy. In summary, our framework shows that complementarity is attainable in multi-agent regression, but obstructed in classification under natural conditions on local aggregation and loss functions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 2 minor

Summary. The paper introduces a tree-based formalization of complementarity in multi-agent human-AI (HAI) interactions. HAI protocols are represented as ordered agent-role configurations together with rooted planar binary trees whose leaves are decorated by prediction vectors. Local binary composition rules are evaluated recursively along the tree to yield a tree-relative complementarity functional relative to a pointwise-min oracle benchmark. Four theorems are proved: selector-based HAIs cannot achieve complementarity; in regression under squared loss, complementarity is equivalent to Euclidean distance minimization from the ground-truth vector (with closed-form optimal linear-pooling weight for N=2); under linear local composition every protocol tree defines a barycentric coordinate chart on the simplex of leaf weights, Tamari-cover reparameterizations preserve complementarity, and the pentagon identity holds for N=4; and in binary classification no internal local composition achieves complementarity under endpoint-monotone losses (including standard Bregman and many finite Bernoulli f-divergence losses), with an analogous obstruction for multiclass aggregation under cross-entropy. In summary, complementarity is attainable in multi-agent regression but obstructed in classification under the stated conditions.

Significance. If the theorems hold, the work supplies a precise formal framework for studying complementarity in multi-agent HAI settings, an area where existing research has been limited. The clean separation between attainability results in regression and obstruction results in classification under natural conditions on losses and compositions is a substantive contribution. The explicit link to barycentric coordinates and Tamari covers, together with the provision of machine-checkable-style theorems, strengthens the paper; these elements allow the claims to be stated and potentially verified without post-hoc parameter fitting.

minor comments (2)
  1. The abstract is dense with specialized terminology (e.g., "tree-relative complementarity functional," "endpoint-monotone losses"); a brief definitional paragraph or glossary in the introduction would improve accessibility for readers outside the immediate subfield.
  2. A concrete small-N example (e.g., N=2 or N=3) with an explicit tree diagram, agent roles, and numerical predictions would make the protocol representation and recursive evaluation easier to follow.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the detailed and accurate summary of our manuscript, the positive assessment of its significance, and the recommendation for minor revision. No specific major comments were raised in the report.

Circularity Check

0 steps flagged

No significant circularity; framework is self-contained formal model

full rationale

The paper defines an HAI protocol via ordered agent-role configurations and rooted planar binary trees with recursive local composition rules, then measures complementarity relative to a pointwise-min oracle. All four theorems are stated as formal consequences inside this model (selector-based HAIs cannot achieve complementarity; regression equivalence to Euclidean minimization; barycentric charts and Tamari reparameterizations; obstructions under endpoint-monotone losses in classification). No parameter fitting, no self-citation chains, and no reduction of a claimed prediction to its own inputs by construction. The modeling choices are explicit definitions, not hidden premises that collapse the derivations.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The central claim rests on a domain-specific modeling assumption that HAI protocols are captured by planar binary trees with local binary composition; no free parameters are introduced because the regression weights are derived in closed form and the obstructions are proved for broad classes of losses.

axioms (2)
  • domain assumption HAI protocols are represented by ordered agent-role configurations together with rooted planar binary trees whose leaves hold prediction vectors
    This is the foundational modeling choice stated in the abstract that enables the recursive evaluation of the complementarity functional.
  • domain assumption Complementarity is measured relative to a pointwise-min oracle benchmark
    The functional is defined with respect to this benchmark throughout the claimed results.
invented entities (1)
  • tree-relative complementarity functional no independent evidence
    purpose: Quantifies the performance gain of the tree protocol over the pointwise-min benchmark
    Newly defined object that carries the main results; no independent evidence outside the framework is supplied.

pith-pipeline@v0.9.1-grok · 5831 in / 1403 out tokens · 39376 ms · 2026-06-28T05:59:30.653511+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

79 extracted references · 3 canonical work pages · 3 internal anchors

  1. [1]

    2026 , publisher=

    Beyond transparency: Computational reliabilism as an externalist epistemology of algorithms , author=. 2026 , publisher=

    2026

  2. [2]

    Social science & medicine , volume=

    The use of QALYs in health care decision making , author=. Social science & medicine , volume=. 1989 , publisher=

    1989

  3. [3]

    Does the whole exceed its parts?

    Bansal, Gagan and Wu, Tongshuang and Zhou, Joyce and Fok, Raymond and Nushi, Besmira and Kamar, Ece and Ribeiro, Marco Tulio and Weld, Daniel , booktitle=. Does the whole exceed its parts?

