Markov decision processes (MDPs) are a standard model for sequential decision-making problems and are widely used across many scientific areas, including formal methods and artificial intelligence (AI). MDPs do, however, come with the restrictive assumption that the transition probabilities need to be precisely known. Robust MDPs (RMDPs) overcome this assumption by instead defining the transition probabilities to belong to some uncertainty set. We present a gentle survey on RMDPs, providing a tutorial covering their fundamentals. In particular, we discuss RMDP semantics and how to solve them by extending standard MDP methods such as value iteration and policy iteration. We also discuss how RMDPs relate to other models and how they are used in several contexts, including reinforcement learning and abstraction techniques. We conclude with some challenges for future work on RMDPs.
Reinforcement learning (RL) is a sub-domain of machine learning, mainly concerned with solving sequential decision-making problems by a learning agent that interacts with the decision environment to improve its behavior through the reward it receives from the environment. This learning paradigm is, however, well-known for being time-consuming due to the necessity of collecting a large amount of data, making RL suffer from sample inefficiency and difficult generalization. Furthermore, the construction of an explicit reward function that accounts for the trade-off between multiple desiderata of a decision problem is often a laborious task. These challenges have been recently addressed utilizing transfer and inverse reinforcement learning (T-IRL). In this regard, this paper is devoted to a comprehensive review of realizing the sample efficiency and generalization of RL algorithms through T-IRL. Following a brief introduction to RL, the fundamental T-IRL methods are presented and the most recent advancements in each research field have been extensively reviewed. Our findings denote that a majority of recent research works have dealt with the aforementioned challenges by utilizing human-in-the-loop and sim-to-real strategies for the efficient transfer of knowledge from source domains to the target domain under the transfer learning scheme. Under the IRL structure, training schemes that require a low number of experience transitions and extension of such frameworks to multi-agent and multi-intention problems have been the priority of researchers in recent years.
We consider fair resource allocation in sequential decision-making environments modeled as weakly coupled Markov decision processes, where resource constraints couple the action spaces of $N$ sub-Markov decision processes (sub-MDPs) that would otherwise operate independently. We adopt a fairness definition using the generalized Gini function instead of the traditional utilitarian (total-sum) objective. After introducing a general but computationally prohibitive solution scheme based on linear programming, we focus on the homogeneous case where all sub-MDPs are identical. For this case, we show for the first time that the problem reduces to optimizing the utilitarian objective over the class of "permutation invariant" policies. This result is particularly useful as we can exploit Whittle index policies in the restless bandits setting while, for the more general setting, we introduce a count-proportion-based deep reinforcement learning approach. Finally, we validate our theoretical findings with comprehensive experiments, confirming the effectiveness of our proposed method in achieving fairness.
Wireless ray-tracing (RT) is emerging as a key tool for three-dimensional (3D) wireless channel modeling, driven by advances in graphical rendering. Current approaches struggle to accurately model beyond 5G (B5G) network signaling, which often operates at higher frequencies and is more susceptible to environmental conditions and changes. Existing online learning solutions require real-time environmental supervision during training, which is both costly and incompatible with GPU-based processing. In response, we propose a novel approach that redefines ray trajectory generation as a sequential decision-making problem, leveraging generative models to jointly learn the optical, physical, and signal properties within each designated environment. Our work introduces the Scene-Aware Neural Decision Wireless Channel Raytracing Hierarchy (SANDWICH), an innovative offline, fully differentiable approach that can be trained entirely on GPUs. SANDWICH offers superior performance compared to existing online learning methods, outperforms the baseline by 4e^-2 radian in RT accuracy, and only fades 0.5 dB away from toplined channel gain estimation.
We focus on collaborative and federated black-box optimization (BBOpt), where agents optimize their heterogeneous black-box functions through collaborative sequential experimentation. From a Bayesian optimization perspective, we address the fundamental challenges of distributed experimentation, heterogeneity, and privacy within BBOpt, and propose three unifying frameworks to tackle these issues: (i) a global framework where experiments are centrally coordinated, (ii) a local framework that allows agents to make decisions based on minimal shared information, and (iii) a predictive framework that enhances local surrogates through collaboration to improve decision-making. We categorize existing methods within these frameworks and highlight key open questions to unlock the full potential of federated BBOpt. Our overarching goal is to shift federated learning from its predominantly descriptive/predictive paradigm to a prescriptive one, particularly in the context of BBOpt - an inherently sequential decision-making problem.
