arXiv subjects

Cyrus Cousins

Publications and source records attributed to Cyrus Cousins.

At least 19 recordsLinked to original sources

arXiv2026

Position: We Need Practical AI Alignment Methods to Mirror Human Reasoning

AI systems are increasingly employed as decision aids, decision delegates, or autonomous decision-makers. This position paper argues that in many settings, particularly high-stakes decision-making, we need accurate cognitively-aligned AI systems that reason similarly to their users, and faithfully communicate their reasoning. We review evidence that cognitive alignment improves understandability and trustworthiness, and provide new survey data showing that many users find cognitive alignment "essential" when an AI's rationale for a judgment or action is important to them. We outline the gaps between existing alignment methods and what is needed to achieve cognitive alignment, and present a research agenda to address these gaps. We argue that cognitive misalignment represents a likely impediment to AI adoption in many envisioned applications, and that addressing it is important for creating AI systems on which users are both willing and justified to rely.

cs.AI↗

arXiv2025

Moral Change or Noise? On Problems of Aligning AI With Temporally Unstable Human Feedback

Alignment methods in moral domains seek to elicit moral preferences of human stakeholders and incorporate them into AI. This presupposes moral preferences as static targets, but such preferences often evolve over time. Proper alignment of AI to dynamic human preferences should ideally account for "legitimate" changes to moral reasoning, while ignoring changes related to attention deficits, cognitive biases, or other arbitrary factors. However, common AI alignment approaches largely neglect temporal changes in preferences, posing serious challenges to proper alignment, especially in high-stakes applications of AI, e.g., in healthcare domains, where misalignment can jeopardize the trustworthiness of the system and yield serious individual and societal harms. This work investigates the extent to which people's moral preferences change over time, and the impact of such changes on AI alignment. Our study is grounded in the kidney allocation domain, where we elicit responses to pairwise comparisons of hypothetical kidney transplant patients from over 400 participants across 3-5 sessions. We find that, on average, participants change their response to the same scenario presented at different times around 6-20% of the time (exhibiting "response instability"). Additionally, we observe significant shifts in several participants' retrofitted decision-making models over time (capturing "model instability"). The predictive performance of simple AI models decreases as a function of both response and model instability. Moreover, predictive performance diminishes over time, highlighting the importance of accounting for temporal changes in preferences during training. These findings raise fundamental normative and technical challenges relevant to AI alignment, highlighting the need to better understand the object of alignment (what to align to) when user preferences change significantly over time.

cs.HC↗

arXiv2025

Towards Cognitively-Faithful Decision-Making Models to Improve AI Alignment

Recent AI trends seek to align AI models to learned human-centric objectives, such as personal preferences, utility, or societal values. Using standard preference elicitation methods, researchers and practitioners build models of human decisions and judgments, to which AI models are aligned. However, standard elicitation methods often fail to capture the cognitive processes behind human decision making, such as heuristics or simplifying structured thought patterns. To address this failure, we take an axiomatic approach to learning cognitively faithful decision processes from pairwise comparisons. Building on the literature analyzing cognitive processes that shape human decision-making, we derive a model class in which features are first processed with learned rules, then aggregated via a fixed rule, such as the Bradley-Terry rule, to produce a decision. This structured processing of information ensures that such models are realistic and feasible candidates to represent underlying human decision-making processes. We demonstrate the efficacy of this modeling approach by learning interpretable models of human decision making in a kidney allocation task, and show that our proposed models match or surpass the accuracy of prior models of human pairwise decision-making.

cs.LG↗

arXiv2025

Deploying Fair and Efficient Course Allocation Mechanisms

Universities regularly face the challenging task of assigning classes to thousands of students while considering their preferences, along with course schedules and capacities. Ensuring the effectiveness and fairness of course allocation mechanisms is crucial to guaranteeing student satisfaction and optimizing resource utilization. We approach this problem from an economic perspective, using formal justice criteria to evaluate different algorithmic frameworks. To evaluate our frameworks, we conduct a large scale survey of university students at University of Massachusetts Amherst, collecting over 1,000 student preferences. This is, to our knowledge, the largest publicly available dataset of student preferences. We develop software for generating synthetic student preferences over courses, and implement four allocation algorithms: the serial dictatorship algorithm used by University of Massachusetts Amherst; Round Robin; an Integer Linear Program; and the Yankee Swap algorithm. We propose improvements to the Yankee Swap framework to handle scenarios with item multiplicities. Through experimentation with the Fall 2024 Computer Science course schedule at University of Massachusetts Amherst, we evaluate each algorithm's performance relative to standard justice criteria, providing insights into fair course allocation in large university settings.

