While a number of existing approaches for building foundation model agents rely on prompting or fine-tuning with human demonstrations, it is not sufficient in dynamic environments (e.g., mobile device control). On-policy reinforcement learning (RL) should address these limitations, but collecting actual rollouts in an environment is often undesirable in truly open-ended agentic problems such as mobile device control or interacting with humans, where each unit of interaction is associated with a cost. In such scenarios, a method for policy learning that can utilize off-policy experience by learning a trained action-value function is much more effective. In this paper, we develop an approach, called Digi-Q, to train VLM-based action-value Q-functions which are then used to extract the agent policy. We study our approach in the mobile device control setting. Digi-Q trains the Q-function using offline temporal-difference (TD) learning, on top of frozen, intermediate-layer features of a VLM. Compared to fine-tuning the whole VLM, this approach saves us compute and enhances scalability. To make the VLM features amenable for representing the Q-function, we need to employ an initial phase of fine-tuning to amplify coverage over actionable information needed for value function. Once trained, we use this Q-function via a Best-of-N policy extraction operator that imitates the best action out of multiple candidate actions from the current policy as ranked by the value function, enabling policy improvement without environment interaction. Digi-Q outperforms several prior methods on user-scale device control tasks in Android-in-the-Wild, attaining 21.2% improvement over prior best-performing method. In some cases, our Digi-Q approach already matches state-of-the-art RL methods that require interaction. The project is open-sourced at https://github.com/DigiRL-agent/digiq
The growing reliance on AI-driven pricing algorithms has raised antitrust concerns, particularly that third-party use of nonpublic competitor data may facilitate information sharing and undermine market competition. This paper examines a setting in which firms receive demand information from a common third party and incorporate it into their pricing algorithms. Specifically, I analyze the pricing behavior of Q-learning agents when demand shocks are observable. The simulation results show that at high discount factors, agents consistently charge constant supracompetitive prices across demand states, resulting in price rigidity. This pricing pattern is sustained through collusive strategies that effectively punish deviations and feature a discontinuous switch from punishment back to cooperation. At medium discount factors, however, agents lower prices during booms to reduce incentives to deviate and raise them during downturns, resulting in countercyclical pricing consistent with Rotemberg and Saloner (1986). The findings highlight both the pricing adaptability of Q-learning to fluctuating market demand and the robustness of algorithmic collusion. Compared to settings with unobserved demand shocks, demand observability reduces firm profits and enhances consumer surplus, suggesting that access to market information under algorithmic pricing does not necessarily harm consumers.
This paper investigates the instability of Q-learning in continuous environments, a challenge frequently encountered by practitioners. Traditionally, this instability is attributed to bootstrapping and regression model errors. Using a representative reinforcement learning benchmark, we systematically examine the effects of bootstrapping and model inaccuracies by incrementally eliminating these potential error sources. Our findings reveal that even in relatively simple benchmarks, the fundamental task of Q-learning - iteratively learning a Q-function from policy-specific target values - can be inherently ill-posed and prone to failure. These insights cast doubt on the reliability of Q-learning as a universal solution for reinforcement learning problems.
We study the problem of solving fixed-point equations for seminorm-contractive operators and establish foundational results on the non-asymptotic behavior of iterative algorithms in both deterministic and stochastic settings. Specifically, in the deterministic setting, we prove a fixed-point theorem for seminorm-contractive operators, showing that iterates converge geometrically to the kernel of the seminorm. In the stochastic setting, we analyze the corresponding stochastic approximation (SA) algorithm under seminorm-contractive operators and Markovian noise, providing a finite-sample analysis for various stepsize choices. A benchmark for equation solving is linear systems of equations, where the convergence behavior of fixed-point iteration is closely tied to the stability of linear dynamical systems. In this special case, our results provide a complete characterization of system stability with respect to a seminorm, linking it to the solution of a Lyapunov equation in terms of positive semi-definite matrices. In the stochastic setting, we establish a finite-sample analysis for linear Markovian SA without requiring the Hurwitzness assumption. Our theoretical results offer a unified framework for deriving finite-sample bounds for various reinforcement learning algorithms in the average reward setting, including TD($\lambda$) for policy evaluation (which is a special case of solving a Poisson equation) and Q-learning for control.