Deciding on appropriate mechanical ventilator management strategies significantly impacts the health outcomes for patients with respiratory diseases. Acute Respiratory Distress Syndrome (ARDS) is one such disease that requires careful ventilator operation to be effectively treated. In this work, we frame the management of ventilators for patients with ARDS as a sequential decision making problem using the Markov decision process framework. We implement and compare controllers based on clinical guidelines contained in the ARDSnet protocol, optimal control theory, and learned latent dynamics represented as neural networks. The Pulse Physiology Engine's respiratory dynamics simulator is used to establish a repeatable benchmark, gather simulated data, and quantitatively compare these controllers. We score performance in terms of measured improvement in established ARDS health markers (pertaining to improved respiratory rate, oxygenation, and vital signs). Our results demonstrate that techniques leveraging neural networks and optimal control can automatically discover effective ventilation management strategies without access to explicit ventilator management procedures or guidelines (such as those defined in the ARDSnet protocol).
Conventional methods for Bayesian optimization (BO) primarily involve one-step optimal decisions (e.g., maximizing expected improvement of the next step). To avoid myopic behavior, multi-step lookahead BO algorithms such as rollout strategies consider the sequential decision-making nature of BO, i.e., as a stochastic dynamic programming (SDP) problem, demonstrating promising results in recent years. However, owing to the curse of dimensionality, most of these methods make significant approximations or suffer scalability issues, e.g., being limited to two-step lookahead. This paper presents a novel reinforcement learning (RL)-based framework for multi-step lookahead BO in high-dimensional black-box optimization problems. The proposed method enhances the scalability and decision-making quality of multi-step lookahead BO by efficiently solving the SDP of the BO process in a near-optimal manner using RL. We first introduce an Attention-DeepSets encoder to represent the state of knowledge to the RL agent and employ off-policy learning to accelerate its initial training. We then propose a multi-task, fine-tuning procedure based on end-to-end (encoder-RL) on-policy learning. We evaluate the proposed method, EARL-BO (Encoder Augmented RL for Bayesian Optimization), on both synthetic benchmark functions and real-world hyperparameter optimization problems, demonstrating significantly improved performance compared to existing multi-step lookahead and high-dimensional BO methods.
Link prediction is a critical problem in graph learning with broad applications such as recommender systems and knowledge graph completion. Numerous research efforts have been directed at solving this problem, including approaches based on similarity metrics and Graph Neural Networks (GNN). However, most existing solutions are still rooted in conventional supervised learning, which makes it challenging to adapt over time to changing customer interests and to address the inherent dilemma of exploitation versus exploration in link prediction. To tackle these challenges, this paper reformulates link prediction as a sequential decision-making process, where each link prediction interaction occurs sequentially. We propose a novel fusion algorithm, PRB (PageRank Bandits), which is the first to combine contextual bandits with PageRank for collaborative exploitation and exploration. We also introduce a new reward formulation and provide a theoretical performance guarantee for PRB. Finally, we extensively evaluate PRB in both online and offline settings, comparing it with bandit-based and graph-based methods. The empirical success of PRB demonstrates the value of the proposed fusion approach. Our code is released at https://github.com/jiaruzouu/PRB.
