11 Search Results for "Sharan, Vatsal"

Document
DOI: 10.4230/LIPIcs.FORC.2026.11
Limitations on Accurate, Trusted, Human-Level Reasoning

Authors: Rina Panigrahy and Vatsal Sharan

Published in: LIPIcs, Volume 368, 7th Symposium on Foundations of Responsible Computing (FORC 2026)

Abstract
We identify a fundamental incompatibility between the goals of accuracy, trust, and human-level reasoning in artificial intelligence (AI) systems, for strict mathematical definitions of these notions. We define accuracy of a system as the property that it never makes any false claims when it has the ability to abstain from making a prediction on any input, and trust as the assumption that the system is accurate. We define human-level reasoning as the property of an AI system always matching or exceeding human capability. Our core finding is that - for our formal definitions of these notions - an accurate and trusted AI system cannot be a human-level reasoning system: for such an accurate, trusted system there are task instances which are easily and provably solvable by a human but not by the system. Our proofs draw parallels to Gödel’s incompleteness theorems and Turing’s proof of the undecidability of the halting problem, and can be regarded as interpretations of Gödel’s and Turing’s results. Key to our proof is the formalization of the notion of trust, which allows us to separate the intrinsic property of a system (being accurate) from its epistemic status (being trusted).
Cite as

Rina Panigrahy and Vatsal Sharan. Limitations on Accurate, Trusted, Human-Level Reasoning. In 7th Symposium on Foundations of Responsible Computing (FORC 2026). Leibniz International Proceedings in Informatics (LIPIcs), Volume 368, pp. 11:1-11:21, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2026)

Copy BibTex To Clipboard

@InProceedings{panigrahy_et_al:LIPIcs.FORC.2026.11,
  author =	{Panigrahy, Rina and Sharan, Vatsal},
  title =	{{Limitations on Accurate, Trusted, Human-Level Reasoning}},
  booktitle =	{7th Symposium on Foundations of Responsible Computing (FORC 2026)},
  pages =	{11:1--11:21},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-419-2},
  ISSN =	{1868-8969},
  year =	{2026},
  volume =	{368},
  editor =	{Lin, Huijia (Rachel)},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.FORC.2026.11},
  URN =		{urn:nbn:de:0030-drops-259840},
  doi =		{10.4230/LIPIcs.FORC.2026.11},
  annote =	{Keywords: Accuracy, Safety, Trust, Complexity-theoretic limitations}
}
Document
DOI: 10.4230/LIPIcs.FORC.2026.14
Fair Multi-Agent Persuasion with Submodular Constraints

Authors: Yannan Bai, Kamesh Munagala, Yiheng Shen, and Davidson Zhu

Published in: LIPIcs, Volume 368, 7th Symposium on Foundations of Responsible Computing (FORC 2026)

Abstract
We study the problem of selection in the context of Bayesian persuasion. We are given multiple agents with hidden values (or quality scores), to whom resources must be allocated by a welfare-maximizing decision-maker. An intermediary with knowledge of the agents' values seeks to influence the outcome of the selection by designing informative signals and providing tie-breaking policies, so that when the receiver maximizes welfare over the resulting posteriors, the expected utilities of the agents (where utility is defined as allocation times value) achieve certain fairness properties. The fairness measure we will use is majorization, which simultaneously approximately maximizes all symmetric, monotone, concave functions of the utilities. We consider the general setting where the allocation to the agents needs to respect arbitrary submodular constraints, as given by the corresponding polymatroid. We present a signaling policy that achieves a logarithmically approximate majorized policy in this setting, assuming the receiver is a (1+ε) approximate welfare maximizer. The approximation ratio is almost best possible, and that significantly outperforms generic results that only yield linear approximations. A key component of our result is a structural characterization showing that the vector of agent utilities for a given signaling policy defines the base polytope of a different polymatroid, a result that may be of independent interest. In addition, we show that an arbitrarily good additive approximation to this vector can be produced in (weakly) polynomial time via the multiplicative weights update method.
Cite as

