arXiv subjects

Cédric Langbort

Publications and source records attributed to Cédric Langbort.

12 recordsLinked to original sources

arXiv2026

Steady-state Based Approach to Online Non-stochastic Control

We study the problem of online non-stochastic control (ONC), which is the control of a linear system under adversarial disturbances and adversarial cost functions, with the aim of minimizing the total cost incurred. A recent line of literature in ONC develops algorithms that enjoy sublinear regret with respect to a benchmark based on the set of steady-states that are attainable by a constant input. In this work, we extend this research direction by giving an algorithm that enjoys $\mathcal{O}(\sqrt{T})$ regret with respect to a richer benchmark set, namely the set of steady-states attainable under an \emph{affine controller}. Since this benchmark substantially broadens the comparison class, it provides significantly stronger performance guarantees. Our proposed algorithm combines a Follow-The-Perturbed-Leader-style online non-convex optimization approach with a batching method that maintains stability despite changing policies. Although our proposed algorithm requires solving non-convex subproblems, we show that an approximate solution to this subproblem is sufficient to ensure $\mathcal{O}(\sqrt{T})$ regret. Furthermore, numerical experiments show that our algorithm enjoys lower total cost and similar computation to existing methods in certain settings.

math.OC↗

arXiv2025

Revisiting Regret Benchmarks in Online Non-Stochastic Control

In the online non-stochastic control problem, an agent sequentially selects control inputs for a linear dynamical system when facing unknown and adversarially selected convex costs and disturbances. A common metric for evaluating control policies in this setting is policy regret, defined relative to the best-in-hindsight linear feedback controller. However, for general convex costs, this benchmark may be less meaningful since linear controllers can be highly suboptimal. To address this, we introduce an alternative, more suitable benchmark--the performance of the best fixed input. We show that this benchmark can be viewed as a natural extension of the standard benchmark used in online convex optimization and propose a novel online control algorithm that achieves sublinear regret with respect to this new benchmark. We also discuss the connections between our method and the original one proposed by Agarwal et al. in their seminal work introducing the online non-stochastic control problem, and compare the performance of both approaches through numerical simulations.

math.OC↗

arXiv2024

Responding to Promises: No-regret learning against followers with memory

We consider a repeated Stackelberg game setup where the leader faces a sequence of followers of unknown types and must learn what commitments to make. While previous works have considered followers that best respond to the commitment announced by the leader in every round, we relax this setup in two ways. Motivated by natural scenarios where the leader's reputation factors into how the followers choose their response, we consider followers with memory. Specifically, we model followers that base their response on not just the leader's current commitment but on an aggregate of their past commitments. In developing learning strategies that the leader can employ against such followers, we make the second relaxation and assume boundedly rational followers. In particular, we focus on followers employing quantal responses. Interestingly, we observe that the smoothness property offered by the quantal response (QR) model helps in addressing the challenge posed by learning against followers with memory. Utilizing techniques from online learning, we develop algorithms that guarantee $O(\sqrt{T})$ regret for quantal responding memory-less followers and $O(\sqrt{BT})$ regret for followers with bounded memory of length $B$ with both scaling polynomially in game parameters.

cs.GT↗

arXiv2024

Almost-Bayesian Quadratic Persuasion (Extended Version)

In this article, we relax the Bayesianity assumption in the now-traditional model of Bayesian Persuasion introduced by Kamenica & Gentzkow. Unlike preexisting approaches -- which have tackled the possibility of the receiver (Bob) being non-Bayesian by considering that his thought process is not Bayesian yet known to the sender (Alice), possibly up to a parameter -- we let Alice merely assume that Bob behaves 'almost like' a Bayesian agent, in some sense, without resorting to any specific model. Under this assumption, we study Alice's strategy when both utilities are quadratic and the prior is isotropic. We show that, contrary to the Bayesian case, Alice's optimal response may not be linear anymore. This fact is unfortunate as linear policies remain the only ones for which the induced belief distribution is known. What is more, evaluating linear policies proves difficult except in particular cases, let alone finding an optimal one. Nonetheless, we derive bounds that prove linear policies are near-optimal and allow Alice to compute a near-optimal linear policy numerically. With this solution in hand, we show that Alice shares less information with Bob as he departs more from Bayesianity, much to his detriment.

cs.GT↗

arXiv2024

On Network Congestion Reduction Using Public Signals Under Boundedly Rational User Equilibria (Full Version)

Boundedly Rational User Equilibria (BRUE) capture situations where all agents on a transportation network are electing the fastest option up to some time indifference, and serve as a relaxation of User Equilibria (UE), where each agent exactly minimizes their travel time. We study how the social cost under BRUE departs from that of UE in the context of static demand and stochastic costs, along with the implications of BRUE on the optimal signaling scheme of a benevolent central planner. We show that the average excess time is sublinear in the maximum time indifference of the agents, though such aggregate may hide disparity between populations and the sublinearity constant depends on the topology of the network. Regarding the design of public signals, even though in the limit where agents are totally indifferent, it is optimal to not reveal any information, there is in general no trend in how much information is optimally disclosed to agents. What is more, an increase in information disclosed may either harm or benefit agents as a whole.

cs.GT↗

arXiv2024

Misinformation Regulation in the Presence of Competition between Social Media Platforms (Extended Version)