  4. [4]

    Is the most accurate

    Bansal, Gagan and Nushi, Besmira and Kamar, Ece and Horvitz, Eric and Weld, Daniel S , booktitle=. Is the most accurate

  5. [5]

    Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency , pages=

    Human-algorithm collaboration: Achieving complementarity and avoiding unfairness , author=. Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency , pages=

    2022

  6. [6]

    When combinations of humans and

    Vaccaro, Michelle and Almaatouq, Abdullah and Malone, Thomas , journal=. When combinations of humans and. 2024 , publisher=

    2024

  7. [7]

    Complementarity in human-

    Philipp Hemmer and Michael Schemmer and Niklas K. Complementarity in human-. 2025 , pages =

    2025

  8. [8]

    2025 , volume =

    Michaela Benk and Sophie Kerstan and Florian von Wangenheim and Andrea Ferrario , title =. 2025 , volume =

    2025

  9. [9]

    Grounds for Trust:

    Juan Manuel Dur. Grounds for Trust:. 2018 , volume =

    2018

  10. [10]

    Proceedings of the 2020 CHI conference on human factors in computing systems , pages=

    A human-centered evaluation of a deep learning system deployed in clinics for the detection of diabetic retinopathy , author=. Proceedings of the 2020 CHI conference on human factors in computing systems , pages=

    2020

  11. [11]

    The Algorithmic Automation Problem: Prediction, Triage, and Human Effort

    The algorithmic automation problem: Prediction, triage, and human effort , author=. arXiv preprint arXiv:1903.12220 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 1903

  12. [12]

    2024 , volume =

    Andrea Ferrario , title =. 2024 , volume =

    2024

  13. [13]

    Investigating Human + Machine Complementarity for Recidivism Predictions

    Investigating human+ machine complementarity for recidivism predictions , author=. arXiv preprint arXiv:1808.09123 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  14. [14]

    Studying human-AI collaboration protocols:

    Cabitza, Federico and Campagner, Andrea and Sconfienza, Luca Maria , journal=. Studying human-AI collaboration protocols:. 2021 , publisher=

    2021

  15. [15]

    Communications of the Association for Information Systems , volume=

    Reflecting on the ``dark side''' of information technology use , author=. Communications of the Association for Information Systems , volume=

  16. [16]

    Cognitive challenges in human--artificial intelligence collaboration:

    F. Cognitive challenges in human--artificial intelligence collaboration:. Information Systems Research , volume=. 2022 , publisher=

    2022

  17. [17]

    Prejudiced against the machine?

    Turel, Ofir and Kalhan, Shivam , journal=. Prejudiced against the machine?. 2023 , publisher=

    2023

  18. [18]

    Hemmer, Patrick and Schemmer, Max and V. Human-. PACIS , volume=

  19. [19]

    Communications of the ACM , volume=

    The effects of mixing machine learning and human judgment , author=. Communications of the ACM , volume=. 2019 , publisher=

    2019

  20. [20]

    Effect of confidence and explanation on accuracy and trust calibration in

    Zhang, Yunfeng and Liao, Q Vera and Bellamy, Rachel KE , booktitle=. Effect of confidence and explanation on accuracy and trust calibration in

  21. [21]

    Proceedings of the 28th International Conference on Intelligent User Interfaces , pages =

    Schemmer, Max and Kuehl, Niklas and Benz, Carina and Bartos, Andrea and Satzger, Gerhard , title =. Proceedings of the 28th International Conference on Intelligent User Interfaces , pages =. 2023 , publisher =

    2023

  22. [22]

    Finding consensus on trust in

    Starke, Georg and Gille, Felix and Termine, Alberto and Aquino, Yves Saint James and Chavarriaga, Ricardo and Ferrario, Andrea and Hastings, Janna and Jongsma, Karin and Kellmeyer, Philipp and Kulynych, Bogdan and others , journal=. Finding consensus on trust in. 2025 , publisher=

    2025

  23. [23]

    Experts or authorities?

