arXiv subjects

Argyris Oikonomou

Publications and source records attributed to Argyris Oikonomou.

11 recordsLinked to original sources

arXiv2025

Last-Iterate Convergence of Adaptive Riemannian Gradient Descent for Equilibrium Computation

Equilibrium computation on Riemannian manifolds provides a unifying framework for numerous problems in machine learning and data analytics. One of the simplest yet most fundamental methods is Riemannian gradient descent (RGD). While its Euclidean counterpart has been extensively studied, it remains unclear how the manifold curvature affects RGD in game-theoretic settings. This paper addresses this gap by establishing new convergence results for \textit{geodesic strongly monotone} games. Our key result shows that RGD attains last-iterate linear convergence in a \textit{geometry-agnostic} fashion, a key property for applications in machine learning. We extend this guarantee to stochastic and adaptive variants -- SRGD and FARGD -- and establish that: (i) the sample complexity of SRGD is geometry-agnostic and optimal with respect to noise; (ii) FARGD matches the convergence rate of its non-adaptive counterpart up to constant factors, while avoiding reliance on the condition number. Overall, this paper presents the first geometry-agnostic last-iterate convergence analysis for games beyond the Euclidean settings, underscoring the surprising power of RGD -- despite its simplicity -- in solving a wide spectrum of machine learning problems.

math.OC↗

arXiv2025

Prediction-Augmented Trees for Reliable Statistical Inference

The remarkable success of machine learning (ML) in predictive tasks has led scientists to incorporate ML predictions as a core component of the scientific discovery pipeline. This was exemplified by the landmark achievement of AlphaFold (Jumper et al. (2021)). In this paper, we study how ML predictions can be safely used in statistical analysis of data towards scientific discovery. In particular, we follow the framework introduced by Angelopoulos et al. (2023). In this framework, we assume access to a small set of $n$ gold-standard labeled samples, a much larger set of $N$ unlabeled samples, and a ML model that can be used to impute the labels of the unlabeled data points. We introduce two new learning-augmented estimators: (1) Prediction-Augmented Residual Tree (PART), and (2) Prediction-Augmented Quadrature (PAQ). Both estimators have significant advantages over existing estimators like PPI and PPI++ introduced by Angelopoulos et al. (2023) and Angelopoulos et al. (2024), respectively. PART is a decision-tree based estimator built using a greedy criterion. We first characterize PART's asymptotic distribution and demonstrate how to construct valid confidence intervals. Then we show that PART outperforms existing methods in real-world datasets from ecology, astronomy, and census reports, among other domains. This leads to estimators with higher confidence, which is the result of using both the gold-standard samples and the machine learning predictions. Finally, we provide a formal proof of the advantage of PART by exploring PAQ, an estimation that arises when considering the limit of PART when the depth its tree grows to infinity. Under appropriate assumptions in the input data we show that the variance of PAQ shrinks at rate of $O(N^{-1} + n^{-4})$, improving significantly on the $O(N^{-1}+n^{-1})$ rate of existing methods.

stat.ML↗

arXiv2025

COMAL: A Convergent Meta-Algorithm for Aligning LLMs with General Preferences

Many alignment methods, including reinforcement learning from human feedback (RLHF), rely on the Bradley-Terry reward assumption, which is not always sufficient to capture the full range and complexity of general human preferences. We explore RLHF under a general preference framework by modeling the alignment problem as a two-player zero-sum game in a game-theoretic framework, where the Nash equilibrium policy guarantees a 50% win rate against any competing policy. However, previous self-play algorithms for finding the Nash policy either diverge or only converge to a Nash policy in a modified game, even in a simple synthetic setting, thereby failing to maintain the 50% win rate guarantee against all other policies. We propose a meta-algorithm, Convergent Meta Alignment Algorithm (COMAL), for language model alignment with general preferences, inspired by convergent algorithms in game theory. We provide theoretical analysis that our meta-algorithm converges to an exact Nash policy in the last iterate and demonstrate its effectiveness on a range of synthetic and preference optimization datasets. COMAL is simple and can be integrated with many existing methods designed for preference optimization with minimal changes, and empirically it consistently maintains above 60.2% and 56.8% win rates, when applied to Llama-3-8B-Instruct and Qwen2.5-7B, against all compared algorithms under controlled evaluations.

