Offline meta reinforcement learning (OMRL) has emerged as a promising approach for interaction avoidance and strong generalization performance by leveraging pre-collected data and meta-learning techniques. Previous context-based approaches predominantly rely on the intuition that alternating optimization between the context encoder and the policy can lead to performance improvements, as long as the context encoder follows the principle of maximizing the mutual information between the task variable $M$ and its latent representation $Z$ ($I(Z;M)$) while the policy adopts the standard offline reinforcement learning (RL) algorithms conditioning on the learned task representation.Despite promising results, the theoretical justification of performance improvements for such intuition remains underexplored.Inspired by the return discrepancy scheme in the model-based RL field, we find that the previous optimization framework can be linked with the general RL objective of maximizing the expected return, thereby explaining performance improvements. Furthermore, after scrutinizing this optimization framework, we observe that the condition for monotonic performance improvements does not consider the variation of the task representation. When these variations are considered, the previously established condition may no longer be sufficient to ensure monotonicity, thereby impairing the optimization process.We name this issue task representation shift and theoretically prove that the monotonic performance improvements can be guaranteed with appropriate context encoder updates.Our work opens up a new avenue for OMRL, leading to a better understanding between the task representation and performance improvements.
We introduce the ``soft Deep MaxPain'' (softDMP) algorithm, which integrates the optimization of long-term policy entropy into reward-punishment reinforcement learning objectives. Our motivation is to facilitate a smoother variation of operators utilized in the updating of action values beyond traditional ``max'' and ``min'' operators, where the goal is enhancing sample efficiency and robustness. We also address two unresolved issues from the previous Deep MaxPain method. Firstly, we investigate how the negated (``flipped'') pain-seeking sub-policy, derived from the punishment action value, collaborates with the ``min'' operator to effectively learn the punishment module and how softDMP's smooth learning operator provides insights into the ``flipping'' trick. Secondly, we tackle the challenge of data collection for learning the punishment module to mitigate inconsistencies arising from the involvement of the ``flipped'' sub-policy (pain-avoidance sub-policy) in the unified behavior policy. We empirically explore the first issue in two discrete Markov Decision Process (MDP) environments, elucidating the crucial advancements of the DMP approach and the necessity for soft treatments on the hard operators. For the second issue, we propose a probabilistic classifier based on the ratio of the pain-seeking sub-policy to the sum of the pain-seeking and goal-reaching sub-policies. This classifier assigns roll-outs to separate replay buffers for updating reward and punishment action-value functions, respectively. Our framework demonstrates superior performance in Turtlebot 3's maze navigation tasks under the ROS Gazebo simulation.
Multi-agent reinforcement learning (MARL) algorithms have accomplished remarkable breakthroughs in solving large-scale decision-making tasks. Nonetheless, most existing MARL algorithms are model-free, limiting sample efficiency and hindering their applicability in more challenging scenarios. In contrast, model-based reinforcement learning (MBRL), particularly algorithms integrating planning, such as MuZero, has demonstrated superhuman performance with limited data in many tasks. Hence, we aim to boost the sample efficiency of MARL by adopting model-based approaches. However, incorporating planning and search methods into multi-agent systems poses significant challenges. The expansive action space of multi-agent systems often necessitates leveraging the nearly-independent property of agents to accelerate learning. To tackle this issue, we propose the MAZero algorithm, which combines a centralized model with Monte Carlo Tree Search (MCTS) for policy search. We design a novel network structure to facilitate distributed execution and parameter sharing. To enhance search efficiency in deterministic environments with sizable action spaces, we introduce two novel techniques: Optimistic Search Lambda (OS($\lambda$)) and Advantage-Weighted Policy Optimization (AWPO). Extensive experiments on the SMAC benchmark demonstrate that MAZero outperforms model-free approaches in terms of sample efficiency and provides comparable or better performance than existing model-based methods in terms of both sample and computational efficiency. Our code is available at https://github.com/liuqh16/MAZero.
