Meta-reinforcement learning (meta-RL) is a promising framework for tackling challenging domains requiring efficient exploration. Existing meta-RL algorithms are characterized by low sample efficiency, and mostly focus on low-dimensional task distributions. In parallel, model-based RL methods have been successful in solving partially observable MDPs, of which meta-RL is a special case. In this work, we leverage this success and propose a new model-based approach to meta-RL, based on elements from existing state-of-the-art model-based and meta-RL methods. We demonstrate the effectiveness of our approach on common meta-RL benchmark domains, attaining greater return with better sample efficiency (up to $15\times$) while requiring very little hyperparameter tuning. In addition, we validate our approach on a slate of more challenging, higher-dimensional domains, taking a step towards real-world generalizing agents.
A key aspect of intelligence is the ability to demonstrate a broad spectrum of behaviors for adapting to unexpected situations. Over the past decade, advancements in deep reinforcement learning have led to groundbreaking achievements to solve complex continuous control tasks. However, most approaches return only one solution specialized for a specific problem. We introduce Quality-Diversity Actor-Critic (QDAC), an off-policy actor-critic deep reinforcement learning algorithm that leverages a value function critic and a successor features critic to learn high-performing and diverse behaviors. In this framework, the actor optimizes an objective that seamlessly unifies both critics using constrained optimization to (1) maximize return, while (2) executing diverse skills. Compared with other Quality-Diversity methods, QDAC achieves significantly higher performance and more diverse behaviors on six challenging continuous control locomotion tasks. We also demonstrate that we can harness the learned skills to adapt better than other baselines to five perturbed environments. Finally, qualitative analyses showcase a range of remarkable behaviors: adaptive-intelligent-robotics.github.io/QDAC.
There is a growing interest in the application of Reinforcement Learning (RL) techniques to AI planning with the aim to come up with general policies. Typically, the mapping of the transition model of AI planning to the state transition system of a Markov Decision Process is established by assuming a one-to-one correspondence of the respective action spaces. In this paper, we introduce the concept of meta-operator as the result of simultaneously applying multiple planning operators, and we show that including meta-operators in the RL action space enables new planning perspectives to be addressed using RL, such as parallel planning. Our research aims to analyze the performance and complexity of including meta-operators in the RL process, concretely in domains where satisfactory outcomes have not been previously achieved using usual generalized planning models. The main objective of this article is thus to pave the way towards a redefinition of the RL action space in a manner that is more closely aligned with the planning perspective.
The Intelligent Transportation System (ITS) environment is known to be dynamic and distributed, where participants (vehicle users, operators, etc.) have multiple, changing and possibly conflicting objectives. Although Reinforcement Learning (RL) algorithms are commonly applied to optimize ITS applications such as resource management and offloading, most RL algorithms focus on single objectives. In many situations, converting a multi-objective problem into a single-objective one is impossible, intractable or insufficient, making such RL algorithms inapplicable. We propose a multi-objective, multi-agent reinforcement learning (MARL) algorithm with high learning efficiency and low computational requirements, which automatically triggers adaptive few-shot learning in a dynamic, distributed and noisy environment with sparse and delayed reward. We test our algorithm in an ITS environment with edge cloud computing. Empirical results show that the algorithm is quick to adapt to new environments and performs better in all individual and system metrics compared to the state-of-the-art benchmark. Our algorithm also addresses various practical concerns with its modularized and asynchronous online training method. In addition to the cloud simulation, we test our algorithm on a single-board computer and show that it can make inference in 6 milliseconds.
Deep reinforcement learning (DRL) has shown significant promise for uncovering sophisticated control policies that interact in environments with complicated dynamics, such as stabilizing the magnetohydrodynamics of a tokamak fusion reactor or minimizing the drag force exerted on an object in a fluid flow. However, these algorithms require an abundance of training examples and may become prohibitively expensive for many applications. In addition, the reliance on deep neural networks often results in an uninterpretable, black-box policy that may be too computationally expensive to use with certain embedded systems. Recent advances in sparse dictionary learning, such as the sparse identification of nonlinear dynamics (SINDy), have shown promise for creating efficient and interpretable data-driven models in the low-data regime. In this work we introduce SINDy-RL, a unifying framework for combining SINDy and DRL to create efficient, interpretable, and trustworthy representations of the dynamics model, reward function, and control policy. We demonstrate the effectiveness of our approaches on benchmark control environments and challenging fluids problems. SINDy-RL achieves comparable performance to state-of-the-art DRL algorithms using significantly fewer interactions in the environment and results in an interpretable control policy orders of magnitude smaller than a deep neural network policy.