Recent years have witnessed the rapid development of Neuro-Symbolic (NeSy) AI systems, which integrate symbolic reasoning into deep neural networks. However, most of the existing benchmarks for NeSy AI fail to provide long-horizon reasoning tasks with complex multi-agent interactions. Furthermore, they are usually constrained by fixed and simplistic logical rules over limited entities, making them far from real-world complexities. To address these crucial gaps, we introduce LogiCity, the first simulator based on customizable first-order logic (FOL) for an urban-like environment with multiple dynamic agents. LogiCity models diverse urban elements using semantic and spatial concepts, such as IsAmbulance(X) and IsClose(X, Y). These concepts are used to define FOL rules that govern the behavior of various agents. Since the concepts and rules are abstractions, they can be universally applied to cities with any agent compositions, facilitating the instantiation of diverse scenarios. Besides, a key feature of LogiCity is its support for user-configurable abstractions, enabling customizable simulation complexities for logical reasoning. To explore various aspects of NeSy AI, LogiCity introduces two tasks, one features long-horizon sequential decision-making, and the other focuses on one-step visual reasoning, varying in difficulty and agent behaviors. Our extensive evaluation reveals the advantage of NeSy frameworks in abstract reasoning. Moreover, we highlight the significant challenges of handling more complex abstractions in long-horizon multi-agent scenarios or under high-dimensional, imbalanced data. With its flexible design, various features, and newly raised challenges, we believe LogiCity represents a pivotal step forward in advancing the next generation of NeSy AI. All the code and data are open-sourced at our website.
Unit commitment (UC) optimizes the start-up and shutdown schedules of generating units to meet load demand while minimizing costs. However, the increasing integration of renewable energy introduces uncertainties for real-time scheduling. Existing solutions face limitations both in modeling and algorithmic design. At the modeling level, they fail to incorporate widely adopted virtual power plants (VPPs) as flexibility resources, missing the opportunity to proactively mitigate potential real-time imbalances or ramping constraints through foresight-seeing decision-making. At the algorithmic level, existing probabilistic optimization, multi-stage approaches, and machine learning, face challenges in computational complexity and adaptability. To address these challenges, this study proposes a novel two-stage UC framework that incorporates foresight-seeing sequential decision-making in both day-ahead and real-time scheduling, leveraging VPPs as flexibility resources to proactively reserve capacity and ramping flexibility for upcoming renewable energy uncertainties over several hours. In particular, we develop quantum reinforcement learning (QRL) algorithms that integrate the foresight-seeing sequential decision-making and scalable computation advantages of deep reinforcement learning (DRL) with the parallel and high-efficiency search capabilities of quantum computing. Experimental results demonstrate that the proposed QRL-based approach outperforms in computational efficiency, real-time responsiveness, and solution quality.
This paper considers the problem of annotating datapoints using an expert with only a few annotation rounds in a label-scarce setting. We propose soliciting reliable feedback on difficulty in annotating a datapoint from the expert in addition to ground truth label. Existing literature in active learning or coreset selection turns out to be less relevant to our setting since they presume the existence of a reliable trained model, which is absent in the label-scarce regime. However, the literature on coreset selection emphasizes the presence of difficult data points in the training set to perform supervised learning in downstream tasks (Mindermann et al., 2022). Therefore, for a given fixed annotation budget of $\mathsf{T}$ rounds, we model the sequential decision-making problem of which (difficult) datapoints to choose for annotation in a sparse linear bandits framework with the constraint that no arm can be pulled more than once (blocking constraint). With mild assumptions on the datapoints, our (computationally efficient) Explore-Then-Commit algorithm BSLB achieves a regret guarantee of $\widetilde{\mathsf{O}}(k^{\frac{1}{3}} \mathsf{T}^{\frac{2}{3}} +k^{-\frac{1}{2}} \beta_k + k^{-\frac{1}{12}} \beta_k^{\frac{1}{2}}\mathsf{T}^{\frac{5}{6}})$ where the unknown parameter vector has tail magnitude $\beta_k$ at sparsity level $k$. To this end, we show offline statistical guarantees of Lasso estimator with mild Restricted Eigenvalue (RE) condition that is also robust to sparsity. Finally, we propose a meta-algorithm C-BSLB that does not need knowledge of the optimal sparsity parameters at a no-regret cost. We demonstrate the efficacy of our BSLB algorithm for annotation in the label-scarce setting for an image classification task on the PASCAL-VOC dataset, where we use real-world annotation difficulty scores.