Yannan Bai, Kamesh Munagala, Yiheng Shen, and Davidson Zhu. Fair Multi-Agent Persuasion with Submodular Constraints. In 7th Symposium on Foundations of Responsible Computing (FORC 2026). Leibniz International Proceedings in Informatics (LIPIcs), Volume 368, pp. 14:1-14:22, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2026)

Copy BibTex To Clipboard

@InProceedings{bai_et_al:LIPIcs.FORC.2026.14,
  author =	{Bai, Yannan and Munagala, Kamesh and Shen, Yiheng and Zhu, Davidson},
  title =	{{Fair Multi-Agent Persuasion with Submodular Constraints}},
  booktitle =	{7th Symposium on Foundations of Responsible Computing (FORC 2026)},
  pages =	{14:1--14:22},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-419-2},
  ISSN =	{1868-8969},
  year =	{2026},
  volume =	{368},
  editor =	{Lin, Huijia (Rachel)},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.FORC.2026.14},
  URN =		{urn:nbn:de:0030-drops-259872},
  doi =		{10.4230/LIPIcs.FORC.2026.14},
  annote =	{Keywords: Bayesian Persuasion, Fair Division, Submodular Optimization}
}
Document
DOI: 10.4230/LIPIcs.FORC.2026.21
The Importance of Being Smoothly Calibrated

Authors: Parikshit Gopalan, Konstantinos Stavropoulos, Kunal Talwar, and Pranay Tankala

Published in: LIPIcs, Volume 368, 7th Symposium on Foundations of Responsible Computing (FORC 2026)

Abstract
Recent work has highlighted the centrality of smooth calibration [Sham Kakade and Dean Foster, 2008] as a robust measure of calibration error. We generalize, unify, and extend previous results on smooth calibration, both as a robust calibration measure, and as a step towards omniprediction, which enables predictions with low regret for downstream decision makers seeking to optimize some proper loss unknown to the predictor. - We present a new omniprediction guarantee for smoothly calibrated predictors, for the class of all bounded proper losses. We smooth the predictor by adding some noise to it, and compete against smoothed versions of any benchmark predictor on the space, where we add some noise to the predictor and then post-process it arbitrarily. The omniprediction error is bounded by the smooth calibration error of the predictor and the earth mover’s distance from the benchmark. We exhibit instances showing that this dependence cannot, in general, be improved. We show how this unifies and extends prior results [Dean P. Foster and Rakesh V. Vohra, 1998; Jason D. Hartline et al., 2025] on omniprediction from smooth calibration. - We present a crisp new characterization of smooth calibration in terms of the earth mover’s distance to the closest perfectly calibrated joint distribution of predictions and labels. This also yields a simpler proof of the relation to the lower distance to calibration from [Jaroslaw Blasiok et al., 2023]. - We use this to show that the upper distance to calibration cannot be estimated within a quadratic factor with sample complexity independent of the support size of the predictions. This is in contrast to the distance to calibration, where the corresponding problem was known to be information-theoretically impossible: no finite number of samples suffice [Jaroslaw Blasiok et al., 2023].
Cite as

Parikshit Gopalan, Konstantinos Stavropoulos, Kunal Talwar, and Pranay Tankala. The Importance of Being Smoothly Calibrated. In 7th Symposium on Foundations of Responsible Computing (FORC 2026). Leibniz International Proceedings in Informatics (LIPIcs), Volume 368, pp. 21:1-21:22, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2026)

Copy BibTex To Clipboard

@InProceedings{gopalan_et_al:LIPIcs.FORC.2026.21,
  author =	{Gopalan, Parikshit and Stavropoulos, Konstantinos and Talwar, Kunal and Tankala, Pranay},
  title =	{{The Importance of Being Smoothly Calibrated}},
  booktitle =	{7th Symposium on Foundations of Responsible Computing (FORC 2026)},
  pages =	{21:1--21:22},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-419-2},
  ISSN =	{1868-8969},
  year =	{2026},
  volume =	{368},
  editor =	{Lin, Huijia (Rachel)},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.FORC.2026.21},
  URN =		{urn:nbn:de:0030-drops-259945},
  doi =		{10.4230/LIPIcs.FORC.2026.21},
  annote =	{Keywords: Smooth Calibration, Omniprediction, Distance to Calibration}
}
Document
DOI: 10.4230/LIPIcs.ITCS.2026.48
Auditability and the Landscape of Distance to Multicalibration