    Ferrario, Andrea and Facchini, Alessandro and Termine, Alberto , journal=. Experts or authorities?. 2024 , publisher=

    2024

  24. [24]

    Explainable

    Miller, Tim , booktitle=. Explainable

  25. [25]

    A Well-Founded Solution to the Generality Problem , journal =

    Juan Comesa. A Well-Founded Solution to the Generality Problem , journal =. 2006 , volume =

    2006

  26. [26]

    Beebe , title =

    James R. Beebe , title =. No. 2004 , volume =

    2004

  27. [27]

    Goldman , title =

    Alvin I. Goldman , title =. 1986 , publisher =

    1986

  28. [28]

    Midwest Studies in Philosophy , year =

    Laurence BonJour , title =. Midwest Studies in Philosophy , year =

  29. [29]

    1995 , publisher=

    Cognition in the Wild , author=. 1995 , publisher=

    1995

  30. [30]

    2023 , publisher=

    AI as an epistemic technology , author=. 2023 , publisher=

    2023

  31. [31]

    2015 , publisher=

    Dimensions of integration in embedded and extended cognitive systems , author=. 2015 , publisher=

    2015

  32. [32]

    Analysis , volume=

    The extended mind , author=. Analysis , volume=. 1998 , publisher=

    1998

  33. [33]

    The Extended Mind , editor =

    Sutton, John , title =. The Extended Mind , editor =. 2010 , publisher =

    2010

  34. [34]

    Australasian Journal of Philosophy , year =

    Holton, Richard , title =. Australasian Journal of Philosophy , year =

  35. [35]

    Analyse & Kritik , year =

    Pettit, Philip , title =. Analyse & Kritik , year =

  36. [36]

    The Routledge Handbook of Trust and Philosophy , pages=

    Trust and reliance 1 , author=. The Routledge Handbook of Trust and Philosophy , pages=. 2020 , publisher=

    2020

  37. [37]

    2000 , publisher=

    Distributed cognition: Toward a new foundation for human-computer interaction research , author=. 2000 , publisher=

    2000

  38. [38]

    Minds and Machines , year =

    Andrea Ferrario and Alessandro Facchini and Alberto Termine , title =. Minds and Machines , year =

  39. [39]

    Philosophy & Technology , volume=

    Trust and Trustworthiness in AI , author=. Philosophy & Technology , volume=. 2025 , publisher=

    2025

  40. [40]

    Science and Engineering Ethics , volume=

    In AI we trust: ethics, artificial intelligence, and reliability , author=. Science and Engineering Ethics , volume=. 2020 , publisher=

    2020

  41. [41]

    Enhancing the reliability and accuracy of

    Dvijotham, Krishnamurthy and Winkens, Jim and Barsbey, Melih and Ghaisas, Sumedh and Stanforth, Robert and Pawlowski, Nick and Strachan, Patricia and Ahmed, Zahra and Azizi, Shekoofeh and Bachrach, Yoram and others , journal=. Enhancing the reliability and accuracy of. 2023 , publisher=

    2023

  42. [42]

    Who should I trust:

    Ma, Shuai and Lei, Ying and Wang, Xinru and Zheng, Chengbo and Shi, Chuhan and Yin, Ming and Ma, Xiaojuan , booktitle=. Who should I trust:

  43. [43]

    Advances in Neural Information Processing Systems , volume=

    Partial success in closing the gap between human and machine vision , author=. Advances in Neural Information Processing Systems , volume=

  44. [44]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Does explainable artificial intelligence improve human decision-making? , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  45. [45]

    Advancing human-

    Inkpen, Kori and Chappidi, Shreya and Mallari, Keri and Nushi, Besmira and Ramesh, Divya and Michelucci, Pietro and Mandava, Vani and Vep. Advancing human-. ACM Transactions on Computer-Human Interaction , volume=. 2023 , publisher=

    2023

  46. [46]

    European Journal for Philosophy of Science , volume=

    In defense of reliabilist epistemology of algorithms , author=. European Journal for Philosophy of Science , volume=. 2025 , publisher=

    2025

  47. [47]

    Grounds for trust:

    Dur. Grounds for trust:. 2018 , publisher=

    2018

  48. [48]

    How explainability contributes to trust in

    Ferrario, Andrea and Loi, Michele , booktitle=. How explainability contributes to trust in

  49. [49]

    Forensic Science International: Synergy , volume=

    From understanding to justifying: Computational reliabilism for AI-based forensic evidence evaluation , author=. Forensic Science International: Synergy , volume=. 2024 , publisher=

    2024

  50. [50]

    AI and Ethics , volume=

    What kind of trust does AI deserve, if any? , author=. AI and Ethics , volume=. 2023 , publisher=

    2023

  51. [51]

    2025 , publisher=

    Being pragmatic about reliance and trust in artificial intelligence , author=. 2025 , publisher=

    2025

  52. [52]

    Challenges for computational reliabilism in

    Alvarado, Ram. Challenges for computational reliabilism in. 2026 , publisher=

    2026

  53. [53]

    2024 , month = jun, institution =

    2024

  54. [54]

    The generality problem, statistical relevance and the tri-level hypothesis , author=. No. 2004 , publisher=

    2004

  55. [55]