Reinforcement learning (RL) is a promising approach for aligning large language models (LLMs) knowledge with sequential decision-making tasks. However, few studies have thoroughly investigated the impact on LLM agents capabilities of fine-tuning them with RL in a specific environment. In this paper, we propose a novel framework to analyze the sensitivity of LLMs to prompt formulations following RL training in a textual environment. Our findings reveal that the performance of LLMs degrades when faced with prompt formulations different from those used during the RL training phase. Besides, we analyze the source of this sensitivity by examining the model's internal representations and salient tokens. Finally, we propose to use a contrastive loss to mitigate this sensitivity and improve the robustness and generalization capabilities of LLMs.
Thompson sampling is one of the most popular learning algorithms for online sequential decision-making problems and has rich real-world applications. However, current Thompson sampling algorithms are limited by the assumption that the rewards received are uncorrupted, which may not be true in real-world applications where adversarial reward poisoning exists. To make Thompson sampling more reliable, we want to make it robust against adversarial reward poisoning. The main challenge is that one can no longer compute the actual posteriors for the true reward, as the agent can only observe the rewards after corruption. In this work, we solve this problem by computing pseudo-posteriors that are less likely to be manipulated by the attack. We propose robust algorithms based on Thompson sampling for the popular stochastic and contextual linear bandit settings in both cases where the agent is aware or unaware of the budget of the attacker. We theoretically show that our algorithms guarantee near-optimal regret under any attack strategy.
Masked prediction has emerged as a promising pretraining paradigm in offline reinforcement learning (RL) due to its versatile masking schemes, enabling flexible inference across various downstream tasks with a unified model. Despite the versatility of masked prediction, it remains unclear how to balance the learning of skills at different levels of complexity. To address this, we propose CurrMask, a curriculum masking pretraining paradigm for sequential decision making. Motivated by how humans learn by organizing knowledge in a curriculum, CurrMask adjusts its masking scheme during pretraining for learning versatile skills. Through extensive experiments, we show that CurrMask exhibits superior zero-shot performance on skill prompting tasks, goal-conditioned planning tasks, and competitive finetuning performance on offline RL tasks. Additionally, our analysis of training dynamics reveals that CurrMask gradually acquires skills of varying complexity by dynamically adjusting its masking scheme.
The multi-armed bandit (MAB) problem is a foundational framework in sequential decision-making under uncertainty, extensively studied for its applications in areas such as clinical trials, online advertising, and resource allocation. Traditional MAB formulations, however, do not adequately capture scenarios where decisions are structured hierarchically, involve multi-level constraints, or feature context-dependent action spaces. In this paper, we introduce the hierarchical constrained bandits (HCB) framework, which extends the contextual bandit problem to incorporate hierarchical decision structures and multi-level constraints. We propose the hierarchical constrained upper confidence bound (HC-UCB) algorithm, designed to address the complexities of the HCB problem by leveraging confidence bounds within a hierarchical setting. Our theoretical analysis establishes sublinear regret bounds for HC-UCB and provides high-probability guarantees for constraint satisfaction at all hierarchical levels. Furthermore, we derive a minimax lower bound on the regret for the HCB problem, demonstrating the near-optimality of our algorithm. The results are significant for real-world applications where decision-making processes are inherently hierarchical and constrained, offering a robust and efficient solution that balances exploration and exploitation across multiple levels of decision-making.
Behavioral diversity, expert imitation, fairness, safety goals and others give rise to preferences in sequential decision making domains that do not decompose additively across time. We introduce the class of convex Markov games that allow general convex preferences over occupancy measures. Despite infinite time horizon and strictly higher generality than Markov games, pure strategy Nash equilibria exist. Furthermore, equilibria can be approximated empirically by performing gradient descent on an upper bound of exploitability. Our experiments reveal novel solutions to classic repeated normal-form games, find fair solutions in a repeated asymmetric coordination game, and prioritize safe long-term behavior in a robot warehouse environment. In the prisoner's dilemma, our algorithm leverages transient imitation to find a policy profile that deviates from observed human play only slightly, yet achieves higher per-player utility while also being three orders of magnitude less exploitable.