Scalability remains a challenge in multi-agent reinforcement learning and is currently under active research. A framework named mean-field reinforcement learning (MFRL) could alleviate the scalability problem by employing the Mean Field Theory to turn a many-agent problem into a two-agent problem. However, this framework lacks the ability to identify essential interactions under nonstationary environments. Causality contains relatively invariant mechanisms behind interactions, though environments are nonstationary. Therefore, we propose an algorithm called causal mean-field Q-learning (CMFQ) to address the scalability problem. CMFQ is ever more robust toward the change of the number of agents though inheriting the compressed representation of MFRL's action-state space. Firstly, we model the causality behind the decision-making process of MFRL into a structural causal model (SCM). Then the essential degree of each interaction is quantified via intervening on the SCM. Furthermore, we design the causality-aware compact representation for behavioral information of agents as the weighted sum of all behavioral information according to their causal effects. We test CMFQ in a mixed cooperative-competitive game and a cooperative game. The result shows that our method has excellent scalability performance in both training in environments containing a large number of agents and testing in environments containing much more agents.
Mean field equilibrium (MFE) has emerged as a computationally tractable solution concept for large dynamic games. However, computing MFE remains challenging due to nonlinearities and the absence of contraction properties, limiting its reliability for counterfactual analysis and comparative statics. This paper focuses on MFE in dynamic models where agents interact through a scalar function of the population distribution, referred to as the scalar interaction function. Such models naturally arise in a wide range of applications involving market dynamics and strategic competition. The main contribution of this paper is to introduce iterative algorithms that leverage the scalar interaction structure and are guaranteed to converge to the MFE under mild assumptions. Leveraging this structure, we also establish an MFE existence result for non-compact state spaces and analytical comparative statics. To the best of our knowledge, these are the first algorithms with global convergence guarantees in such settings. Unlike existing approaches, our algorithms do not rely on monotonicity or contraction properties, significantly broadening their applicability. Furthermore, we provide a model-free algorithm that learns the MFE via simulation and reinforcement learning techniques such as Q-learning and policy gradient methods without requiring prior knowledge of payoff or transition functions. We apply our algorithms to classic models of dynamic competition, such as capacity competition, and to competitive models motivated by online marketplaces, including ridesharing and inventory competition, as well as to social learning models. We show how key market parameters influence equilibrium outcomes through reliable comparative statics in these representative models, providing insights into the design of competitive systems.
As a data-driven approach, offline MARL learns superior policies solely from offline datasets, ideal for domains rich in historical data but with high interaction costs and risks. However, most existing methods are task-specific, requiring retraining for new tasks, leading to redundancy and inefficiency. To address this issue, in this paper, we propose a task-efficient multi-task offline MARL algorithm, Skill-Discovery Conservative Q-Learning (SD-CQL). Unlike existing offline skill-discovery methods, SD-CQL discovers skills by reconstructing the next observation. It then evaluates fixed and variable actions separately and employs behavior-regularized conservative Q-learning to execute the optimal action for each skill. This approach eliminates the need for local-global alignment and enables strong multi-task generalization from limited small-scale source tasks. Substantial experiments on StarCraftII demonstrates the superior generalization performance and task-efficiency of SD-CQL. It achieves the best performance on $\textbf{10}$ out of $14$ task sets, with up to $\textbf{65%}$ improvement on individual task sets, and is within $4\%$ of the best baseline on the remaining four.