Social media platforms have diverse content moderation policies, with many prominent actors hesitant to impose strict regulations. A key reason for this reluctance could be the competitive advantage that comes with lax regulation. A popular platform that starts enforcing content moderation rules may fear that it could lose users to less-regulated alternative platforms. Moreover, if users continue harmful activities on other platforms, regulation ends up being futile. This article examines the competitive aspect of content moderation by considering the motivations of all involved players (platformer, news source, and social media users), identifying the regulation policies sustained in equilibrium, and evaluating the information quality available on each platform. Applied to simple yet relevant social networks such as stochastic block models, our model reveals the conditions for a popular platform to enforce strict regulation without losing users. Effectiveness of regulation depends on the diffusive property of news posts, friend interaction qualities in social media, the sizes and cohesiveness of communities, and how much sympathizers appreciate surprising news from influencers.

cs.GT↗

arXiv2023

Learn and Control while Switching: with Guaranteed Stability and Sublinear Regret

Over-actuated systems often make it possible to achieve specific performances by switching between different subsets of actuators. However, when the system parameters are unknown, transferring authority to different subsets of actuators is challenging due to stability and performance efficiency concerns. This paper presents an efficient algorithm to tackle the so-called "learn and control while switching between different actuating modes" problem in the Linear Quadratic (LQ) setting. Our proposed strategy is constructed upon Optimism in the Face of Uncertainty (OFU) based algorithm equipped with a projection toolbox to keep the algorithm efficient, regret-wise. Along the way, we derive an optimum duration for the warm-up phase, thanks to the existence of a stabilizing neighborhood. The stability of the switched system is also guaranteed by designing a minimum average dwell time. The proposed strategy is proved to have a regret bound of $\mathcal{\bar{O}}\big(\sqrt{T}\big)+\mathcal{O}\big(ns\sqrt{T}\big)$ in horizon $T$ with $(ns)$ number of switches, provably outperforming naively applying the basic OFU algorithm.

eess.SY↗

arXiv2019

Strategic Inference with a Single Private Sample

Motivated by applications in cyber security, we develop a simple game model for describing how a learning agent's private information influences an observing agent's inference process. The model describes a situation in which one of the agents (attacker) is deciding which of two targets to attack, one with a known reward and another with uncertain reward. The attacker receives a single private sample from the uncertain target's distribution and updates its belief of the target quality. The other agent (defender) knows the true rewards, but does not see the sample that the attacker has received. This leads to agents possessing asymmetric information: the attacker is uncertain over the parameter of the distribution, whereas the defender is uncertain about the observed sample. After the attacker updates its belief, both the attacker and the defender play a simultaneous move game based on their respective beliefs. We offer a characterization of the pure strategy equilibria of the game and explain how the players' decisions are influenced by their prior knowledge and the payoffs/costs.

cs.GT↗

arXiv2015

Faithful Implementations of Distributed Algorithms and Control Laws

When a distributed algorithm must be executed by strategic agents with misaligned interests, a social leader needs to introduce an appropriate tax/subsidy mechanism to incentivize agents to faithfully implement the intended algorithm so that a correct outcome is obtained. We discuss the incentive issues of implementing economically efficient distributed algorithms using the framework of indirect mechanism design theory. In particular, we show that indirect Groves mechanisms are not only sufficient but also necessary to achieve incentive compatibility. This result can be viewed as a generalization of the Green-Laffont theorem to indirect mechanisms. Then we introduce the notion of asymptotic incentive compatibility as an appropriate solution concept to faithfully implement distributed and iterative optimization algorithms. We consider two special types of optimization algorithms: dual decomposition algorithms for resource allocation and average consensus algorithms.

cs.GT↗

arXiv2013

A Faithful Distributed Implementation of Dual Decomposition and Average Consensus Algorithms

We consider large scale cost allocation problems and consensus seeking problems for multiple agents, in which agents are suggested to collaborate in a distributed algorithm to find a solution. If agents are strategic to minimize their own individual cost rather than the global social cost, they are endowed with an incentive not to follow the intended algorithm, unless the tax/subsidy mechanism is carefully designed. Inspired by the classical Vickrey-Clarke-Groves mechanism and more recent algorithmic mechanism design theory, we propose a tax mechanism that incentivises agents to faithfully implement the intended algorithm. In particular, a new notion of asymptotic incentive compatibility is introduced to characterize a desirable property of such class of mechanisms. The proposed class of tax mechanisms provides a sequence of mechanisms that gives agents a diminishing incentive to deviate from suggested algorithm.

math.OC↗

arXiv2012

Nash Equilibria for Stochastic Games with Asymmetric Information-Part 1: Finite Games

A model of stochastic games where multiple controllers jointly control the evolution of the state of a dynamic system but have access to different information about the state and action processes is considered. The asymmetry of information among the controllers makes it difficult to compute or characterize Nash equilibria. Using common information among the controllers, the game with asymmetric information is shown to be equivalent to another game with symmetric information. Further, under certain conditions, a Markov state is identified for the equivalent symmetric information game and its Markov perfect equilibria are characterized. This characterization provides a backward induction algorithm to find Nash equilibria of the original game with asymmetric information in pure or behavioral strategies. Each step of this algorithm involves finding Bayesian Nash equilibria of a one-stage Bayesian game. The class of Nash equilibria of the original game that can be characterized in this backward manner are named common information based Markov perfect equilibria.

cs.GT↗

arXiv2010

Task Release Control for Decision Making Queues

We consider the optimal duration allocation in a decision making queue. Decision making tasks arrive at a given rate to a human operator. The correctness of the decision made by human evolves as a sigmoidal function of the duration allocated to the task. Each task in the queue loses its value continuously. We elucidate on this trade-off and determine optimal policies for the human operator. We show the optimal policy requires the human to drop some tasks. We present a receding horizon optimization strategy, and compare it with the greedy policy.

math.OC↗