Authors: Nathan Derhake, Siddartha Devic, Dutch Hansen, Kuan Liu, and Vatsal Sharan

Published in: LIPIcs, Volume 362, 17th Innovations in Theoretical Computer Science Conference (ITCS 2026)

Abstract
Calibration is a critical property for establishing the trustworthiness of predictors that provide uncertainty estimates. Multicalibration is a strengthening of calibration which requires that predictors be calibrated on a potentially overlapping collection of subsets of the domain. As multicalibration grows in popularity with practitioners, an essential question is: how do we measure how multicalibrated a predictor is? Błasiok et al. [Błasiok et al., 2023] considered this question for standard calibration by introducing the distance to calibration framework (dCE) to understand how calibration metrics relate to each other and the ground truth. Building on the dCE framework, we consider the auditability of the distance to multicalibration of a predictor f. We begin by considering what are perhaps the two most natural generalizations of dCE to multiple subgroups: worst group dCE (wdMC), and distance to multicalibration (dMC). Using wdMC and dMC as a guiding path, we argue that there are two essential properties of any multicalibration error metric: 1) the metric should capture how much f would need to be modified in order to be perfectly multicalibrated; and 2) the metric should be auditable in an information theoretic sense (i.e., with some finite sample complexity). We show that wdMC and dMC each fail to satisfy one of these two properties, and that similar barriers arise when considering the auditability of general distance to multigroup fairness notions (e.g. multiaccuracy or low-degree multicalibration). We then propose two (equivalent) multicalibration metrics which do satisfy these requirements: 1) a continuized variant of dMC; and 2) a distance to intersection multicalibration, which leans on intersectional fairness desiderata. Along the way, we shed light on the loss-landscape of distance to multicalibration and the geometry of the set of perfectly multicalibrated predictors. We also demonstrate that the loss surface of any metric which captures how much f would need to be modified to be perfectly multicalibrated often satisfies a local minima are global minima property. Our findings may have implications for the development of stronger multicalibration algorithms, as well as multicalibration auditing more generally.
Cite as

Nathan Derhake, Siddartha Devic, Dutch Hansen, Kuan Liu, and Vatsal Sharan. Auditability and the Landscape of Distance to Multicalibration. In 17th Innovations in Theoretical Computer Science Conference (ITCS 2026). Leibniz International Proceedings in Informatics (LIPIcs), Volume 362, pp. 48:1-48:23, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2026)

Copy BibTex To Clipboard

@InProceedings{derhake_et_al:LIPIcs.ITCS.2026.48,
  author =	{Derhake, Nathan and Devic, Siddartha and Hansen, Dutch and Liu, Kuan and Sharan, Vatsal},
  title =	{{Auditability and the Landscape of Distance to Multicalibration}},
  booktitle =	{17th Innovations in Theoretical Computer Science Conference (ITCS 2026)},
  pages =	{48:1--48:23},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-410-9},
  ISSN =	{1868-8969},
  year =	{2026},
  volume =	{362},
  editor =	{Saraf, Shubhangi},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.ITCS.2026.48},
  URN =		{urn:nbn:de:0030-drops-253351},
  doi =		{10.4230/LIPIcs.ITCS.2026.48},
  annote =	{Keywords: Multicalibration, Auditability, Fairness, Classification, Calibration}
}
Document
DOI: 10.4230/LIPIcs.CCC.2025.19
Characterizing the Distinguishability of Product Distributions Through Multicalibration

Authors: Cassandra Marcussen, Aaron Putterman, and Salil Vadhan

Published in: LIPIcs, Volume 339, 40th Computational Complexity Conference (CCC 2025)

