Searcharxiv⌕ Search

arXiv subjects

Xiaowu Dai

Publications and source records attributed to Xiaowu Dai.

At least 19 recordsLinked to original sources

Exploratory creative pre-ranking: Hierarchical empirical-Bayes Thompson sampling for real-time creative optimization

Dynamic creative optimization decides which creative of an ad to render for each request. Recent work places that choice before ad ranking, but relies on deterministic point predictors with no calibrated uncertainty. AI-generated content has made creatives nearly free, while feedback per creative remains sparse and selection faces a tight latency budget. We introduce a creative pre-ranker that scores every recalled creative and shortlists each campaign's pool before the per-user ranker runs. Creatives sit in a hierarchy of audience segment, advertised item, campaign and creative, so a cold creative falls back on what its parent levels have earned, and how much it borrows from its parents is set in closed form from historical logs, with one parameter per level and nothing tuned. To stay within the latency budget, the pre-ranker predicts for an audience segment rather than an individual user, from per-attribute counters. It selects by a single Thompson draw per creative, balancing exploration and exploitation so that a new creative still has a chance to be learned. Replaying Roblox Sponsored-ads traffic over campaigns with competing creatives, the shortlist lifts predicted play-through rate by 8.14% against 2.80% for an equally sized random shortlist, forwards more than 40% fewer creatives than the full pool, and adds only 0.2% to serving time. After an A/B test on live traffic, the method now serves in production at Roblox.

cs.LG↗

Persistent Memory in Multi-Agent LLM Inference: What It Costs, What It Buys, and When You Can Tell

Decomposing long-context inference across cooperating agents bounds the active KV cache per call rather than total evidence, which matters when KV-cache memory binds. Many such systems add a persistent tier storing and recalling reasoning traces, usually validated by an ablation reporting an accuracy gain. We measure both on one three-tier agent architecture. Decomposition delivers: peak KV working set of 14.3 MiB per query against 35.5 and 35.3 MiB for single-pass and retrieval-augmented baselines. The persistent tier does not: across eight controlled dataset pairs at n=100 per arm it costs +0.368 MiB [+0.167, +0.590] of peak cache and produces no detectable accuracy change (+0.015, 95% CI [-0.011, +0.046]). We argue the null is structural: single-question benchmarks supply each item with its own evidence and score it independently, and correctness requires resetting stored traces between conditions, so recall has nothing informative to retrieve. Reaching it took four measurement corrections -- three inflating the apparent benefit, the fourth making an effect that size look resolvable -- none visible in the results table. We give the conditions an agent-memory ablation must satisfy and detection procedures that need no knowledge of the specific defect.

cs.AI↗

Prediction-Powered Conditional Inference

We study prediction-powered conditional inference in the setting where labeled data are scarce, unlabeled covariates are abundant, and a black-box machine-learning predictor is available. The goal is to perform statistical inference on conditional functionals evaluated at a fixed target point, such as conditional means, without imposing a parametric model for the conditional relationship. Our approach combines localization with prediction-based variance reduction. First, we introduce an RKHS localization method that learns a data-adaptive weight from covariates and reformulates the target conditional moment at the target point as a weighted unconditional moment. Second, we incorporate machine-learning predictions through a correction-based decomposition of this localized moment, yielding a prediction-powered estimator and confidence interval that reduce variance when the predictor is informative while preserving validity regardless of predictor accuracy. We establish nonasymptotic error bounds and, in the abundant-unlabeled regime, minimax-optimal convergence rates for the resulting estimator, prove pointwise asymptotic normality with consistent variance estimation, and provide an explicit variance decomposition that characterizes how machine-learning predictions and unlabeled covariates improve statistical efficiency. Numerical experiments on simulated and real datasets demonstrate valid conditional coverage and substantially sharper confidence intervals than alternative methods.

stat.ML↗

Incentivized Exploration with Stochastic Covariates: A Two-Stage Mechanism Design for Recommender System