This research paper delves into the application of Deep Reinforcement Learning (DRL) in asset-class agnostic portfolio optimization, integrating industry-grade methodologies with quantitative finance. At the heart of this integration is our robust framework that not only merges advanced DRL algorithms with modern computational techniques but also emphasizes stringent statistical analysis, software engineering and regulatory compliance. To the best of our knowledge, this is the first study integrating financial Reinforcement Learning with sim-to-real methodologies from robotics and mathematical physics, thus enriching our frameworks and arguments with this unique perspective. Our research culminates with the introduction of AlphaOptimizerNet, a proprietary Reinforcement Learning agent (and corresponding library). Developed from a synthesis of state-of-the-art (SOTA) literature and our unique interdisciplinary methodology, AlphaOptimizerNet demonstrates encouraging risk-return optimization across various asset classes with realistic constraints. These preliminary results underscore the practical efficacy of our frameworks. As the finance sector increasingly gravitates towards advanced algorithmic solutions, our study bridges theoretical advancements with real-world applicability, offering a template for ensuring safety and robust standards in this technologically driven future.
With the continuous expansion of the scale of cloud computing applications, artificial intelligence technologies such as Deep Learning and Reinforcement Learning have gradually become the key tools to solve the automated task scheduling of large-scale cloud computing systems. Aiming at the complexity and real-time requirement of task scheduling in large-scale cloud computing system, this paper proposes an automatic task scheduling scheme based on deep learning and reinforcement learning. Firstly, the deep learning technology is used to monitor and predict the parameters in the cloud computing system in real time to obtain the system status information. Then, combined with reinforcement learning algorithm, the task scheduling strategy is dynamically adjusted according to the real-time system state and task characteristics to achieve the optimal utilization of system resources and the maximum of task execution efficiency. This paper verifies the effectiveness and performance advantages of the proposed scheme in experiments, and proves the potential and application prospect of deep learning and reinforcement learning in automatic task scheduling in large-scale cloud computing systems.
The development of Large Language Models (LLMs) often confronts challenges stemming from the heavy reliance on human annotators in the reinforcement learning with human feedback (RLHF) framework, or the frequent and costly external queries tied to the self-instruct paradigm. In this work, we pivot to Reinforcement Learning (RL) -- but with a twist. Diverging from the typical RLHF, which refines LLMs following instruction data training, we use RL to directly generate the foundational instruction dataset that alone suffices for fine-tuning. Our method, TeaMs-RL, uses a suite of textual operations and rules, prioritizing the diversification of training datasets. It facilitates the generation of high-quality data without excessive reliance on external advanced models, paving the way for a single fine-tuning step and negating the need for subsequent RLHF stages. Our findings highlight key advantages of our approach: reduced need for human involvement and fewer model queries (only $5.73\%$ of the strong baseline's total), along with enhanced capabilities of LLMs in crafting and comprehending complex instructions compared to strong baselines, and substantially improved model privacy protection. Code is available at the link: https://github.com/SafeRL-Lab/TeaMs-RL
Reinforcement Learning from AI Feedback (RLAIF) has the advantages of shorter annotation cycles and lower costs over Reinforcement Learning from Human Feedback (RLHF), making it highly efficient during the rapid strategy iteration periods of large language model (LLM) training. Using ChatGPT as a labeler to provide feedback on open-domain prompts in RLAIF training, we observe an increase in human evaluators' preference win ratio for model responses, but a decrease in evaluators' satisfaction rate. Analysis suggests that the decrease in satisfaction rate is mainly due to some responses becoming less helpful, particularly in terms of correctness and truthfulness, highlighting practical limitations of basic RLAIF. In this paper, we propose Hybrid Reinforcement Learning from AI Feedback (HRLAIF). This method enhances the accuracy of AI annotations for responses, making the model's helpfulness more robust in training process. Additionally, it employs AI for Red Teaming, further improving the model's harmlessness. Human evaluation results show that HRLAIF inherits the ability of RLAIF to enhance human preference for outcomes at a low cost while also improving the satisfaction rate of responses. Compared to the policy model before Reinforcement Learning (RL), it achieves an increase of 2.08\% in satisfaction rate, effectively addressing the issue of a decrease of 4.58\% in satisfaction rate after basic RLAIF.
