In the field of high-performance computing (HPC), there has been recent exploration into the use of deep reinforcement learning for cluster scheduling (DRL scheduling), which has demonstrated promising outcomes. However, a significant challenge arises from the lack of interpretability in deep neural networks (DNN), rendering them as black-box models to system managers. This lack of model interpretability hinders the practical deployment of DRL scheduling. In this work, we present a framework called IRL (Interpretable Reinforcement Learning) to address the issue of interpretability of DRL scheduling. The core idea is to interpret DNN (i.e., the DRL policy) as a decision tree by utilizing imitation learning. Unlike DNN, decision tree models are non-parametric and easily comprehensible to humans. To extract an effective and efficient decision tree, IRL incorporates the Dataset Aggregation (DAgger) algorithm and introduces the notion of critical state to prune the derived decision tree. Through trace-based experiments, we demonstrate that IRL is capable of converting a black-box DNN policy into an interpretable rulebased decision tree while maintaining comparable scheduling performance. Additionally, IRL can contribute to the setting of rewards in DRL scheduling.
Navigating human-filled spaces is crucial for the interactive social robots to support advanced services, such as cooperative carrying, which enables service provision in complex and crowded environments while adapting behavior based on real-time human language commands or feedback. However, existing social robot navigation planners face two major challenges: managing real-time user inputs and ensuring socially compliant behaviors in unfamiliar, zero-shot environments. In response, we introduce SALM, an interactive, human-in-loop Socially-Aware navigation Large Language Model framework that dynamically integrates deep reinforcement learning (DRL) with large language model (LLM) capabilities. SALM leverages contextual semantic understanding from real-time human-robot interactions to convert high-level user commands into precise, low-level control actions. A high-level LLM module parses user input, guiding the simultaneous generation of navigation commands by both a large language navigation model (LNM) and a DRL-based navigation model (RLNM). A memory mechanism archives temporal data for continuous refinement, while a multi-step graph-of-thoughts inference-based large language feedback model adaptively fuses the strengths of both planning approaches. Experimental evaluations demonstrate that SALM not only enhances navigational precision in crowded, dynamic environments but also significantly improves system adaptability, offering tailored behaviors that align with individual user preferences and real-time feedback. More details and videos about this work are available at: https://sites.google.com/view/navi-salm.
Portfolio optimization involves determining the optimal allocation of portfolio assets in order to maximize a given investment objective. Traditionally, some form of mean-variance optimization is used with the aim of maximizing returns while minimizing risk, however, more recently, deep reinforcement learning formulations have been explored. Increasingly, investors have demonstrated an interest in incorporating ESG objectives when making investment decisions, and modifications to the classical mean-variance optimization framework have been developed. In this work, we study the use of deep reinforcement learning for responsible portfolio optimization, by incorporating ESG states and objectives, and provide comparisons against modified mean-variance approaches. Our results show that deep reinforcement learning policies can provide competitive performance against mean-variance approaches for responsible portfolio allocation across additive and multiplicative utility functions of financial and ESG responsibility objectives.
Driven by the great advances in metaverse and edge computing technologies, vehicular edge metaverses are expected to disrupt the current paradigm of intelligent transportation systems. As highly computerized avatars of Vehicular Metaverse Users (VMUs), the Vehicle Twins (VTs) deployed in edge servers can provide valuable metaverse services to improve driving safety and on-board satisfaction for their VMUs throughout journeys. To maintain uninterrupted metaverse experiences, VTs must be migrated among edge servers following the movements of vehicles. This can raise concerns about privacy breaches during the dynamic communications among vehicular edge metaverses. To address these concerns and safeguard location privacy, pseudonyms as temporary identifiers can be leveraged by both VMUs and VTs to realize anonymous communications in the physical space and virtual spaces. However, existing pseudonym management methods fall short in meeting the extensive pseudonym demands in vehicular edge metaverses, thus dramatically diminishing the performance of privacy preservation. To this end, we present a cross-metaverse empowered dual pseudonym management framework. We utilize cross-chain technology to enhance management efficiency and data security for pseudonyms. Furthermore, we propose a metric to assess the privacy level and employ a Multi-Agent Deep Reinforcement Learning (MADRL) approach to obtain an optimal pseudonym generating strategy. Numerical results demonstrate that our proposed schemes are high-efficiency and cost-effective, showcasing their promising applications in vehicular edge metaverses.