cs.LG↗

arXiv2025

Provable Partially Observable Reinforcement Learning with Privileged Information

Partial observability of the underlying states generally presents significant challenges for reinforcement learning (RL). In practice, certain \emph{privileged information}, e.g., the access to states from simulators, has been exploited in training and has achieved prominent empirical successes. To better understand the benefits of privileged information, we revisit and examine several simple and practically used paradigms in this setting. Specifically, we first formalize the empirical paradigm of \emph{expert distillation} (also known as \emph{teacher-student} learning), demonstrating its pitfall in finding near-optimal policies. We then identify a condition of the partially observable environment, the \emph{deterministic filter condition}, under which expert distillation achieves sample and computational complexities that are \emph{both} polynomial. Furthermore, we investigate another useful empirical paradigm of \emph{asymmetric actor-critic}, and focus on the more challenging setting of observable partially observable Markov decision processes. We develop a belief-weighted asymmetric actor-critic algorithm with polynomial sample and quasi-polynomial computational complexities, in which one key component is a new provable oracle for learning belief states that preserve \emph{filter stability} under a misspecified model, which may be of independent interest. Finally, we also investigate the provable efficiency of partially observable multi-agent RL (MARL) with privileged information. We develop algorithms featuring \emph{centralized-training-with-decentralized-execution}, a popular framework in empirical MARL, with polynomial sample and (quasi-)polynomial computational complexities in both paradigms above. Compared with a few recent related theoretical studies, our focus is on understanding practically inspired algorithmic paradigms, without computationally intractable oracles.

cs.LG↗

arXiv2024

Accelerated Algorithms for Constrained Nonconvex-Nonconcave Min-Max Optimization and Comonotone Inclusion

We study constrained comonotone min-max optimization, a structured class of nonconvex-nonconcave min-max optimization problems, and their generalization to comonotone inclusion. In our first contribution, we extend the Extra Anchored Gradient (EAG) algorithm, originally proposed by Yoon and Ryu (2021) for unconstrained min-max optimization, to constrained comonotone min-max optimization and comonotone inclusion, achieving an optimal convergence rate of $O\left(\frac{1}{T}\right)$ among all first-order methods. Additionally, we prove that the algorithm's iterations converge to a point in the solution set. In our second contribution, we extend the Fast Extra Gradient (FEG) algorithm, as developed by Lee and Kim (2021), to constrained comonotone min-max optimization and comonotone inclusion, achieving the same $O\left(\frac{1}{T}\right)$ convergence rate. This rate is applicable to the broadest set of comonotone inclusion problems yet studied in the literature. Our analyses are based on simple potential function arguments, which might be useful for analyzing other accelerated algorithms.

math.OC↗

arXiv2024

Injecting Undetectable Backdoors in Obfuscated Neural Networks and Language Models

As ML models become increasingly complex and integral to high-stakes domains such as finance and healthcare, they also become more susceptible to sophisticated adversarial attacks. We investigate the threat posed by undetectable backdoors, as defined in Goldwasser et al. (FOCS '22), in models developed by insidious external expert firms. When such backdoors exist, they allow the designer of the model to sell information on how to slightly perturb their input to change the outcome of the model. We develop a general strategy to plant backdoors to obfuscated neural networks, that satisfy the security properties of the celebrated notion of indistinguishability obfuscation. Applying obfuscation before releasing neural networks is a strategy that is well motivated to protect sensitive information of the external expert firm. Our method to plant backdoors ensures that even if the weights and architecture of the obfuscated model are accessible, the existence of the backdoor is still undetectable. Finally, we introduce the notion of undetectable backdoors to language models and extend our neural network backdoor attacks to such models based on the existence of steganographic functions.

cs.LG↗

arXiv2024

Dynamics and Contracts for an Agent with Misspecified Beliefs

We study a single-agent contracting environment where the agent has misspecified beliefs about the outcome distributions for each chosen action. First, we show that for a myopic Bayesian learning agent with only two possible actions, the empirical frequency of the chosen actions converges to a Berk-Nash equilibrium. However, through a constructed example, we illustrate that this convergence in action frequencies fails when the agent has three or more actions. Furthermore, with multiple actions, even computing an $\varepsilon$-Berk-Nash equilibrium requires at least quasi-polynomial time under the Exponential Time Hypothesis (ETH) for the PPAD-class. This finding poses a significant challenge to the existence of simple learning dynamics that converge in action frequencies. Motivated by this challenge, we focus on the contract design problems for an agent with misspecified beliefs and two possible actions. We show that the revenue-optimal contract, under a Berk-Nash equilibrium, can be computed in polynomial time. Perhaps surprisingly, we show that even a minor degree of misspecification can result in a significant reduction in optimal revenue.

cs.GT↗

arXiv2022

Tight Last-Iterate Convergence of the Extragradient and the Optimistic Gradient Descent-Ascent Algorithm for Constrained Monotone Variational Inequalities