Large-scale generative models have achieved remarkable success in a number of domains. However, for sequential decision-making problems, such as robotics, action-labelled data is often scarce and therefore scaling-up foundation models for decision-making remains a challenge. A potential solution lies in leveraging widely-available unlabelled videos to train world models that simulate the consequences of actions. If the world model is accurate, it can be used to optimize decision-making in downstream tasks. Image-to-video diffusion models are already capable of generating highly realistic synthetic videos. However, these models are not action-conditioned, and the most powerful models are closed-source which means they cannot be finetuned. In this work, we propose to adapt pretrained video diffusion models to action-conditioned world models, without access to the parameters of the pretrained model. Our approach, AVID, trains an adapter on a small domain-specific dataset of action-labelled videos. AVID uses a learned mask to modify the intermediate outputs of the pretrained model and generate accurate action-conditioned videos. We evaluate AVID on video game and real-world robotics data, and show that it outperforms existing baselines for diffusion model adaptation.1 Our results demonstrate that if utilized correctly, pretrained video models have the potential to be powerful tools for embodied AI.
We address the challenge of explaining counterfactual outcomes in multi-agent Markov decision processes. In particular, we aim to explain the total counterfactual effect of an agent's action on the outcome of a realized scenario through its influence on the environment dynamics and the agents' behavior. To achieve this, we introduce a novel causal explanation formula that decomposes the counterfactual effect by attributing to each agent and state variable a score reflecting their respective contributions to the effect. First, we show that the total counterfactual effect of an agent's action can be decomposed into two components: one measuring the effect that propagates through all subsequent agents' actions and another related to the effect that propagates through the state transitions. Building on recent advancements in causal contribution analysis, we further decompose these two effects as follows. For the former, we consider agent-specific effects -- a causal concept that quantifies the counterfactual effect of an agent's action that propagates through a subset of agents. Based on this notion, we use Shapley value to attribute the effect to individual agents. For the latter, we consider the concept of structure-preserving interventions and attribute the effect to state variables based on their "intrinsic" contributions. Through extensive experimentation, we demonstrate the interpretability of our approach in a Gridworld environment with LLM-assisted agents and a sepsis management simulator.
The past years have seen Large Language Models (LLMs) strive not only as generative models but also as agents solving textual sequential decision-making tasks. When facing complex environments where their zero-shot abilities are insufficient, recent work showed online Reinforcement Learning (RL) could be used for the LLM agent to discover and learn efficient strategies interactively. However, most prior work sticks to on-policy algorithms, which greatly reduces the scope of methods such agents could use for both exploration and exploitation, such as experience replay and hindsight relabeling. Yet, such methods may be key for LLM learning agents, and in particular when designing autonomous intrinsically motivated agents sampling and pursuing their own goals (i.e. autotelic agents). This paper presents and studies an adaptation of Soft Actor-Critic and hindsight relabeling to LLM agents. Our method not only paves the path towards autotelic LLM agents that learn online but can also outperform on-policy methods in more classic multi-goal RL environments.
Wireless Networked Control Systems (WNCSs) are essential to Industry 4.0, enabling flexible control in applications, such as drone swarms and autonomous robots. The interdependence between communication and control requires integrated design, but traditional methods treat them separately, leading to inefficiencies. Current codesign approaches often rely on simplified models, focusing on single-loop or independent multi-loop systems. However, large-scale WNCSs face unique challenges, including coupled control loops, time-correlated wireless channels, trade-offs between sensing and control transmissions, and significant computational complexity. To address these challenges, we propose a practical WNCS model that captures correlated dynamics among multiple control loops with spatially distributed sensors and actuators sharing limited wireless resources over multi-state Markov block-fading channels. We formulate the codesign problem as a sequential decision-making task that jointly optimizes scheduling and control inputs across estimation, control, and communication domains. To solve this problem, we develop a Deep Reinforcement Learning (DRL) algorithm that efficiently handles the hybrid action space, captures communication-control correlations, and ensures robust training despite sparse cross-domain variables and floating control inputs. Extensive simulations show that the proposed DRL approach outperforms benchmarks and solves the large-scale WNCS codesign problem, providing a scalable solution for industrial automation.