Recommender systems play a crucial role in internet economies by connecting users with relevant products. However, designing effective recommender systems faces the key challenges: the exploration-exploitation tradeoff in securing incentive to explore new products against user's self-interested preferences. While prior work addresses Bayesian Incentive Compatibility (BIC) in fixed-design linear bandits (Sellke & Slivkins, 2023), we tackle the challenge of stochastic user covariates sampled online. Unlike standard black-box reductions (Mansour et al., 2020), our two-stage framework exploits the linear reward structure to achieve sublinear regret while satisfying incentive constraints. To address it, we propose a two-stage algorithm that integrates incentivized exploration with any efficient plug-in offline learning algorithms. In the first stage, it explores products while maintaining incentive compatibility to gather optimal samples. The second stage employs inverse proportional gap sampling strategy (IPGS) integrated with any efficient learning methods to secure sublinear regret. Theoretically, we prove that algorithm RCB achieves $O(\sqrt{KdT})$ regret and simultaneously satisfies incentive constraints, and discovers the tradeoff between incentive budget and regret, validating in experiments. We demonstrate RCB's strong incentive gain, sublinear regret, and robustness through a real application on personalized warfarin dosing and simulations.

cs.IR↗

Common-agency Games for Multi-Objective Test-Time Alignment

Aligning large language models (LLMs) with human preferences is inherently multi-objective: different users and evaluation criteria impose heterogeneous and often conflicting requirements on model outputs. We propose CAGE (Common-Agency Games for Alignment), a training-free, game-theoretic framework for multi-objective test-time alignment. CAGE models alignment objectives as strategic principals that allocate token-level incentives to a shared LLM, inducing an equilibrium policy that captures the joint effect of competing objectives. We develop an efficient algorithm based on equilibrium problems with equilibrium constraints (EPEC) to compute this equilibrium, and establish theoretical guarantees including existence and uniqueness of the equilibrium policy, convergence and stability of the algorithm, and no-regret learning dynamics. Empirically, CAGE enables flexible and fine-grained trade-offs across objectives at inference time, consistently outperforming existing test-time alignment methods while requiring no retraining. It further supports weak-to-strong generalization, making multi-objective alignment practical in resource-constrained settings.

cs.GT↗

Mechanism Design for Quality-Preserving LLM Advertising

Embedding advertisements into large language model (LLM) outputs introduces a fundamental tension: revenue optimization can distort content and degrade user experience. Existing approaches largely ignore this trade-off, often forcing irrelevant ads into responses. We propose a quality-preserving auction framework that explicitly integrates content fidelity into the mechanism design. Built on retrieval-augmented generation (RAG), our approach treats organic content as a reference and derives an endogenous reserve price that screens out ads with non-positive marginal social welfare contributions. We develop a KL-regularized single-allocation mechanism with Myerson payments and a screened VCG multi-allocation mechanism, both satisfying dominant-strategy incentive compatibility and individual rationality. Experiments across diverse scenarios demonstrate that our mechanisms outperform existing baselines in metrics such as revenue per ad and semantic similarity to no-ad responses. Our results establish a new paradigm for LLM advertising that enables monetization without compromising output quality.

cs.GT↗

A Statistical Market-Design Framework for Academic Job Markets

The academic job market for new statisticians is highly congested at the interview stage, where departments must rank and select candidates from large applicant pools without credible signals of candidate interest. As a result, interviews and offers are often misallocated, leading to unfilled positions and poor mutual fit. We frame interview allocation as a statistical ranking problem under uncertainty and propose a market-design framework that incorporates structured preference signaling into interview selection. Candidates submit a single standardized questionnaire describing preferences over interpretable job characteristics, which departments combine with traditional application materials and historical hiring data to estimate candidate-specific acceptance probabilities and expected utilities. To account for estimation uncertainty, we employ a confidence-calibrated ranking procedure based on pairwise utility comparisons that provides statistical guarantees for candidate ranking. We establish that truthful participation is optimal for candidates and that preference information improves departmental outcomes and matching stability. We use a dataset of U.S. statistics departments to show that the proposed framework substantially increases matching rates, improves match quality, and reduces hiring failures relative to the current practice.