Reinforcement Learning (RL) for control has become increasingly popular due to its ability to learn rich feedback policies that take into account uncertainty and complex representations of the environment. When considering safety constraints, constrained optimization approaches, where agents are penalized for constraint violations, are commonly used. In such methods, if agents are initialized in, or must visit, states where constraint violation might be inevitable, it is unclear how much they should be penalized. We address this challenge by formulating a constraint on the counterfactual harm of the learned policy compared to a default, safe policy. In a philosophical sense this formulation only penalizes the learner for constraint violations that it caused; in a practical sense it maintains feasibility of the optimal control problem. We present simulation studies on a rover with uncertain road friction and a tractor-trailer parking environment that demonstrate our constraint formulation enables agents to learn safer policies than contemporary constrained RL methods.
Animals and robots exist in a physical world and must coordinate their bodies to achieve behavioral objectives. With recent developments in deep reinforcement learning, it is now possible for scientists and engineers to obtain sensorimotor strategies (policies) for specific tasks using physically simulated bodies and environments. However, the utility of these methods goes beyond the constraints of a specific task; they offer an exciting framework for understanding the organization of an animal sensorimotor system in connection to its morphology and physical interaction with the environment, as well as for deriving general design rules for sensing and actuation in robotic systems. Algorithms and code implementing both learning agents and environments are increasingly available, but the basic assumptions and choices that go into the formulation of an embodied feedback control problem using deep reinforcement learning may not be immediately apparent. Here, we present a concise exposition of the mathematical and algorithmic aspects of model-free reinforcement learning, specifically through the use of \textit{actor-critic} methods, as a tool for investigating the feedback control underlying animal and robotic behavior.
Over the last decade, data-driven methods have surged in popularity, emerging as valuable tools for control theory. As such, neural network approximations of control feedback laws, system dynamics, and even Lyapunov functions have attracted growing attention. With the ascent of learning based control, the need for accurate, fast, and easy-to-use benchmarks has increased. In this work, we present the first learning-based environment for boundary control of PDEs. In our benchmark, we introduce three foundational PDE problems - a 1D transport PDE, a 1D reaction-diffusion PDE, and a 2D Navier-Stokes PDE - whose solvers are bundled in an user-friendly reinforcement learning gym. With this gym, we then present the first set of model-free, reinforcement learning algorithms for solving this series of benchmark problems, achieving stability, although at a higher cost compared to model-based PDE backstepping. With the set of benchmark environments and detailed examples, this work significantly lowers the barrier to entry for learning-based PDE control - a topic largely unexplored by the data-driven control community. The entire benchmark is available on Github along with detailed documentation and the presented reinforcement learning models are open sourced.
This paper presents a study of robust policy networks in deep reinforcement learning. We investigate the benefits of policy parameterizations that naturally satisfy constraints on their Lipschitz bound, analyzing their empirical performance and robustness on two representative problems: pendulum swing-up and Atari Pong. We illustrate that policy networks with smaller Lipschitz bounds are more robust to disturbances, random noise, and targeted adversarial attacks than unconstrained policies composed of vanilla multi-layer perceptrons or convolutional neural networks. However, the structure of the Lipschitz layer is important. We find that the widely-used method of spectral normalization is too conservative and severely impacts clean performance, whereas more expressive Lipschitz layers such as the recently-proposed Sandwich layer can achieve improved robustness without sacrificing clean performance.
As large language models (LLMs) continue to grow by scaling laws, reinforcement learning from human feedback (RLHF) has gained significant attention due to its outstanding performance. However, unlike pretraining or fine-tuning a single model, scaling reinforcement learning from human feedback (RLHF) for training large language models poses coordination challenges across four models. We present OpenRLHF, an open-source framework enabling efficient RLHF scaling. Unlike existing RLHF frameworks that co-locate four models on the same GPUs, OpenRLHF re-designs scheduling for the models beyond 70B parameters using Ray, vLLM, and DeepSpeed, leveraging improved resource utilization and diverse training approaches. Integrating seamlessly with Hugging Face, OpenRLHF provides an out-of-the-box solution with optimized algorithms and launch scripts, which ensures user-friendliness. OpenRLHF implements RLHF, DPO, rejection sampling, and other alignment techniques. Empowering state-of-the-art LLM development, OpenRLHF's code is available at \url{https://github.com/OpenRLHF/OpenRLHF}.