Extensive behavioral experiments reveal that conditional cooperation is a prevalent phenomenon. Previous game-theoretical studies have predominantly relied on hard-manner models, where cooperation is triggered only upon reaching a specific threshold. However, this approach contrasts with the observed flexibility of human behaviors, where individuals adapt their strategies dynamically based on their surroundings. To capture this adaptability, we introduce a soft form of conditional cooperation by integrating the Q-learning algorithm from reinforcement learning. In this form, players not only reciprocate mutual cooperation but may also defect in highly cooperative environments or cooperate in less cooperative settings to maximize rewards. To explore the effects of hard and soft conditional cooperators, we examine their interactions in two scenarios: structural mixture (SM) and probabilistic mixture (PM), where the two behavioral modes are fixed and probabilistically adopted, respectively. In SM, hard conditional cooperators enhance cooperation when the threshold is low but hinder it otherwise. Surprisingly, in PM, the cooperation prevalence exhibits two first-order phase transitions as the probability is varied, leading to high, low, and vanishing levels of cooperation. Analysis of Q-tables offers insights into the "psychological shifts" of soft conditional cooperators and the overall evolutionary dynamics. Model extensions confirm the robustness of our findings. These results highlight the novel complexities arising from the diversity of conditional cooperators.
Vision-Language-Action (VLA) models have shown substantial potential in real-world robotic manipulation. However, fine-tuning these models through supervised learning struggles to achieve robust performance due to limited, inconsistent demonstrations, especially in contact-rich environments. In this paper, we propose a reinforced fine-tuning approach for VLA models, named ConRFT, which consists of offline and online fine-tuning with a unified consistency-based training objective, to address these challenges. In the offline stage, our method integrates behavior cloning and Q-learning to effectively extract policy from a small set of demonstrations and stabilize value estimating. In the online stage, the VLA model is further fine-tuned via consistency policy, with human interventions to ensure safe exploration and high sample efficiency. We evaluate our approach on eight diverse real-world manipulation tasks. It achieves an average success rate of 96.3% within 45-90 minutes of online fine-tuning, outperforming prior supervised methods with a 144% improvement in success rate and 1.9x shorter episode length. This work highlights the potential of integrating reinforcement learning to enhance the performance of VLA models for real-world robotic applications. Videos and code are available at our project website https://cccedric.github.io/conrft/.
We introduce VistaFlow, a scalable three-dimensional imaging technique capable of reconstructing fully interactive 3D volumetric images from a set of 2D photographs. Our model synthesizes novel viewpoints through a differentiable rendering system capable of dynamic resolution management on photorealistic 3D scenes. We achieve this through the introduction of QuiQ, a novel intermediate video controller trained through Q-learning to maintain a consistently high framerate by adjusting render resolution with millisecond precision. Notably, VistaFlow runs natively on integrated CPU graphics, making it viable for mobile and entry-level devices while still delivering high-performance rendering. VistaFlow bypasses Neural Radiance Fields (NeRFs), using the PlenOctree data structure to render complex light interactions such as reflection and subsurface scattering with minimal hardware requirements. Our model is capable of outperforming state-of-the-art methods with novel view synthesis at a resolution of 1080p at over 100 frames per second on consumer hardware. By tailoring render quality to the capabilities of each device, VistaFlow has the potential to improve the efficiency and accessibility of photorealistic 3D scene rendering across a wide spectrum of hardware, from high-end workstations to inexpensive microcontrollers.
Data preprocessing is a critical yet frequently neglected aspect of machine learning, often paid little attention despite its potentially significant impact on model performance. While automated machine learning pipelines are starting to recognize and integrate data preprocessing into their solutions for classification and regression tasks, this integration is lacking for more specialized tasks like survival or time-to-event models. As a result, survival analysis not only faces the general challenges of data preprocessing but also suffers from the lack of tailored, automated solutions in this area. To address this gap, this paper presents 'CleanSurvival', a reinforcement-learning-based solution for optimizing preprocessing pipelines, extended specifically for survival analysis. The framework can handle continuous and categorical variables, using Q-learning to select which combination of data imputation, outlier detection and feature extraction techniques achieves optimal performance for a Cox, random forest, neural network or user-supplied time-to-event model. The package is available on GitHub: https://github.com/datasciapps/CleanSurvival Experimental benchmarks on real-world datasets show that the Q-learning-based data preprocessing results in superior predictive performance to standard approaches, finding such a model up to 10 times faster than undirected random grid search. Furthermore, a simulation study demonstrates the effectiveness in different types and levels of missingness and noise in the data.