stat.AP↗

Learn then Decide: A Learning Approach for Designing Data Marketplaces

As data marketplaces become increasingly central to the digital economy, it is crucial to design efficient pricing mechanisms that optimize revenue while ensuring fair and adaptive pricing. We introduce the Maximum Auction-to-Posted Price (MAPP) mechanism, a novel two-stage approach that first estimates the bidders' value distribution through auctions and then determines the optimal posted price based on the learned distribution. We establish that MAPP is individually rational and incentive-compatible, ensuring truthful bidding while balancing revenue maximization with minimal price discrimination. On the theoretical side, we establish a statistical viewpoint that recasts revenue optimization as a valuation density estimation problem: we show that revenue regret can be controlled by uniform error in estimating the valuation density. MAPP achieves a regret of $O_p(n^{-1}(\log n)^2)$ when incorporating historical bid data, where $n$ is the number of bids in the current round. For sequential dataset sales over $T$ rounds, we propose an online MAPP mechanism that dynamically adjusts pricing across datasets with varying value distributions. Our approach achieves no-regret learning, with the average cumulative regret converging at a rate of $O_p(T^{-1/2}(\log T)^2)$. We validate the effectiveness of MAPP through simulations and real-world data from the FCC AWS-3 spectrum auction.

stat.ML↗

A Robust Multi-Item Auction Design with Statistical Learning

We propose a novel statistical learning method for multi-item auctions that incorporates credible intervals. Our approach employs nonparametric density estimation to estimate credible intervals for bidder types based on historical data. We introduce two new strategies that leverage these credible intervals to reduce the time cost of implementing auctions. The first strategy screens potential winners' value regions within the credible intervals, while the second strategy simplifies the type distribution when the length of the interval is below a threshold value. These strategies are easy to implement and ensure fairness, dominant-strategy incentive compatibility, and dominant-strategy individual rationality with a high probability, while simultaneously reducing implementation costs. We demonstrate the effectiveness of our strategies using the Vickrey-Clarke-Groves mechanism and evaluate their performance through simulation experiments. Our results show that the proposed strategies consistently outperform alternative methods, achieving both revenue maximization and cost reduction objectives.

cs.GT↗

Prompt-Dependent Ranking of Large Language Models with Uncertainty Quantification

Rankings derived from pairwise comparisons are central to many economic and computational systems. In the context of large language models (LLMs), rankings are typically constructed from human preference data and presented as leaderboards that guide deployment decisions. However, existing approaches rely on point estimates, implicitly treating rankings as fixed objects despite substantial estimation noise and context-dependent performance variation. Acting on such rankings can lead to misallocation and welfare loss when apparent differences are not statistically meaningful. We study prompt-dependent ranking inference under pairwise human preferences and develop a framework for decision-safe rankings with statistically valid uncertainty guarantees. We model preferences using a contextual Bradley-Terry-Luce model in which the latent utility of each model depends on the input prompt. Rather than targeting point estimates of utilities, we directly conduct inference on induced rankings, constructing confidence sets based on simultaneous confidence intervals for pairwise utility differences. This approach yields statistically valid marginal and simultaneous confidence sets for prompt-specific ranks. Our framework connects recent advances in rank inference to contextual preference learning and provides tools for robust ranking-based decision-making. Empirically, using large-scale human preference data from LLM evaluations, we show that rankings vary substantially across prompt characteristics and that many apparent rank differences are not statistically distinguishable. We further demonstrate how uncertainty-aware rankings identify dominance only when supported by the data and otherwise return partial orders.