Abstract
Given a sequence of samples x_1, … , x_k promised to be drawn from one of two distributions X₀, X₁, a well-studied problem in statistics is to decide which distribution the samples are from. Information theoretically, the maximum advantage in distinguishing the two distributions given k samples is captured by the total variation distance between X₀^{⊗k} and X₁^{⊗k}. However, when we restrict our attention to efficient distinguishers (i.e., small circuits) of these two distributions, exactly characterizing the ability to distinguish X₀^{⊗k} and X₁^{⊗k} is more involved and less understood. In this work, we give a general way to reduce bounds on the computational indistinguishability of X₀ and X₁ to bounds on the information-theoretic indistinguishability of some specific, related variables X̃₀ and X̃₁. As a consequence, we prove a new, tight characterization of the number of samples k needed to efficiently distinguish X₀^{⊗k} and X₁^{⊗k} with constant advantage as k = Θ(d_H^{-2}(X̃₀, X̃₁)), which is the inverse of the squared Hellinger distance d_H between two distributions X̃₀ and X̃₁ that are computationally indistinguishable from X₀ and X₁. Likewise, our framework can be used to re-derive a result of Halevi and Rabin (TCC 2008) and Geier (TCC 2022), proving nearly-tight bounds on how computational indistinguishability scales with the number of samples for arbitrary product distributions. At the heart of our work is the use of the Multicalibration Theorem (Hébert-Johnson, Kim, Reingold, Rothblum 2018) in a way inspired by recent work of Casacuberta, Dwork, and Vadhan (STOC 2024). Multicalibration allows us to relate the computational indistinguishability of X₀, X₁ to the statistical indistinguishability of X̃₀, X̃₁ (for lower bounds on k) and construct explicit circuits to distinguish between X̃₀, X̃₁ and consequently X₀, X₁ (for upper bounds on k).
Cite as

Cassandra Marcussen, Aaron Putterman, and Salil Vadhan. Characterizing the Distinguishability of Product Distributions Through Multicalibration. In 40th Computational Complexity Conference (CCC 2025). Leibniz International Proceedings in Informatics (LIPIcs), Volume 339, pp. 19:1-19:19, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2025)

Copy BibTex To Clipboard

@InProceedings{marcussen_et_al:LIPIcs.CCC.2025.19,
  author =	{Marcussen, Cassandra and Putterman, Aaron and Vadhan, Salil},
  title =	{{Characterizing the Distinguishability of Product Distributions Through Multicalibration}},
  booktitle =	{40th Computational Complexity Conference (CCC 2025)},
  pages =	{19:1--19:19},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-379-9},
  ISSN =	{1868-8969},
  year =	{2025},
  volume =	{339},
  editor =	{Srinivasan, Srikanth},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.CCC.2025.19},
  URN =		{urn:nbn:de:0030-drops-237130},
  doi =		{10.4230/LIPIcs.CCC.2025.19},
  annote =	{Keywords: Multicalibration, computational distinguishability}
}
Document
DOI: 10.4230/LIPIcs.FORC.2025.16
Smooth Calibration and Decision Making

Authors: Jason Hartline, Yifan Wu, and Yunran Yang

Published in: LIPIcs, Volume 329, 6th Symposium on Foundations of Responsible Computing (FORC 2025)

Abstract
Calibration requires predictor outputs to be consistent with their Bayesian posteriors. For machine learning predictors that do not distinguish between small perturbations, calibration errors are continuous in predictions, e.g. smooth calibration error [Foster and Hart, 2018], distance to calibration [Błasiok et al., 2023]. On the contrary, decision-makers who use predictions make optimal decisions discontinuously in probabilistic space, experiencing loss from miscalibration discontinuously. Calibration errors for decision-making are thus discontinuous, e.g., Expected Calibration Error [Foster and Vohra, 1997], and Calibration Decision Loss [Hu and Wu, 2024]. Thus, predictors with a low calibration error for machine learning may suffer a high calibration error for decision-making, i.e. they may not be trustworthy for decision-makers optimizing assuming their predictions are correct. It is natural to ask if post-processing a predictor with a low calibration error for machine learning is without loss to achieve a low calibration error for decision-making. In our paper, we show post-processing an online predictor with ε distance to calibration achieves O(√{ε}) ECE and CDL, which is asymptotically optimal. The post-processing algorithm adds noise to make predictions differentially private. The optimal bound from low distance to calibration predictors from post-processing is non-optimal compared with existing online calibration algorithms that directly optimize for ECE and CDL.
Cite as