We present the first gap-dependent analysis of regret and communication cost for on-policy federated $Q$-Learning in tabular episodic finite-horizon Markov decision processes (MDPs). Existing FRL methods focus on worst-case scenarios, leading to $\sqrt{T}$-type regret bounds and communication cost bounds with a $\log T$ term scaling with the number of agents $M$, states $S$, and actions $A$, where $T$ is the average total number of steps per agent. In contrast, our novel framework leverages the benign structures of MDPs, such as a strictly positive suboptimality gap, to achieve a $\log T$-type regret bound and a refined communication cost bound that disentangles exploration and exploitation. Our gap-dependent regret bound reveals a distinct multi-agent speedup pattern, and our gap-dependent communication cost bound removes the dependence on $MSA$ from the $\log T$ term. Notably, our gap-dependent communication cost bound also yields a better global switching cost when $M=1$, removing $SA$ from the $\log T$ term.
This paper investigates the anti-jamming channel access problem in complex and unknown jamming environments, where the jammer could dynamically adjust its strategies to target different channels. Traditional channel hopping anti-jamming approaches using fixed patterns are ineffective against such dynamic jamming attacks. Although the emerging deep reinforcement learning (DRL) based dynamic channel access approach could achieve the Nash equilibrium under fast-changing jamming attacks, it requires extensive training episodes. To address this issue, we propose a fast adaptive anti-jamming channel access approach guided by the intuition of ``learning faster than the jammer", where a synchronously updated coarse-grained spectrum prediction serves as an auxiliary task for the deep Q learning (DQN) based anti-jamming model. This helps the model identify a superior Q-function compared to standard DRL while significantly reducing the number of training episodes. Numerical results indicate that the proposed approach significantly accelerates the rate of convergence in model training, reducing the required training episodes by up to 70% compared to standard DRL. Additionally, it also achieves a 10% improvement in throughput over NE strategies, owing to the effective use of coarse-grained spectrum prediction.
Triangular arbitrage is a profitable trading strategy in financial markets that exploits discrepancies in currency exchange rates. Traditional methods for detecting triangular arbitrage opportunities, such as exhaustive search algorithms and linear programming solvers, often suffer from high computational complexity and may miss potential opportunities in dynamic markets. In this paper, we propose a novel approach to triangular arbitrage detection using Graph Neural Networks (GNNs). By representing the currency exchange network as a graph, we leverage the powerful representation and learning capabilities of GNNs to identify profitable arbitrage opportunities more efficiently. Specifically, we formulate the triangular arbitrage problem as a graph-based optimization task and design a GNN architecture that captures the complex relationships between currencies and exchange rates. We introduce a relaxed loss function to enable more flexible learning and integrate Deep Q-Learning principles to optimize the expected returns. Our experiments on a synthetic dataset demonstrate that the proposed GNN-based method achieves a higher average yield with significantly reduced computational time compared to traditional methods. This work highlights the potential of using GNNs for solving optimization problems in finance and provides a promising approach for real-time arbitrage detection in dynamic financial markets.
In this paper, we investigate an accurate synchronization between a physical network and its digital network twin (DNT), which serves as a virtual representation of the physical network. The considered network includes a set of base stations (BSs) that must allocate its limited spectrum resources to serve a set of users while also transmitting its partially observed physical network information to a cloud server to generate the DNT. Since the DNT can predict the physical network status based on its historical status, the BSs may not need to send their physical network information at each time slot, allowing them to conserve spectrum resources to serve the users. However, if the DNT does not receive the physical network information of the BSs over a large time period, the DNT's accuracy in representing the physical network may degrade. To this end, each BS must decide when to send the physical network information to the cloud server to update the DNT, while also determining the spectrum resource allocation policy for both DNT synchronization and serving the users. We formulate this resource allocation task as an optimization problem, aiming to maximize the total data rate of all users while minimizing the asynchronization between the physical network and the DNT. To address this problem, we propose a method based on the GRUs and the value decomposition network (VDN). Simulation results show that our GRU and VDN based algorithm improves the weighted sum of data rates and the similarity between the status of the DNT and the physical network by up to 28.96%, compared to a baseline method combining GRU with the independent Q learning.