cs.CL↗

Auto-bidding under Return-on-Spend Constraints with Uncertainty Quantification

Auto-bidding systems are widely used in advertising to automatically determine bid values under constraints such as total budget and Return-on-Spend (RoS) targets. Existing works often assume that the value of an ad impression, such as the conversion rate, is known. This paper considers the more realistic scenario where the true value is unknown. We propose a novel method that uses conformal prediction to quantify the uncertainty of these values based on machine learning methods trained on historical bidding data with contextual features, without assuming the data are i.i.d. This approach is compatible with current industry systems that use machine learning to predict values. Building on prediction intervals, we introduce an adjusted value estimator derived from machine learning predictions, and show that it provides performance guarantees without requiring knowledge of the true value. We apply this method to enhance existing auto-bidding algorithms with budget and RoS constraints, and establish theoretical guarantees for achieving high reward while keeping RoS violations low. Empirical results on both simulated and real-world industrial datasets demonstrate that our approach improves performance while maintaining computational efficiency.

cs.LG↗

Uncertainty-Aware Multimodal Learning via Conformal Shapley Intervals

Multimodal learning combines information from multiple data modalities to improve predictive performance. However, modalities often contribute unequally and in a data dependent way, making it unclear which data modalities are genuinely informative and to what extent their contributions can be trusted. Quantifying modality level importance together with uncertainty is therefore central to interpretable and reliable multimodal learning. We introduce conformal Shapley intervals, a framework that combines Shapley values with conformal inference to construct uncertainty-aware importance intervals for each modality. Building on these intervals, we propose a modality selection procedure with a provable optimality guarantee: conditional on the observed features, the selected subset of modalities achieves performance close to that of the optimal subset. We demonstrate the effectiveness of our approach on multiple datasets, showing that it provides meaningful uncertainty quantification and strong predictive performance while relying on only a small number of informative modalities.

stat.ML↗

ALIGN: Aligned Delegation with Performance Guarantees for Multi-Agent LLM Reasoning

LLMs often underperform on complex reasoning tasks when relying on a single generation-and-selection pipeline. Inference-time ensemble methods can improve performance by sampling diverse reasoning paths or aggregating multiple candidate answers, but they typically treat candidates independently and provide no formal guarantees that ensembling improves reasoning quality. We propose a novel method, Aligned Delegation for Multi-Agent LLM Reasoning (ALIGN), which formulates LLM reasoning as an aligned delegation game. In ALIGN, a principal delegates a task to multiple agents that generate candidate solutions under designed incentives, and then selects among their outputs to produce a final answer. This formulation induces structured interaction among agents while preserving alignment between agent and principal objectives. We establish theoretical guarantees showing that, under a fair comparison with equal access to candidate solutions, ALIGN provably improves expected performance over single-agent generation. Our analysis accommodates correlated candidate answers and relaxes independence assumptions that are commonly used in prior work. Empirical results across a broad range of LLM reasoning benchmarks consistently demonstrate that ALIGN outperforms strong single-agent and ensemble baselines.

cs.LG↗

Fairness-aware kidney exchange and kidney paired donation

The kidney paired donation (KPD) program provides an innovative solution to overcome incompatibility challenges in kidney transplants by matching incompatible donor-patient pairs and facilitating kidney exchanges. To address unequal access to transplant opportunities, there are two widely used fairness criteria: group fairness and individual fairness. However, these criteria do not consider protected patient features, which refer to characteristics legally or ethically recognized as needing protection from discrimination, such as race and gender. Motivated by the calibration principle in machine learning, we introduce a new fairness criterion: the matching outcome should be conditionally independent of the protected feature, given the sensitization level. We integrate this fairness criterion as a constraint within the KPD optimization framework and propose a computationally efficient solution using linearization strategies and column-generation methods. Theoretically, we analyze the associated price of fairness using random graph models. Empirically, we compare our fairness criterion with group fairness and individual fairness through both simulations and a real-data example.