    Philosophical Studies: An International Journal for Philosophy in the Analytic Tradition , volume=

    A well-founded solution to the generality problem , author=. Philosophical Studies: An International Journal for Philosophy in the Analytic Tradition , volume=. 2006 , publisher=

    2006

  56. [56]

    Midwest studies in philosophy , volume=

    Externalist theories of empirical knowledge , author=. Midwest studies in philosophy , volume=. 1980 , publisher=

    1980

  57. [57]

    1986 , publisher=

    Epistemology and cognition , author=. 1986 , publisher=

    1986

  58. [58]

    Philosophical Topics , volume=

    How to think about reliability , author=. Philosophical Topics , volume=. 1995 , publisher=

    1995

  59. [59]

    Philosophy Compass , volume=

    Reliability in machine learning , author=. Philosophy Compass , volume=. 2024 , publisher=

    2024

  60. [60]

    A taxonomy of human and ML strengths in decision-making to investigate human-

    Rastogi, Charvi and Leqi, Liu and Holstein, Kenneth and Heidari, Hoda , booktitle=. A taxonomy of human and ML strengths in decision-making to investigate human-

  61. [61]

    Toward a science of human--

    Gonzalez, Cleotilde and Donahue, Kate and Goldstein, Daniel G and Heidari, Hoda and Jalali, Mohammad S and Schelble, Beau and Singh, Aarti and Woolley, Anita Williams , journal=. Toward a science of human--. 2026 , publisher=

    2026

  62. [62]

    Advances in Neural Information Processing Systems , volume=

    Human expertise in algorithmic prediction , author=. Advances in Neural Information Processing Systems , volume=

  63. [63]

    Tamari, Dov , journal=. Probl

  64. [64]

    Epistemology gives a Future to Complementarity in Human-AI Interactions

    Ferrario, Andrea and Facchini, Alessandro and Dur. Epistemology gives a future to complementarity in human-. arXiv preprint arXiv:2601.09871 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  65. [65]

    Transactions of the American Mathematical Society , volume =

    Stasheff, James Dillon , title =. Transactions of the American Mathematical Society , volume =

  66. [66]

    On the notion of mean , author=

  67. [67]

    No-regret learning with unbounded losses:

    Neyman, Eric and Roughgarden, Tim , journal=. No-regret learning with unbounded losses:

  68. [68]

    Operations Research , volume=

    From proper scoring rules to max-min optimal forecast aggregation , author=. Operations Research , volume=. 2023 , publisher=

    2023

  69. [69]

    Fundamental properties of

    Furuichi, Shigeru and Yanagi, Kenjiro and Kuriyama, Ken , journal=. Fundamental properties of. 2004 , publisher=

    2004

  70. [70]

    Possible generalization of

    Tsallis, Constantino , journal=. Possible generalization of. 1988 , publisher=

    1988

  71. [71]

    USSR Computational Mathematics and Mathematical Physics , volume=

    The relaxation method of finding the common point of convex sets and its application to the solution of problems in convex programming , author=. USSR Computational Mathematics and Mathematical Physics , volume=. 1967 , publisher=

    1967

  72. [72]

    Proceedings of the fourth Berkeley symposium on mathematical statistics and probability, volume 1: contributions to the theory of statistics , volume=

    On measures of entropy and information , author=. Proceedings of the fourth Berkeley symposium on mathematical statistics and probability, volume 1: contributions to the theory of statistics , volume=. 1961 , organization=

    1961

  73. [73]

    1930 , organization=

    Nagumo, Mitio , booktitle=. 1930 , organization=

    1930

  74. [74]

    Forming Effective Human-

    Hemmer, Patrick and Schellhammer, Sebastian and V. Forming Effective Human-. Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence , pages =

  75. [75]

    International Conference on Artificial Intelligence and Statistics , pages=

    Learning to defer to multiple experts: Consistent surrogate losses, confidence calibration, and conformal ensembles , author=. International Conference on Artificial Intelligence and Statistics , pages=. 2023 , organization=

    2023

  76. [76]

    Conformal Set-based Human-

    Paat, Helbert and Shen, Guohao , journal=. Conformal Set-based Human-

  77. [77]

    A no free lunch theorem for human-

    Peng, Kenny and Garg, Nikhil and Kleinberg, Jon , booktitle=. A no free lunch theorem for human-

  78. [78]

    Journal of the Royal Statistical Society: Series B (Methodological) , volume=

    A general class of coefficients of divergence of one distribution from another , author=. Journal of the Royal Statistical Society: Series B (Methodological) , volume=. 1966 , publisher=

    1966

  79. [79]

    Yanofsky, Noson S , year=