cs.GT↗

arXiv2024

Fair and Welfare-Efficient Constrained Multi-matchings under Uncertainty

We study fair allocation of constrained resources, where a market designer optimizes overall welfare while maintaining group fairness. In many large-scale settings, utilities are not known in advance, but are instead observed after realizing the allocation. We therefore estimate agent utilities using machine learning. Optimizing over estimates requires trading-off between mean utilities and their predictive variances. We discuss these trade-offs under two paradigms for preference modeling -- in the stochastic optimization regime, the market designer has access to a probability distribution over utilities, and in the robust optimization regime they have access to an uncertainty set containing the true utilities with high probability. We discuss utilitarian and egalitarian welfare objectives, and we explore how to optimize for them under stochastic and robust paradigms. We demonstrate the efficacy of our approaches on three publicly available conference reviewer assignment datasets. The approaches presented enable scalable constrained resource allocation under uncertainty for many combinations of objectives and preference models.

cs.GT↗

arXiv2024

Algorithms and Analysis for Optimizing Robust Objectives in Fair Machine Learning

The original position or veil of ignorance argument of John Rawls, perhaps the most famous argument for egalitarianism, states that our concept of fairness, justice, or welfare should be decided from behind a veil of ignorance, and thus must consider everyone impartially (invariant to our identity). This can be posed as a zero-sum game, where a Daemon constructs a world, and an adversarial Angel then places the Daemon into the world. This game incentivizes the Daemon to maximize the minimum utility over all people (i.e., to maximize egalitarian welfare). In some sense, this is the most extreme form of risk aversion or robustness, and we show that by weakening the Angel, milder robust objectives arise, which we argue are effective robust proxies for fair learning or allocation tasks. In particular, the utilitarian, Gini, and power-mean welfare concepts arise from special cases of the adversarial game, which has philosophical implications for the understanding of each of these concepts. We also motivate a new fairness concept that essentially fuses the nonlinearity of the power-mean with the piecewise nature of the Gini class. Then, exploiting the relationship between fairness and robustness, we show that these robust fairness concepts can all be efficiently optimized under mild conditions via standard maximin optimization techniques. Finally, we show that such methods apply in machine learning contexts, and moreover we show generalization bounds for robust fair machine learning tasks.

cs.GT↗

arXiv2024

Percentile Criterion Optimization in Offline Reinforcement Learning

In reinforcement learning, robust policies for high-stakes decision-making problems with limited data are usually computed by optimizing the \emph{percentile criterion}. The percentile criterion is approximately solved by constructing an \emph{ambiguity set} that contains the true model with high probability and optimizing the policy for the worst model in the set. Since the percentile criterion is non-convex, constructing ambiguity sets is often challenging. Existing work uses \emph{Bayesian credible regions} as ambiguity sets, but they are often unnecessarily large and result in learning overly conservative policies. To overcome these shortcomings, we propose a novel Value-at-Risk based dynamic programming algorithm to optimize the percentile criterion without explicitly constructing any ambiguity sets. Our theoretical and empirical results show that our algorithm implicitly constructs much smaller ambiguity sets and learns less conservative robust policies.

cs.LG↗

arXiv2024

To Pool or Not To Pool: Analyzing the Regularizing Effects of Group-Fair Training on Shared Models

In fair machine learning, one source of performance disparities between groups is over-fitting to groups with relatively few training samples. We derive group-specific bounds on the generalization error of welfare-centric fair machine learning that benefit from the larger sample size of the majority group. We do this by considering group-specific Rademacher averages over a restricted hypothesis class, which contains the family of models likely to perform well with respect to a fair learning objective (e.g., a power-mean). Our simulations demonstrate these bounds improve over a naive method, as expected by theory, with particularly significant improvement for smaller group sizes.

cs.LG↗

arXiv2023

Towards AI Transparency and Accountability: A Global Framework for Exchanging Information on AI Systems