stat.ME↗

Surrogate-Powered Inference: Regularization and Adaptivity

High-quality labeled data are essential for reliable statistical inference, but are often limited by validation costs. While surrogate labels provide cost-effective alternatives, their noise can introduce non-negligible bias. To address this challenge, we propose the surrogate-powered inference (SPI) toolbox, a unified framework that leverages both the validity of high-quality labels and the abundance of surrogates to enable reliable statistical inference. SPI comprises three progressively enhanced versions. Base-SPI integrates validated labels and surrogates through augmentation to improve estimation efficiency. SPI+ incorporates regularized regression to safely handle multiple surrogates, preventing performance degradation due to error accumulation. SPI++ further optimizes efficiency under limited validation budgets through an adaptive, multiwave labeling procedure that prioritizes informative subjects for labeling. Compared to traditional methods, SPI substantially reduces the estimation error and increases the power in risk factor identification. These results demonstrate the value of SPI in improving the reproducibility. Theoretical guarantees and extensive simulation studies further illustrate the properties of our approach.

stat.ME↗

A Data Envelopment Analysis Approach for Assessing Fairness in Resource Allocation: Application to Kidney Exchange Programs

Kidney exchange programs have substantially increased transplantation rates but also raise critical concerns about fairness in organ allocation. We propose a novel framework leveraging Data Envelopment Analysis (DEA) to evaluate multiple dimensions of fairness-Priority, Access, and Outcome-within a unified model. This approach captures complexities often missed in single-metric analyses. Using data from the United Network for Organ Sharing, we separately quantify fairness across these dimensions: Priority fairness through waitlist durations, Access fairness via the Living Kidney Donor Profile Index (LKDPI) scores, and Outcome fairness based on graft lifespan. We then apply our conditional DEA model with covariate adjustment to demonstrate significant disparities in kidney allocation efficiency across ethnic groups. To quantify uncertainty, we employ conformal prediction within a novel Reference Frontier Mapping (RFM) framework, yielding group-conditional prediction intervals with finite-sample coverage guarantees. Our findings show notable differences in efficiency distributions between ethnic groups. Our study provides a rigorous framework for evaluating fairness in complex resource allocation systems with resource scarcity and mutual compatibility constraints.

cs.CY↗

Variance Reduction via Resampling and Experience Replay

Experience replay is a foundational technique in reinforcement learning that enhances learning stability by storing past experiences in a replay buffer and reusing them during training. Despite its practical success, its theoretical properties remain underexplored. In this paper, we present a theoretical framework that models experience replay using resampled $U$- and $V$-statistics, providing rigorous variance reduction guarantees. We apply this framework to policy evaluation tasks using the Least-Squares Temporal Difference (LSTD) algorithm and a Partial Differential Equation (PDE)-based model-free algorithm, demonstrating significant improvements in stability and efficiency, particularly in data-scarce scenarios. Beyond policy evaluation, we extend the framework to kernel ridge regression, showing that the experience replay-based method reduces the computational cost from the traditional $O(n^3)$ in time to as low as $O(n^2)$ in time while simultaneously reducing variance. Extensive numerical experiments validate our theoretical findings, demonstrating the broad applicability and effectiveness of experience replay in diverse machine learning tasks.

stat.ML↗

Incentivizing Truthful Language Models via Peer Elicitation Games

Large Language Models (LLMs) have demonstrated strong generative capabilities but remain prone to inconsistencies and hallucinations. We introduce Peer Elicitation Games (PEG), a training-free, game-theoretic framework for aligning LLMs through a peer elicitation mechanism involving a generator and multiple discriminators instantiated from distinct base models. Discriminators interact in a peer evaluation setting, where utilities are computed using a determinant-based mutual information score that provably incentivizes truthful reporting without requiring ground-truth labels. We establish theoretical guarantees showing that each agent, via online learning, achieves sublinear regret in the sense their cumulative performance approaches that of the best fixed truthful strategy in hindsight. Moreover, we prove last-iterate convergence to a truthful Nash equilibrium, ensuring that the actual policies used by agents converge to stable and truthful behavior over time. Empirical evaluations across multiple benchmarks demonstrate significant improvements in factual accuracy. These results position PEG as a practical approach for eliciting truthful behavior from LLMs without supervision or fine-tuning.

cs.LG↗