Station-Keeping short-duration high-altitude balloons (HABs) in a region of interest is a challenging path-planning problem due to partially observable, complex, and dynamic wind flows. Deep reinforcement learning is a popular strategy for solving the station-keeping problem. A custom simulation environment was developed to train and evaluate Deep Q-Learning (DQN) for short-duration HAB agents in the simulation. To train the agents on realistic winds, synthetic wind forecasts were generated from aggregated historical radiosonde data to apply horizontal kinematics to simulated agents. The synthetic forecasts were closely correlated with ECWMF ERA5 Reanalysis forecasts, providing a realistic simulated wind field and seasonal and altitudinal variances between the wind models. DQN HAB agents were then trained and evaluated across different seasonal months. To highlight differences and trends in months with vastly different wind fields, a Forecast Score algorithm was introduced to independently classify forecasts based on wind diversity, and trends between station-keeping success and the Forecast Score were evaluated across all seasons.
A wide variety of resource allocation problems operate under resource constraints that are managed by a central arbitrator, with agents who evaluate and communicate preferences over these resources. We formulate this broad class of problems as Distributed Evaluation, Centralized Allocation (DECA) problems and propose methods to learn fair and efficient policies in centralized resource allocation. Our methods are applied to learning long-term fairness in a novel and general framework for fairness in multi-agent systems. We show three different methods based on Double Deep Q-Learning: (1) A joint weighted optimization of fairness and utility, (2) a split optimization, learning two separate Q-estimators for utility and fairness, and (3) an online policy perturbation to guide existing black-box utility functions toward fair solutions. Our methods outperform existing fair MARL approaches on multiple resource allocation domains, even when evaluated using diverse fairness functions, and allow for flexible online trade-offs between utility and fairness.
We present flow Q-learning (FQL), a simple and performant offline reinforcement learning (RL) method that leverages an expressive flow-matching policy to model arbitrarily complex action distributions in data. Training a flow policy with RL is a tricky problem, due to the iterative nature of the action generation process. We address this challenge by training an expressive one-step policy with RL, rather than directly guiding an iterative flow policy to maximize values. This way, we can completely avoid unstable recursive backpropagation, eliminate costly iterative action generation at test time, yet still mostly maintain expressivity. We experimentally show that FQL leads to strong performance across 73 challenging state- and pixel-based OGBench and D4RL tasks in offline RL and offline-to-online RL. Project page: https://seohong.me/projects/fql/
In this paper, we propose a novel censoring policy for energy-efficient transmissions in energy-harvesting sensors. The problem is formulated as an infinite-horizon Markov Decision Process (MDP). The objective to be optimized is the expected sum of the importance (utility) of all transmitted messages. Assuming that such importance can be evaluated at the transmitting node, we show that, under certain conditions on the battery model, the optimal censoring policy is a threshold function on the importance value. Specifically, messages are transmitted only if their importance is above a threshold whose value depends on the battery level. Exploiting this property, we propose a model-based stochastic scheme that approximates the optimal solution, with less computational complexity and faster convergence speed than a conventional Q-learning algorithm. Numerical experiments in single-hop and multi-hop networks confirm the analytical advantages of the proposed scheme.
Overestimation in single-agent reinforcement learning has been extensively studied. In contrast, overestimation in the multiagent setting has received comparatively little attention although it increases with the number of agents and leads to severe learning instability. Previous works concentrate on reducing overestimation in the estimation process of target Q-value. They ignore the follow-up optimization process of online Q-network, thus making it hard to fully address the complex multiagent overestimation problem. To solve this challenge, in this study, we first establish an iterative estimation-optimization analysis framework for multiagent value-mixing Q-learning. Our analysis reveals that multiagent overestimation not only comes from the computation of target Q-value but also accumulates in the online Q-network's optimization. Motivated by it, we propose the Dual Ensembled Multiagent Q-Learning with Hypernet Regularizer algorithm to tackle multiagent overestimation from two aspects. First, we extend the random ensemble technique into the estimation of target individual and global Q-values to derive a lower update target. Second, we propose a novel hypernet regularizer on hypernetwork weights and biases to constrain the optimization of online global Q-network to prevent overestimation accumulation. Extensive experiments in MPE and SMAC show that the proposed method successfully addresses overestimation across various tasks.