Large language models (LLMs) have received considerable attention recently due to their outstanding comprehension and reasoning capabilities, leading to great progress in many fields. The advancement of LLM techniques also offers promising opportunities to automate many tasks in the telecommunication (telecom) field. After pre-training and fine-tuning, LLMs can perform diverse downstream tasks based on human instructions, paving the way to artificial general intelligence (AGI)-enabled 6G. Given the great potential of LLM technologies, this work aims to provide a comprehensive overview of LLM-enabled telecom networks. In particular, we first present LLM fundamentals, including model architecture, pre-training, fine-tuning, inference and utilization, model evaluation, and telecom deployment. Then, we introduce LLM-enabled key techniques and telecom applications in terms of generation, classification, optimization, and prediction problems. Specifically, the LLM-enabled generation applications include telecom domain knowledge, code, and network configuration generation. After that, the LLM-based classification applications involve network security, text, image, and traffic classification problems. Moreover, multiple LLM-enabled optimization techniques are introduced, such as automated reward function design for reinforcement learning and verbal reinforcement learning. Furthermore, for LLM-aided prediction problems, we discussed time-series prediction models and multi-modality prediction problems for telecom. Finally, we highlight the challenges and identify the future directions of LLM-enabled telecom networks.
We consider a constrained Markov Decision Problem (CMDP) where the goal of an agent is to maximize the expected discounted sum of rewards over an infinite horizon while ensuring that the expected discounted sum of costs exceeds a certain threshold. Building on the idea of momentum-based acceleration, we develop the Primal-Dual Accelerated Natural Policy Gradient (PD-ANPG) algorithm that ensures an $\epsilon$ global optimality gap and $\epsilon$ constraint violation with $\tilde{\mathcal{O}}((1-\gamma)^{-7}\epsilon^{-2})$ sample complexity for general parameterized policies where $\gamma$ denotes the discount factor. This improves the state-of-the-art sample complexity in general parameterized CMDPs by a factor of $\mathcal{O}((1-\gamma)^{-1}\epsilon^{-2})$ and achieves the theoretical lower bound in $\epsilon^{-1}$.
Energy storage devices, such as batteries, thermal energy storages, and hydrogen systems, can help mitigate climate change by ensuring a more stable and sustainable power supply. To maximize the effectiveness of such energy storage, determining the appropriate charging and discharging amounts for each time period is crucial. Reinforcement learning is preferred over traditional optimization for the control of energy storage due to its ability to adapt to dynamic and complex environments. However, the continuous nature of charging and discharging levels in energy storage poses limitations for discrete reinforcement learning, and time-varying feasible charge-discharge range based on state of charge (SoC) variability also limits the conventional continuous reinforcement learning. In this paper, we propose a continuous reinforcement learning approach that takes into account the time-varying feasible charge-discharge range. An additional objective function was introduced for learning the feasible action range for each time period, supplementing the objectives of training the actor for policy learning and the critic for value learning. This actively promotes the utilization of energy storage by preventing them from getting stuck in suboptimal states, such as continuous full charging or discharging. This is achieved through the enforcement of the charging and discharging levels into the feasible action range. The experimental results demonstrated that the proposed method further maximized the effectiveness of energy storage by actively enhancing its utilization.
Observing celestial objects and advancing our scientific knowledge about them involves tedious planning, scheduling, data collection and data post-processing. Many of these operational aspects of astronomy are guided and executed by expert astronomers. Reinforcement learning is a mechanism where we (as humans and astronomers) can teach agents of artificial intelligence to perform some of these tedious tasks. In this paper, we will present a state of the art overview of reinforcement learning and how it can benefit astronomy.