Jason Hartline, Yifan Wu, and Yunran Yang. Smooth Calibration and Decision Making. In 6th Symposium on Foundations of Responsible Computing (FORC 2025). Leibniz International Proceedings in Informatics (LIPIcs), Volume 329, pp. 16:1-16:26, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2025)

Copy BibTex To Clipboard

@InProceedings{hartline_et_al:LIPIcs.FORC.2025.16,
  author =	{Hartline, Jason and Wu, Yifan and Yang, Yunran},
  title =	{{Smooth Calibration and Decision Making}},
  booktitle =	{6th Symposium on Foundations of Responsible Computing (FORC 2025)},
  pages =	{16:1--16:26},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-367-6},
  ISSN =	{1868-8969},
  year =	{2025},
  volume =	{329},
  editor =	{Bun, Mark},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.FORC.2025.16},
  URN =		{urn:nbn:de:0030-drops-231438},
  doi =		{10.4230/LIPIcs.FORC.2025.16},
  annote =	{Keywords: Calibration, calibration errors, decision making, differential privacy}
}
Document
DOI: 10.4230/LIPIcs.FORC.2025.14
Model Ensembling for Constrained Optimization

Authors: Ira Globus Harris, Varun Gupta, Michael Kearns, and Aaron Roth

Published in: LIPIcs, Volume 329, 6th Symposium on Foundations of Responsible Computing (FORC 2025)

Abstract
Many instances of decision making under objective uncertainty can be decomposed into two steps: predicting the objective function and then optimizing for the best feasible action under the estimate of the objective vector. We study the problem of ensembling models for optimization of uncertain linear objectives under arbitrary constraints. We imagine we are given a collection of predictive models mapping a feature space to multi-dimensional real-valued predictions, which form the coefficients of a linear objective that we would like to optimize. We give two ensembling methods that can provably result in transparent decisions that strictly improve on all initial policies. The first method operates in the "white box" setting in which we have access to the underlying prediction models and the second in the "black box" setting in which we only have access to the induced decisions (in the downstream optimization problem) of the constituent models, but not their underlying point predictions. They are transparent or trustworthy in the sense that the user can reliably predict long-term ensemble rewards even if the instance by instance predictions are imperfect.
Cite as

Ira Globus Harris, Varun Gupta, Michael Kearns, and Aaron Roth. Model Ensembling for Constrained Optimization. In 6th Symposium on Foundations of Responsible Computing (FORC 2025). Leibniz International Proceedings in Informatics (LIPIcs), Volume 329, pp. 14:1-14:17, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2025)

Copy BibTex To Clipboard

@InProceedings{globusharris_et_al:LIPIcs.FORC.2025.14,
  author =	{Globus Harris, Ira and Gupta, Varun and Kearns, Michael and Roth, Aaron},
  title =	{{Model Ensembling for Constrained Optimization}},
  booktitle =	{6th Symposium on Foundations of Responsible Computing (FORC 2025)},
  pages =	{14:1--14:17},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-367-6},
  ISSN =	{1868-8969},
  year =	{2025},
  volume =	{329},
  editor =	{Bun, Mark},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.FORC.2025.14},
  URN =		{urn:nbn:de:0030-drops-231412},
  doi =		{10.4230/LIPIcs.FORC.2025.14},
  annote =	{Keywords: model ensembling, trustworthy AI, decision-making under uncertainty}
}
Document
DOI: 10.4230/LIPIcs.FORC.2025.19
Mapping the Tradeoffs and Limitations of Algorithmic Fairness

Authors: Etam Benger and Katrina Ligett

Published in: LIPIcs, Volume 329, 6th Symposium on Foundations of Responsible Computing (FORC 2025)