The monotone variational inequality is a central problem in mathematical programming that unifies and generalizes many important settings such as smooth convex optimization, two-player zero-sum games, convex-concave saddle point problems, etc. The extragradient algorithm by Korpelevich [1976] and the optimistic gradient descent-ascent algorithm by Popov [1980] are arguably the two most classical and popular methods for solving monotone variational inequalities. Despite their long histories, the following major problem remains open. What is the last-iterate convergence rate of the extragradient algorithm or the optimistic gradient descent-ascent algorithm for monotone and Lipschitz variational inequalities with constraints? We resolve this open problem by showing that both the extragradient algorithm and the optimistic gradient descent-ascent algorithm have a tight $O\left(\frac{1}{\sqrt{T}}\right)$ last-iterate convergence rate for arbitrary convex feasible sets, which matches the lower bound by Golowich et al. [2020a,b]. Our rate is measured in terms of the standard gap function. At the core of our results lies a non-standard performance measure -- the tangent residual, which can be viewed as an adaptation of the norm of the operator that takes the local constraints into account. We use the tangent residual (or a slight variation of the tangent residual) as the the potential function in our analysis of the extragradient algorithm (or the optimistic gradient descent-ascent algorithm) and prove that it is non-increasing between two consecutive iterates.

math.OC↗

arXiv2022

Computing Simple Mechanisms: Lift-and-Round over Marginal Reduced Forms

We study revenue maximization in multi-item multi-bidder auctions under the natural item-independence assumption - a classical problem in Multi-Dimensional Bayesian Mechanism Design. One of the biggest challenges in this area is developing algorithms to compute (approximately) optimal mechanisms that are not brute-force in the size of the bidder type space, which is usually exponential in the number of items in multi-item auctions. Unfortunately, such algorithms were only known for basic settings of our problem when bidders have unit-demand [CHMS10,CMS15] or additive valuations [Yao15]. In this paper, we significantly improve the previous results and design the first algorithm that runs in time polynomial in the number of items and the number of bidders to compute mechanisms that are $O(1)$-approximations to the optimal revenue when bidders have XOS valuations, resolving an open problem raised in [CM16,CZ17]. Moreover, the computed mechanism has a simple structure: It is either a posted price mechanism or a two-part tariff mechanism. As a corollary of our result, we show how to compute an approximately optimal and simple mechanism efficiently using only sample access to the bidders' value distributions. Our algorithm builds on two innovations that allow us to search over the space of mechanisms efficiently: (i) a new type of succinct representation of mechanisms - the marginal reduced forms, and (ii) a novel Lift-and-Round procedure that concavifies the problem.

cs.GT↗

arXiv2021

On Simple Mechanisms for Dependent Items

We study the problem of selling $n$ heterogeneous items to a single buyer, whose values for different items are dependent. Under arbitrary dependence, Hart and Nisan show that no simple mechanism can achieve a non-negligible fraction of the optimal revenue even with only two items. We consider the setting where the buyer's type is drawn from a correlated distribution that can be captured by a Markov Random Field, one of the most prominent frameworks for modeling high-dimensional distributions with structure. If the buyer's valuation is additive or unit-demand, we extend the result to all MRFs and show that max(SRev,BRev) can achieve an $Ω\left(\frac{1}{e^{O(Δ)}}\right)$-fraction of the optimal revenue, where $Δ$ is a parameter of the MRF that is determined by how much the value of an item can be influenced by the values of the other items. We further show that the exponential dependence on $Δ$ is unavoidable for our approach and a polynomial dependence on $Δ$ is unavoidable for any approach. When the buyer has a XOS valuation, we show that max(Srev,Brev) achieves at least an $Ω\left(\frac{1}{e^{O(Δ)}+\frac{1}{\sqrt{nγ}}}\right)$-fraction of the optimal revenue, where $γ$ is the spectral gap of the Glauber dynamics of the MRF. Note that in the special case of independently distributed items, $Δ=0$ and $\frac{1}{nγ}\leq 1$, and our results recover the known constant factor approximations for a XOS buyer. We further extend our parametric approximation to several other well-studied dependency measures such as the Dobrushin coefficient and the inverse temperature. Our results are based on the Duality-Framework by Cai et al. and a new concentration inequality for XOS functions over dependent random variables.

cs.GT↗

arXiv2020

An Efficient $\varepsilon$-BIC to BIC Transformation and Its Application to Black-Box Reduction in Revenue Maximization

We consider the black-box reduction from multi-dimensional revenue maximization to virtual welfare maximization. Cai et al. show a polynomial-time approximation-preserving reduction, however, the mechanism produced by their reduction is only approximately Bayesian incentive compatible ($\varepsilon$-BIC). We provide two new polynomial time transformations that convert any $\varepsilon$-BIC mechanism to an exactly BIC mechanism with only a negligible revenue loss. Our first transformation applies to any mechanism design setting with downward-closed outcome space and only requires sample access to the agents' type distributions. Our second transformation applies to the fully general outcome space, removing the downward-closed assumption, but requires full access to the agents' type distributions. Both transformations only require query access to the original $\varepsilon$-BIC mechanism. Other $\varepsilon$-BIC to BIC transformations for revenue exist in the literature but all require exponential time to run in both of the settings we consider. As an application of our transformations, we improve the reduction by Cai et al. to generate an exactly BIC mechanism.

cs.GT↗