The software industry is experiencing a surge in the adoption of Continuous Integration (CI) practices, both in commercial and open-source environments. CI practices facilitate the seamless integration of code changes by employing automated building and testing processes. Some frameworks, such as Travis CI and GitHub Actions have significantly contributed to simplifying and enhancing the CI process, rendering it more accessible and efficient for development teams. Despite the availability these CI tools , developers continue to encounter difficulties in accurately flagging commits as either suitable for CI execution or as candidates for skipping especially for large projects with many dependencies. Inaccurate flagging of commits can lead to resource-intensive test and build processes, as even minor commits may inadvertently trigger the Continuous Integration process. The problem of detecting CI-skip commits, can be modeled as binary classification task where we decide to either build a commit or to skip it. This study proposes a novel solution that leverages Deep Reinforcement Learning techniques to construct an optimal Decision Tree classifier that addresses the imbalanced nature of the data. We evaluate our solution by running a within and a cross project validation benchmark on diverse range of Open-Source projects hosted on GitHub which showcased superior results when compared with existing state-of-the-art methods.
We show that discounted methods for solving continuing reinforcement learning problems can perform significantly better if they center their rewards by subtracting out the rewards' empirical average. The improvement is substantial at commonly used discount factors and increases further as the discount factor approaches one. In addition, we show that if a problem's rewards are shifted by a constant, then standard methods perform much worse, whereas methods with reward centering are unaffected. Estimating the average reward is straightforward in the on-policy setting; we propose a slightly more sophisticated method for the off-policy setting. Reward centering is a general idea, so we expect almost every reinforcement-learning algorithm to benefit by the addition of reward centering.
Chaos-based reinforcement learning (CBRL) is a method in which the agent's internal chaotic dynamics drives exploration. This approach offers a model for considering how the biological brain can create variability in its behavior and learn in an exploratory manner. At the same time, it is a learning model that has the ability to automatically switch between exploration and exploitation modes and the potential to realize higher explorations that reflect what it has learned so far. However, the learning algorithms in CBRL have not been well-established in previous studies and have yet to incorporate recent advances in reinforcement learning. This study introduced Twin Delayed Deep Deterministic Policy Gradients (TD3), which is one of the state-of-the-art deep reinforcement learning algorithms that can treat deterministic and continuous action spaces, to CBRL. The validation results provide several insights. First, TD3 works as a learning algorithm for CBRL in a simple goal-reaching task. Second, CBRL agents with TD3 can autonomously suppress their exploratory behavior as learning progresses and resume exploration when the environment changes. Finally, examining the effect of the agent's chaoticity on learning shows that extremely strong chaos negatively impacts the flexible switching between exploration and exploitation.
Value function factorization methods are commonly used in cooperative multi-agent reinforcement learning, with QMIX receiving significant attention. Many QMIX-based methods introduce monotonicity constraints between the joint action value and individual action values to achieve decentralized execution. However, such constraints limit the representation capacity of value factorization, restricting the joint action values it can represent and hindering the learning of the optimal policy. To address this challenge, we propose the Potentially Optimal Joint Actions Weighted QMIX (POWQMIX) algorithm, which recognizes the potentially optimal joint actions and assigns higher weights to the corresponding losses of these joint actions during training. We theoretically prove that with such a weighted training approach the optimal policy is guaranteed to be recovered. Experiments in matrix games, difficulty-enhanced predator-prey, and StarCraft II Multi-Agent Challenge environments demonstrate that our algorithm outperforms the state-of-the-art value-based multi-agent reinforcement learning methods.