Abstract
Sufficiency and separation are two fundamental criteria in classification fairness. For binary classifiers, these concepts correspond to subgroup calibration and equalized odds, respectively, and are known to be incompatible except in trivial cases. In this work, we explore a relaxation of these criteria based on f-divergences between distributions - essentially the same relaxation studied in the literature on approximate multicalibration - analyze their relationships, and derive implications for fair representations and downstream uses (post-processing) of representations. We show that when a protected attribute is determinable from features present in the data, the (relaxed) criteria of sufficiency and separation exhibit a tradeoff, forming a convex Pareto frontier. Moreover, we prove that when a protected attribute is not fully encoded in the data, achieving full sufficiency may be impossible. This finding not only strengthens the case against "fairness through unawareness" but also highlights an important caveat for work on (multi-)calibration.
Cite as

Etam Benger and Katrina Ligett. Mapping the Tradeoffs and Limitations of Algorithmic Fairness. In 6th Symposium on Foundations of Responsible Computing (FORC 2025). Leibniz International Proceedings in Informatics (LIPIcs), Volume 329, pp. 19:1-19:20, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2025)

Copy BibTex To Clipboard

@InProceedings{benger_et_al:LIPIcs.FORC.2025.19,
  author =	{Benger, Etam and Ligett, Katrina},
  title =	{{Mapping the Tradeoffs and Limitations of Algorithmic Fairness}},
  booktitle =	{6th Symposium on Foundations of Responsible Computing (FORC 2025)},
  pages =	{19:1--19:20},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-367-6},
  ISSN =	{1868-8969},
  year =	{2025},
  volume =	{329},
  editor =	{Bun, Mark},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.FORC.2025.19},
  URN =		{urn:nbn:de:0030-drops-231465},
  doi =		{10.4230/LIPIcs.FORC.2025.19},
  annote =	{Keywords: Algorithmic fairness, information theory, sufficiency-separation tradeoff}
}
Document
DOI: 10.4230/LIPIcs.FORC.2025.22
When Does a Predictor Know Its Own Loss?

Authors: Aravind Gollakota, Parikshit Gopalan, Aayush Karan, Charlotte Peale, and Udi Wieder

Published in: LIPIcs, Volume 329, 6th Symposium on Foundations of Responsible Computing (FORC 2025)

Abstract
Given a predictor and a loss function, how well can we predict the loss that the predictor will incur on an input? This is the problem of loss prediction, a key computational task associated with uncertainty estimation for a predictor. In a classification setting, a predictor will typically predict a distribution over labels and hence have its own estimate of the loss that it will incur, given by the entropy of the predicted distribution. Should we trust this estimate? In other words, when does the predictor know what it knows and what it does not know? In this work we study the theoretical foundations of loss prediction. Our main contribution is to establish tight connections between nontrivial loss prediction and certain forms of multicalibration [Ursula Hébert-Johnson et al., 2018], a multigroup fairness notion that asks for calibrated predictions across computationally identifiable subgroups. Formally, we show that a loss predictor that is able to improve on the self-estimate of a predictor yields a witness to a failure of multicalibration, and vice versa. This has the implication that nontrivial loss prediction is in effect no easier or harder than auditing for multicalibration. We support our theoretical results with experiments that show a robust positive correlation between the multicalibration error of a predictor and the efficacy of training a loss predictor.
Cite as

Aravind Gollakota, Parikshit Gopalan, Aayush Karan, Charlotte Peale, and Udi Wieder. When Does a Predictor Know Its Own Loss?. In 6th Symposium on Foundations of Responsible Computing (FORC 2025). Leibniz International Proceedings in Informatics (LIPIcs), Volume 329, pp. 22:1-22:22, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2025)