The Overfitted Brain hypothesis suggests dreams happen to allow generalization in the human brain. Here, we ask if the same is true for reinforcement learning agents as well. Given limited experience in a real environment, we use imagination-based reinforcement learning to train a policy on dream-like episodes, where non-imaginative, predicted trajectories are modified through generative augmentations. Experiments on four ProcGen environments show that, compared to classic imagination and offline training on collected experience, our method can reach a higher level of generalization when dealing with sparsely rewarded environments.
In this paper, we study the cooperative Multi-Agent Reinforcement Learning (MARL) problems using Reward Machines (RMs) to specify the reward functions such that the prior knowledge of high-level events in a task can be leveraged to facilitate the learning efficiency. Unlike the existing work that RMs have been incorporated into MARL for task decomposition and policy learning in relatively simple domains or with an assumption of independencies among the agents, we present Multi-Agent Reinforcement Learning with a Hierarchy of RMs (MAHRM) that is capable of dealing with more complex scenarios when the events among agents can occur concurrently and the agents are highly interdependent. MAHRM exploits the relationship of high-level events to decompose a task into a hierarchy of simpler subtasks that are assigned to a small group of agents, so as to reduce the overall computational complexity. Experimental results in three cooperative MARL domains show that MAHRM outperforms other MARL methods using the same prior knowledge of high-level events.
Reinforcement learning from human feedback (RLHF) is the mainstream paradigm used to align large language models (LLMs) with human preferences. Yet existing RLHF heavily relies on accurate and informative reward models, which are vulnerable and sensitive to noise from various sources, e.g. human labeling errors, making the pipeline fragile. In this work, we improve the effectiveness of the reward model by introducing a penalty term on the reward, named as \textit{contrastive rewards}. %Contrastive rewards Our approach involves two steps: (1) an offline sampling step to obtain responses to prompts that serve as baseline calculation and (2) a contrastive reward calculated using the baseline responses and used in the Proximal Policy Optimization (PPO) step. We show that contrastive rewards enable the LLM to penalize reward uncertainty, improve robustness, encourage improvement over baselines, calibrate according to task difficulty, and reduce variance in PPO. We show empirically contrastive rewards can improve RLHF substantially, evaluated by both GPTs and humans, and our method consistently outperforms strong baselines.
In deep reinforcement learning, estimating the value function to evaluate the quality of states and actions is essential. The value function is often trained using the least squares method, which implicitly assumes a Gaussian error distribution. However, a recent study suggested that the error distribution for training the value function is often skewed because of the properties of the Bellman operator, and violates the implicit assumption of normal error distribution in the least squares method. To address this, we proposed a method called Symmetric Q-learning, in which the synthetic noise generated from a zero-mean distribution is added to the target values to generate a Gaussian error distribution. We evaluated the proposed method on continuous control benchmark tasks in MuJoCo. It improved the sample efficiency of a state-of-the-art reinforcement learning method by reducing the skewness of the error distribution.
Blood Glucose (BG) control involves keeping an individual's BG within a healthy range through extracorporeal insulin injections is an important task for people with type 1 diabetes. However,traditional patient self-management is cumbersome and risky. Recent research has been devoted to exploring individualized and automated BG control approaches, among which Deep Reinforcement Learning (DRL) shows potential as an emerging approach. In this paper, we use an exponential decay model of drug concentration to convert the formalization of the BG control problem, which takes into account the delay and prolongedness of drug effects, from a PAE-POMDP (Prolonged Action Effect-Partially Observable Markov Decision Process) to a MDP, and we propose a novel multi-step DRL-based algorithm to solve the problem. The Prioritized Experience Replay (PER) sampling method is also used in it. Compared to single-step bootstrapped updates, multi-step learning is more efficient and reduces the influence from biasing targets. Our proposed method converges faster and achieves higher cumulative rewards compared to the benchmark in the same training environment, and improves the time-in-range (TIR), the percentage of time the patient's BG is within the target range, in the evaluation phase. Our work validates the effectiveness of multi-step reinforcement learning in BG control, which may help to explore the optimal glycemic control measure and improve the survival of diabetic patients.