Optimal control of parametric partial differential equations (PDEs) is crucial in many applications in engineering and science. In recent years, the progress in scientific machine learning has opened up new frontiers for the control of parametric PDEs. In particular, deep reinforcement learning (DRL) has the potential to solve high-dimensional and complex control problems in a large variety of applications. Most DRL methods rely on deep neural network (DNN) control policies. However, for many dynamical systems, DNN-based control policies tend to be over-parametrized, which means they need large amounts of training data, show limited robustness, and lack interpretability. In this work, we leverage dictionary learning and differentiable L$_0$ regularization to learn sparse, robust, and interpretable control policies for parametric PDEs. Our sparse policy architecture is agnostic to the DRL method and can be used in different policy-gradient and actor-critic DRL algorithms without changing their policy-optimization procedure. We test our approach on the challenging tasks of controlling parametric Kuramoto-Sivashinsky and convection-diffusion-reaction PDEs. We show that our method (1) outperforms baseline DNN-based DRL policies, (2) allows for the derivation of interpretable equations of the learned optimal control laws, and (3) generalizes to unseen parameters of the PDE without retraining the policies.
Even though Deep Reinforcement Learning (DRL) showed outstanding results in the fields of Robotics and Games, it is still challenging to implement it in the optimization of industrial processes like wastewater treatment. One of the challenges is the lack of a simulation environment that will represent the actual plant as accurately as possible to train DRL policies. Stochasticity and non-linearity of wastewater treatment data lead to unstable and incorrect predictions of models over long time horizons. One possible reason for the models' incorrect simulation behavior can be related to the issue of compounding error, which is the accumulation of errors throughout the simulation. The compounding error occurs because the model utilizes its predictions as inputs at each time step. The error between the actual data and the prediction accumulates as the simulation continues. We implemented two methods to improve the trained models for wastewater treatment data, which resulted in more accurate simulators: 1- Using the model's prediction data as input in the training step as a tool of correction, and 2- Change in the loss function to consider the long-term predicted shape (dynamics). The experimental results showed that implementing these methods can improve the behavior of simulators in terms of Dynamic Time Warping throughout a year up to 98% compared to the base model. These improvements demonstrate significant promise in creating simulators for biological processes that do not need pre-existing knowledge of the process but instead depend exclusively on time series data obtained from the system.
In the context of online education, designing an automatic solver for geometric problems has been considered a crucial step towards general math Artificial Intelligence (AI), empowered by natural language understanding and traditional logical inference. In most instances, problems are addressed by adding auxiliary components such as lines or points. However, adding auxiliary components automatically is challenging due to the complexity in selecting suitable auxiliary components especially when pivotal decisions have to be made. The state-of-the-art performance has been achieved by exhausting all possible strategies from the category library to identify the one with the maximum likelihood. However, an extensive strategy search have to be applied to trade accuracy for ef-ficiency. To add auxiliary components automatically and efficiently, we present deep reinforcement learning framework based on the language model, such as BERT. We firstly apply the graph attention mechanism to reduce the strategy searching space, called AttnStrategy, which only focus on the conclusion-related components. Meanwhile, a novel algorithm, named Automatically Adding Auxiliary Components using Reinforcement Learning framework (A3C-RL), is proposed by forcing an agent to select top strategies, which incorporates the AttnStrategy and BERT as the memory components. Results from extensive experiments show that the proposed A3C-RL algorithm can substantially enhance the average precision by 32.7% compared to the traditional MCTS. In addition, the A3C-RL algorithm outperforms humans on the geometric questions from the annual University Entrance Mathematical Examination of China.
Deep reinforcement learning has made significant progress in games with imperfect information, but its performance in the card game Doudizhu (Chinese Poker/Fight the Landlord) remains unsatisfactory. Doudizhu is different from conventional games as it involves three players and combines elements of cooperation and confrontation, resulting in a large state and action space. In 2021, a Doudizhu program called DouZero\cite{zha2021douzero} surpassed previous models without prior knowledge by utilizing traditional Monte Carlo methods and multilayer perceptrons. Building on this work, our study incorporates residual networks into the model, explores different architectural designs, and conducts multi-role testing. Our findings demonstrate that this model significantly improves the winning rate within the same training time. Additionally, we introduce a call scoring system to assist the agent in deciding whether to become a landlord. With these enhancements, our model consistently outperforms the existing version of DouZero and even experienced human players. \footnote{The source code is available at \url{https://github.com/Yingchaol/Douzero_Resnet.git.}
Reinforcement learning (RL) tackles sequential decision-making problems by creating agents that interacts with their environment. However, existing algorithms often view these problem as static, focusing on point estimates for model parameters to maximize expected rewards, neglecting the stochastic dynamics of agent-environment interactions and the critical role of uncertainty quantification. Our research leverages the Kalman filtering paradigm to introduce a novel and scalable sampling algorithm called Langevinized Kalman Temporal-Difference (LKTD) for deep reinforcement learning. This algorithm, grounded in Stochastic Gradient Markov Chain Monte Carlo (SGMCMC), efficiently draws samples from the posterior distribution of deep neural network parameters. Under mild conditions, we prove that the posterior samples generated by the LKTD algorithm converge to a stationary distribution. This convergence not only enables us to quantify uncertainties associated with the value function and model parameters but also allows us to monitor these uncertainties during policy updates throughout the training phase. The LKTD algorithm paves the way for more robust and adaptable reinforcement learning approaches.