We propose that future AI transparency and accountability regulations are based on an open global standard for exchanging information about AI systems, which allows co-existence of potentially conflicting local regulations. Then, we discuss key components of a lightweight and effective AI transparency and/or accountability regulation. To prevent overregulation, the proposed approach encourages collaboration between regulators and industry to create a scalable and cost-efficient mutually beneficial solution. This includes using automated assessments and benchmarks with results transparently communicated through AI cards in an open AI register to facilitate meaningful public comparisons of competing AI systems. Such AI cards should report standardized measures tailored to the specific high-risk applications of AI systems and could be used for conformity assessments under AI transparency and accountability policies such as the European Union's AI Act.

cs.CY↗

arXiv2023

The Good, the Bad and the Submodular: Fairly Allocating Mixed Manna Under Order-Neutral Submodular Preferences

We study the problem of fairly allocating indivisible goods (positively valued items) and chores (negatively valued items) among agents with decreasing marginal utilities over items. Our focus is on instances where all the agents have simple preferences; specifically, we assume the marginal value of an item can be either $-1$, $0$ or some positive integer $c$. Under this assumption, we present an efficient algorithm to compute leximin allocations for a broad class of valuation functions we call order-neutral submodular valuations. Order-neutral submodular valuations strictly contain the well-studied class of additive valuations but are a strict subset of the class of submodular valuations. We show that these leximin allocations are Lorenz dominating and approximately proportional. We also show that, under further restriction to additive valuations, these leximin allocations are approximately envy-free and guarantee each agent their maxmin share. We complement this algorithmic result with a lower bound showing that the problem of computing leximin allocations is NP-hard when $c$ is a rational number.

cs.GT↗

arXiv2023

Dividing Good and Better Items Among Agents with Bivalued Submodular Valuations

We study the problem of fairly allocating a set of indivisible goods among agents with {\em bivalued submodular valuations} -- each good provides a marginal gain of either $a$ or $b$ ($a < b$) and goods have decreasing marginal gains. This is a natural generalization of two well-studied valuation classes -- bivalued additive valuations and binary submodular valuations. We present a simple sequential algorithmic framework, based on the recently introduced Yankee Swap mechanism, that can be adapted to compute a variety of solution concepts, including max Nash welfare (MNW), leximin and $p$-mean welfare maximizing allocations when $a$ divides $b$. This result is complemented by an existing result on the computational intractability of MNW and leximin allocations when $a$ does not divide $b$. We show that MNW and leximin allocations guarantee each agent at least $\frac25$ and $\frac{a}{b+2a}$ of their maximin share, respectively, when $a$ divides $b$. We also show that neither the leximin nor the MNW allocation is guaranteed to be envy free up to one good (EF1). This is surprising since for the simpler classes of bivalued additive valuations and binary submodular valuations, MNW allocations are known to be envy free up to any good (EFX).

cs.GT↗

arXiv2023

Into the Unknown: Assigning Reviewers to Papers with Uncertain Affinities

A successful peer review process requires that qualified and interested reviewers are assigned to each paper. Most automated reviewer assignment approaches estimate a real-valued affinity score for each paper-reviewer pair that acts as a proxy for the quality of the match, and then assign reviewers to maximize the sum of affinity scores. Most affinity score estimation methods are inherently noisy: reviewers can only bid on a small number of papers, and textual similarity models and subject-area matching are inherently noisy estimators. Current paper assignment systems are not designed to rigorously handle noise in the peer-review matching market. In this work, we assume paper-reviewer affinity scores are located in or near a high-probability region called an uncertainty set. We maximize the worst-case sum of scores for a reviewer assignment over the uncertainty set. We demonstrate how to robustly maximize the sum of scores across various classes of uncertainty sets, avoiding potentially serious mistakes in assignment. Our general approach can be used to integrate a large variety of paper-reviewer affinity models into reviewer assignment, opening the door to a much more robust peer review process.