Copy BibTex To Clipboard

@InProceedings{gollakota_et_al:LIPIcs.FORC.2025.22,
  author =	{Gollakota, Aravind and Gopalan, Parikshit and Karan, Aayush and Peale, Charlotte and Wieder, Udi},
  title =	{{When Does a Predictor Know Its Own Loss?}},
  booktitle =	{6th Symposium on Foundations of Responsible Computing (FORC 2025)},
  pages =	{22:1--22:22},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-367-6},
  ISSN =	{1868-8969},
  year =	{2025},
  volume =	{329},
  editor =	{Bun, Mark},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.FORC.2025.22},
  URN =		{urn:nbn:de:0030-drops-231490},
  doi =		{10.4230/LIPIcs.FORC.2025.22},
  annote =	{Keywords: loss prediction, multicalibration, active learning, algorithmic fairness, calibration, predictive uncertainty, uncertainty estimation, machine learning theory}
}
Document
DOI: 10.4230/LIPIcs.FORC.2025.7
Kernel Multiaccuracy

Authors: Carol Xuan Long, Wael Alghamdi, Alexander Glynn, Yixuan Wu, and Flavio P. Calmon

Published in: LIPIcs, Volume 329, 6th Symposium on Foundations of Responsible Computing (FORC 2025)

Abstract
Predefined demographic groups often overlook the subpopulations most impacted by model errors, leading to a growing emphasis on data-driven methods that pinpoint where models underperform. The emerging field of multi-group fairness addresses this by ensuring models perform well across a wide range of group-defining functions, rather than relying on fixed demographic categories. We demonstrate that recently introduced notions of multi-group fairness can be equivalently formulated as integral probability metrics (IPM). IPMs are the common information-theoretic tool that underlie definitions such as multiaccuracy, multicalibration, and outcome indistinguishably. For multiaccuracy, this connection leads to a simple, yet powerful procedure for achieving multiaccuracy with respect to an infinite-dimensional class of functions defined by a reproducing kernel Hilbert space (RKHS): first perform a kernel regression of a model’s errors, then subtract the resulting function from a model’s predictions. We combine these results to develop a post-processing method that improves multiaccuracy with respect to bounded-norm functions in an RKHS, enjoys provable performance guarantees, and, in binary classification benchmarks, achieves favorable multiaccuracy relative to competing methods.
Cite as

Carol Xuan Long, Wael Alghamdi, Alexander Glynn, Yixuan Wu, and Flavio P. Calmon. Kernel Multiaccuracy. In 6th Symposium on Foundations of Responsible Computing (FORC 2025). Leibniz International Proceedings in Informatics (LIPIcs), Volume 329, pp. 7:1-7:23, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2025)

Copy BibTex To Clipboard

@InProceedings{long_et_al:LIPIcs.FORC.2025.7,
  author =	{Long, Carol Xuan and Alghamdi, Wael and Glynn, Alexander and Wu, Yixuan and Calmon, Flavio P.},
  title =	{{Kernel Multiaccuracy}},
  booktitle =	{6th Symposium on Foundations of Responsible Computing (FORC 2025)},
  pages =	{7:1--7:23},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-367-6},
  ISSN =	{1868-8969},
  year =	{2025},
  volume =	{329},
  editor =	{Bun, Mark},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.FORC.2025.7},
  URN =		{urn:nbn:de:0030-drops-231341},
  doi =		{10.4230/LIPIcs.FORC.2025.7},
  annote =	{Keywords: algorithmic fairness, integral probability metrics, information theory}
}
Document
DOI: 10.4230/LIPIcs.ITCS.2022.79
Omnipredictors

Authors: Parikshit Gopalan, Adam Tauman Kalai, Omer Reingold, Vatsal Sharan, and Udi Wieder

Published in: LIPIcs, Volume 215, 13th Innovations in Theoretical Computer Science Conference (ITCS 2022)