Learning from expert demonstrations to flexibly program an autonomous system with complex behaviors or to predict an agent's behavior is a powerful tool, especially in collaborative control settings. A common method to solve this problem is inverse reinforcement learning (IRL), where the observed agent, e.g., a human demonstrator, is assumed to behave according to the optimization of an intrinsic cost function that reflects its intent and informs its control actions. While the framework is expressive, it is also computationally demanding and generally lacks convergence guarantees. We therefore propose a novel, stability-certified IRL approach by reformulating the cost function inference problem to learning control Lyapunov functions (CLF) from demonstrations data. By additionally exploiting closed-form expressions for associated control policies, we are able to efficiently search the space of CLFs by observing the attractor landscape of the induced dynamics. For the construction of the inverse optimal CLFs, we use a Sum of Squares and formulate a convex optimization problem. We present a theoretical analysis of the optimality properties provided by the CLF and evaluate our approach using both simulated and real-world data.
GPU-based heterogeneous architectures are now commonly used in HPC clusters. Due to their architectural simplicity specialized for data-level parallelism, GPUs can offer much higher computational throughput and memory bandwidth than CPUs in the same generation do. However, as the available resources in GPUs have increased exponentially over the past decades, it has become increasingly difficult for a single program to fully utilize them. As a consequence, the industry has started supporting several resource partitioning features in order to improve the resource utilization by co-scheduling multiple programs on the same GPU die at the same time. Driven by the technological trend, this paper focuses on hierarchical resource partitioning on modern GPUs, and as an example, we utilize a combination of two different features available on recent NVIDIA GPUs in a hierarchical manner: MPS (Multi-Process Service), a finer-grained logical partitioning; and MIG (Multi-Instance GPU), a coarse-grained physical partitioning. We propose a method for comprehensively co-optimizing the setup of hierarchical partitioning and the selection of co-scheduling groups from a given set of jobs, based on reinforcement learning using their profiles. Our thorough experimental results demonstrate that our approach can successfully set up job concurrency, partitioning, and co-scheduling group selections simultaneously. This results in a maximum throughput improvement by a factor of 1.87 compared to the time-sharing scheduling.
This paper contributes to the existing literature on hedging American options with Deep Reinforcement Learning (DRL). The study first investigates hyperparameter impact on hedging performance, considering learning rates, training episodes, neural network architectures, training steps, and transaction cost penalty functions. Results highlight the importance of avoiding certain combinations, such as high learning rates with a high number of training episodes or low learning rates with few training episodes and emphasize the significance of utilizing moderate values for optimal outcomes. Additionally, the paper warns against excessive training steps to prevent instability and demonstrates the superiority of a quadratic transaction cost penalty function over a linear version. This study then expands upon the work of Pickard et al. (2024), who utilize a Chebyshev interpolation option pricing method to train DRL agents with market calibrated stochastic volatility models. While the results of Pickard et al. (2024) showed that these DRL agents achieve satisfactory performance on empirical asset paths, this study introduces a novel approach where new agents at weekly intervals to newly calibrated stochastic volatility models. Results show DRL agents re-trained using weekly market data surpass the performance of those trained solely on the sale date. Furthermore, the paper demonstrates that both single-train and weekly-train DRL agents outperform the Black-Scholes Delta method at transaction costs of 1% and 3%. This practical relevance suggests that practitioners can leverage readily available market data to train DRL agents for effective hedging of options in their portfolios.
This paper proposes a framework for training Reinforcement Learning agents using Python in conjunction with Simulink models. Leveraging Python's superior customization options and popular libraries like Stable Baselines3, we aim to bridge the gap between the established Simulink environment and the flexibility of Python for training bleeding edge agents. Our approach is demonstrated on the Quanser Aero 2, a versatile dual-rotor helicopter. We show that policies trained on Simulink models can be seamlessly transferred to the real system, enabling efficient development and deployment of Reinforcement Learning agents for control tasks. Through systematic integration steps, including C-code generation from Simulink, DLL compilation, and Python interface development, we establish a robust framework for training agents on Simulink models. Experimental results demonstrate the effectiveness of our approach, surpassing previous efforts and highlighting the potential of combining Simulink with Python for Reinforcement Learning research and applications.