cs.GT↗

arXiv2022

Regret Pruning for Learning Equilibria in Simulation-Based Games

In recent years, empirical game-theoretic analysis (EGTA) has emerged as a powerful tool for analyzing games in which an exact specification of the utilities is unavailable. Instead, EGTA assumes access to an oracle, i.e., a simulator, which can generate unbiased noisy samples of players' unknown utilities, given a strategy profile. Utilities can thus be empirically estimated by repeatedly querying the simulator. Recently, various progressive sampling (PS) algorithms have been proposed, which aim to produce PAC-style learning guarantees (e.g., approximate Nash equilibria with high probability) using as few simulator queries as possible. One recent work introduces a pruning technique called regret-pruning which further minimizes the number of simulator queries placed in PS algorithms which aim to learn pure Nash equilibria. In this paper, we address a serious limitation of this original regret pruning approach -- it is only able to guarantee that true pure Nash equilibria of the empirical game are approximate equilibria of the true game, and is unable to provide any strong guarantees regarding the efficacy of approximate pure Nash equilibria. This is a significant limitation since in many games, pure Nash equilibria are computationally intractable to find, or even non-existent. We introduce three novel regret pruning variations. The first two variations generalize the original regret pruning approach to yield guarantees for approximate pure Nash equilibria of the empirical game. The third variation goes further to even yield strong guarantees for all approximate mixed Nash equilibria of the empirical game. We use these regret pruning variations to design two novel progressive sampling algorithms, PS-REG+ and PS-REG-M, which experimentally outperform the previous state-of-the-art algorithms for learning pure and mixed equilibria, respectively, of simulation-based games.

cs.GT↗

arXiv2022

Computational and Data Requirements for Learning Generic Properties of Simulation-Based Games

Empirical game-theoretic analysis (EGTA) is primarily focused on learning the equilibria of simulation-based games. Recent approaches have tackled this problem by learning a uniform approximation of the game's utilities, and then applying precision-recall theorems: i.e., all equilibria of the true game are approximate equilibria in the estimated game, and vice-versa. In this work, we generalize this approach to all game properties that are well behaved (i.e., Lipschitz continuous in utilities), including regret (which defines Nash and correlated equilibria), adversarial values, and power-mean and Gini social welfare. Further, we introduce a novel algorithm -- progressive sampling with pruning (PSP) -- for learning a uniform approximation and thus any well-behaved property of a game, which prunes strategy profiles once the corresponding players' utilities are well-estimated, and we analyze its data and query complexities in terms of the a priori unknown utility variances. We experiment with our algorithm extensively, showing that 1) the number of queries that PSP saves is highly sensitive to the utility variance distribution, and 2) PSP consistently outperforms theoretical upper bounds, achieving significantly lower query complexities than natural baselines. We conclude with experiments that uncover some of the remaining difficulties with learning properties of simulation-based games, in spite of recent advances in statistical EGTA methodology, including those developed herein.

cs.GT↗

arXiv2021

Fast Doubly-Adaptive MCMC to Estimate the Gibbs Partition Function with Weak Mixing Time Bounds

We present a novel method for reducing the computational complexity of rigorously estimating the partition functions (normalizing constants) of Gibbs (Boltzmann) distributions, which arise ubiquitously in probabilistic graphical models. A major obstacle to practical applications of Gibbs distributions is the need to estimate their partition functions. The state of the art in addressing this problem is multi-stage algorithms, which consist of a cooling schedule, and a mean estimator in each step of the schedule. While the cooling schedule in these algorithms is adaptive, the mean estimation computations use MCMC as a black-box to draw approximate samples. We develop a doubly adaptive approach, combining the adaptive cooling schedule with an adaptive MCMC mean estimator, whose number of Markov chain steps adapts dynamically to the underlying chain. Through rigorous theoretical analysis, we prove that our method outperforms the state of the art algorithms in several factors: (1) The computational complexity of our method is smaller; (2) Our method is less sensitive to loose bounds on mixing times, an inherent component in these algorithms; and (3) The improvement obtained by our method is particularly significant in the most challenging regime of high-precision estimation. We demonstrate the advantage of our method in experiments run on classic factor graphs, such as voting models and Ising models.

stat.ML↗

arXiv2021

Making mean-estimation more efficient using an MCMC trace variance approach: DynaMITE