Abstract
Loss minimization is a dominant paradigm in machine learning, where a predictor is trained to minimize some loss function that depends on an uncertain event (e.g., "will it rain tomorrow?"). Different loss functions imply different learning algorithms and, at times, very different predictors. While widespread and appealing, a clear drawback of this approach is that the loss function may not be known at the time of learning, requiring the algorithm to use a best-guess loss function. Alternatively, the same classifier may be used to inform multiple decisions, which correspond to multiple loss functions, requiring multiple learning algorithms to be run on the same data. We suggest a rigorous new paradigm for loss minimization in machine learning where the loss function can be ignored at the time of learning and only be taken into account when deciding an action. We introduce the notion of an (L,𝒞)-omnipredictor, which could be used to optimize any loss in a family L. Once the loss function is set, the outputs of the predictor can be post-processed (a simple univariate data-independent transformation of individual predictions) to do well compared with any hypothesis from the class C. The post processing is essentially what one would perform if the outputs of the predictor were true probabilities of the uncertain events. In a sense, omnipredictors extract all the predictive power from the class 𝒞, irrespective of the loss function in L. We show that such "loss-oblivious" learning is feasible through a connection to multicalibration, a notion introduced in the context of algorithmic fairness. A multicalibrated predictor doesn’t aim to minimize some loss function, but rather to make calibrated predictions, even when conditioned on inputs lying in certain sets c belonging to a family 𝒞 which is weakly learnable. We show that a 𝒞-multicalibrated predictor is also an (L,𝒞)-omnipredictor, where L contains all convex loss functions with some mild Lipschitz conditions. The predictors are even omnipredictors with respect to sparse linear combinations of functions in 𝒞. As a corollary, we deduce that distribution-specific weak agnostic learning is complete for a large class of loss minimization tasks. In addition, we show how multicalibration can be viewed as a solution concept for agnostic boosting, shedding new light on past results. Finally, we transfer our insights back to the context of algorithmic fairness by providing omnipredictors for multi-group loss minimization.
Cite as

Parikshit Gopalan, Adam Tauman Kalai, Omer Reingold, Vatsal Sharan, and Udi Wieder. Omnipredictors. In 13th Innovations in Theoretical Computer Science Conference (ITCS 2022). Leibniz International Proceedings in Informatics (LIPIcs), Volume 215, pp. 79:1-79:21, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2022)

Copy BibTex To Clipboard

@InProceedings{gopalan_et_al:LIPIcs.ITCS.2022.79,
  author =	{Gopalan, Parikshit and Kalai, Adam Tauman and Reingold, Omer and Sharan, Vatsal and Wieder, Udi},
  title =	{{Omnipredictors}},
  booktitle =	{13th Innovations in Theoretical Computer Science Conference (ITCS 2022)},
  pages =	{79:1--79:21},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-217-4},
  ISSN =	{1868-8969},
  year =	{2022},
  volume =	{215},
  editor =	{Braverman, Mark},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.ITCS.2022.79},
  URN =		{urn:nbn:de:0030-drops-156755},
  doi =		{10.4230/LIPIcs.ITCS.2022.79},
  annote =	{Keywords: Loss-minimzation, multi-group fairness, agnostic learning, boosting}
}
  • Refine by Type
  • 11 Document/PDF
  • 9 Document/HTML

  • Refine by Publication Year
  • 4 2026
  • 6 2025
  • 1 2022

  • Refine by Author
  • 3 Gopalan, Parikshit
  • 3 Sharan, Vatsal
  • 2 Wieder, Udi
  • 1 Alghamdi, Wael
  • 1 Bai, Yannan
  • Show More...

  • Refine by Series/Journal
  • 11 LIPIcs

  • Refine by Classification
  • 5 Theory of computation → Machine learning theory
  • 2 Mathematics of computing → Information theory
  • 2 Theory of computation → Algorithmic game theory and mechanism design
  • 1 Computing methodologies → Learning settings
  • 1 Computing methodologies → Machine learning
  • Show More...

  • Refine by Keyword
  • 2 Calibration
  • 2 Multicalibration
  • 2 algorithmic fairness
  • 2 information theory
  • 1 Accuracy
  • Show More...

Any Issues?
X

Feedback on the Current Page

CAPTCHA

Thanks for your feedback!

Feedback submitted to Dagstuhl Publishing

Could not send message

Please try again later or send an E-mail
© 2023-2026 Schloss Dagstuhl – LZI GmbH Schloss Dagstuhl – LZI GmbH About DROPS Imprint Privacy Contact