The paper presents a technique using reinforcement learning (RL) to adapt the control gains of a quadcopter controller. Specifically, we employed Proximal Policy Optimization (PPO) to train a policy which adapts the gains of a cascaded feedback controller in-flight. The primary goal of this controller is to minimize tracking error while following a specified trajectory. The paper's key objective is to analyze the effectiveness of the adaptive gain policy and compare it to the performance of a static gain control algorithm, where the Integral Squared Error and Integral Time Squared Error are used as metrics. The results show that the adaptive gain scheme achieves over 40$\%$ decrease in tracking error as compared to the static gain controller.
Exploration is a major challenge in reinforcement learning, especially for high-dimensional domains that require function approximation. We propose exploration objectives -- policy optimization objectives that enable downstream maximization of any reward function -- as a conceptual framework to systematize the study of exploration. Within this framework, we introduce a new objective, $L_1$-Coverage, which generalizes previous exploration schemes and supports three fundamental desiderata: 1. Intrinsic complexity control. $L_1$-Coverage is associated with a structural parameter, $L_1$-Coverability, which reflects the intrinsic statistical difficulty of the underlying MDP, subsuming Block and Low-Rank MDPs. 2. Efficient planning. For a known MDP, optimizing $L_1$-Coverage efficiently reduces to standard policy optimization, allowing flexible integration with off-the-shelf methods such as policy gradient and Q-learning approaches. 3. Efficient exploration. $L_1$-Coverage enables the first computationally efficient model-based and model-free algorithms for online (reward-free or reward-driven) reinforcement learning in MDPs with low coverability. Empirically, we find that $L_1$-Coverage effectively drives off-the-shelf policy optimization algorithms to explore the state space.
Intelligent agents must be generalists, capable of quickly adapting to various tasks. In reinforcement learning (RL), model-based RL learns a dynamics model of the world, in principle enabling transfer to arbitrary reward functions through planning. However, autoregressive model rollouts suffer from compounding error, making model-based RL ineffective for long-horizon problems. Successor features offer an alternative by modeling a policy's long-term state occupancy, reducing policy evaluation under new rewards to linear regression. Yet, zero-shot policy optimization for new tasks with successor features can be challenging. This work proposes a novel class of models, i.e., Distributional Successor Features for Zero-Shot Policy Optimization (DiSPOs), that learn a distribution of successor features of a stationary dataset's behavior policy, along with a policy that acts to realize different successor features achievable within the dataset. By directly modeling long-term outcomes in the dataset, DiSPOs avoid compounding error while enabling a simple scheme for zero-shot policy optimization across reward functions. We present a practical instantiation of DiSPOs using diffusion models and show their efficacy as a new class of transferable models, both theoretically and empirically across various simulated robotics problems. Videos and code available at https://weirdlabuw.github.io/dispo/.
We study paycheck optimization, which examines how to allocate income in order to achieve several competing financial goals. For paycheck optimization, a quantitative methodology is missing, due to a lack of a suitable problem formulation. To deal with this issue, we formulate the problem as a utility maximization problem. The proposed formulation is able to (i) unify different financial goals; (ii) incorporate user preferences regarding the goals; (iii) handle stochastic interest rates. The proposed formulation also facilitates an end-to-end reinforcement learning solution, which is implemented on a variety of problem settings.
Feature selection plays a crucial role in improving predictive accuracy by identifying relevant features while filtering out irrelevant ones. This study investigates the importance of effective feature selection in enhancing the performance of classification models. By employing reinforcement learning (RL) algorithms, specifically Q-learning (QL) and SARSA learning, this paper addresses the feature selection challenge. Using the Breast Cancer Coimbra dataset (BCCDS) and three normalization methods (Min-Max, l1, and l2), the study evaluates the performance of these algorithms. Results show that QL@Min-Max and SARSA@l2 achieve the highest classification accuracies, reaching 87% and 88%, respectively. This highlights the effectiveness of RL-based feature selection methods in optimizing classification tasks, contributing to improved model accuracy and efficiency.
Autonomous spacecraft control via Shielded Deep Reinforcement Learning (SDRL) has become a rapidly growing research area. However, the construction of shields and the definition of tasking remains informal, resulting in policies with no guarantees on safety and ambiguous goals for the RL agent. In this paper, we first explore the use of formal languages, namely Linear Temporal Logic (LTL), to formalize spacecraft tasks and safety requirements. We then define a manner in which to construct a reward function from a co-safe LTL specification automatically for effective training in SDRL framework. We also investigate methods for constructing a shield from a safe LTL specification for spacecraft applications and propose three designs that provide probabilistic guarantees. We show how these shields interact with different policies and the flexibility of the reward structure through several experiments.