We introduce a novel statistical measure for MCMC-mean estimation, the inter-trace variance ${\rm trv}^{(τ_{rel})}({\cal M},f)$, which depends on a Markov chain ${\cal M}$ and a function $f:S\to [a,b]$. The inter-trace variance can be efficiently estimated from observed data and leads to a more efficient MCMC-mean estimator. Prior MCMC mean-estimators receive, as input, upper-bounds on $τ_{mix}$ or $τ_{rel}$, and often also the stationary variance, and their performance is highly dependent to the sharpness of these bounds. In contrast, we introduce DynaMITE, which dynamically adjusts the sample size, it is less sensitive to the looseness of input upper-bounds on $τ_{rel}$, and requires no bound on $v_π$. Receiving only an upper-bound ${\cal T}_{rel}$ on $τ_{rel}$, DynaMITE estimates the mean of $f$ in $\tilde{\cal{O}}\bigl(\smash{\frac{{\cal T}_{rel} R}{\varepsilon}}+\frac{τ_{rel}\cdot {\rm trv}^{(τ{rel})}}{\varepsilon^{2}}\bigr)$ steps, without a priori bounds on the stationary variance $v_π$ or the inter-trace variance ${\rm trv}^{(τrel)}$. Thus we depend minimally on the tightness of ${\cal T}_{mix}$, as the complexity is dominated by $τ_{rel}\rm{trv}^{(τ{rel})}$ as $\varepsilon \to 0$. Note that bounding $τ_{\rm rel}$ is known to be prohibitively difficult, however, DynaMITE is able to reduce its principal dependence on ${\cal T}_{rel}$ to $τ_{rel}$, simply by exploiting properties of the inter-trace variance. To compare our method to known variance-aware bounds, we show ${\rm trv}^{(τ{rel})}({\cal M},f) \leq v_π$. Furthermore, we show when $f$'s image is distributed (semi)symmetrically on ${\cal M}$'s traces, we have ${\rm trv}^{({τ{rel}})}({\cal M},f)=o(v_π(f))$, thus DynaMITE outperforms prior methods in these cases.

cs.DS↗

arXiv2021

An Axiomatic Theory of Provably-Fair Welfare-Centric Machine Learning

We address an inherent difficulty in welfare-theoretic fair machine learning by proposing an equivalently axiomatically-justified alternative and studying the resulting computational and statistical learning questions. Welfare metrics quantify overall wellbeing across a population of one or more groups, and welfare-based objectives and constraints have recently been proposed to incentivize fair machine learning methods to produce satisfactory solutions that consider the diverse needs of multiple groups. Unfortunately, many machine-learning problems are more naturally cast as loss minimization tasks, rather than utility maximization, which complicates direct application of welfare-centric methods to fair machine learning. In this work, we define a complementary measure, termed malfare, measuring overall societal harm (rather than wellbeing), with axiomatic justification via the standard axioms of cardinal welfare. We then cast fair machine learning as malfare minimization over the risk values (expected losses) of each group. Surprisingly, the axioms of cardinal welfare (malfare) dictate that this is not equivalent to simply defining utility as negative loss. Building upon these concepts, we define fair-PAC (FPAC) learning, where an FPAC learner is an algorithm that learns an $\varepsilon$-$δ$ malfare-optimal model with bounded sample complexity, for any data distribution, and for any (axiomatically justified) malfare concept. Finally, we show broad conditions under which, with appropriate modifications, standard PAC-learners may be converted to FPAC learners. This places FPAC learning on firm theoretical ground, as it yields statistical and computational efficiency guarantees for many well-studied machine-learning models, and is also practically relevant, as it democratizes fair ML by providing concrete training algorithms and rigorous generalization guarantees for these models

cs.LG↗

arXiv2021

Learning Competitive Equilibria in Noisy Combinatorial Markets

We present a methodology to robustly estimate the competitive equilibria (CE) of combinatorial markets under the assumption that buyers do not know their precise valuations for bundles of goods, but instead can only provide noisy estimates. We first show tight lower- and upper-bounds on the buyers' utility loss, and hence the set of CE, given a uniform approximation of one market by another. We then develop a learning framework for our setup, and present two probably-approximately-correct algorithms for learning CE, i.e., producing uniform approximations that preserve CE, with finite-sample guarantees. The first is a baseline that uses Hoeffding's inequality to produce a uniform approximation of buyers' valuations with high probability. The second leverages a connection between the first welfare theorem of economics and uniform approximations to adaptively prune value queries when it determines that they are provably not part of a CE. We experiment with our algorithms and find that the pruning algorithm achieves better estimates than the baseline with far fewer samples.

cs.GT↗