Designing a humanoid locomotion controller is challenging and classically split up in sub-problems. Footstep planning is one of those, where the sequence of footsteps is defined. Even in simpler environments, finding a minimal sequence, or even a feasible sequence, yields a complex optimization problem. In the literature, this problem is usually addressed by search-based algorithms (e.g. variants of A*). However, such approaches are either computationally expensive or rely on hand-crafted tuning of several parameters. In this work, at first, we propose an efficient footstep planning method to navigate in local environments with obstacles, based on state-of-the art Deep Reinforcement Learning (DRL) techniques, with very low computational requirements for on-line inference. Our approach is heuristic-free and relies on a continuous set of actions to generate feasible footsteps. In contrast, other methods necessitate the selection of a relevant discrete set of actions. Second, we propose a forecasting method, allowing to quickly estimate the number of footsteps required to reach different candidates of local targets. This approach relies on inherent computations made by the actor-critic DRL architecture. We demonstrate the validity of our approach with simulation results, and by a deployment on a kid-size humanoid robot during the RoboCup 2023 competition.
Deep learning methods have been widely used in robotic applications, making learning-enabled control design for complex nonlinear systems a promising direction. Although deep reinforcement learning methods have demonstrated impressive empirical performance, they lack the stability guarantees that are important in safety-critical situations. One way to provide these guarantees is to learn Lyapunov certificates alongside control policies. There are three related problems: 1) verify that a given Lyapunov function candidate satisfies the conditions for a given controller on a region, 2) find a valid Lyapunov function and controller on a given region, and 3) find a valid Lyapunov function and a controller such that the region of attraction is as large as possible. Previous work has shown that if the dynamics are piecewise linear, it is possible to solve problems 1) and 2) by solving a Mixed-Integer Linear Program (MILP). In this work, we build upon this method by proposing a Lyapunov neural network that considers monotonicity over half spaces in different directions. We 1) propose a specific choice of Lyapunov function architecture that ensures non-negativity and a unique global minimum by construction, and 2) show that this can be leveraged to find the controller and Lyapunov certificates faster and with a larger valid region by maximizing the size of a square inscribed in a given level set. We apply our method to a 2D inverted pendulum, unicycle path following, a 3-D feedback system, and a 4-D cart pole system, and demonstrate it can shorten the training time by half compared to the baseline, as well as find a larger ROA.
We present a novel multi-parent crossover operator in genetic algorithms (GAs) called ``Deep Neural Crossover'' (DNC). Unlike conventional GA crossover operators that rely on a random selection of parental genes, DNC leverages the capabilities of deep reinforcement learning (DRL) and an encoder-decoder architecture to select the genes. Specifically, we use DRL to learn a policy for selecting promising genes. The policy is stochastic, to maintain the stochastic nature of GAs, representing a distribution for selecting genes with a higher probability of improving fitness. Our architecture features a recurrent neural network (RNN) to encode the parental genomes into latent memory states, and a decoder RNN that utilizes an attention-based pointing mechanism to generate a distribution over the next selected gene in the offspring. To improve the training time, we present a pre-training approach, wherein the architecture is initially trained on a single problem within a specific domain and then applied to solving other problems of the same domain. We compare DNC to known operators from the literature over two benchmark domains -- bin packing and graph coloring. We compare with both two- and three-parent crossover, outperforming all baselines. DNC is domain-independent and can be easily applied to other problem domains.
Recently there has been a growing interest in industry and academia, regarding the use of wireless chargers to prolong the operational longevity of unmanned aerial vehicles (commonly knowns as drones). In this paper we consider a charger-assisted drone application: a drone is deployed to observe a set points of interest, while a charger can move to recharge the drone's battery. We focus on the route and charging schedule of the drone and the mobile charger, to obtain high observation utility with the shortest possible time, while ensuring the drone remains operational during task execution. Essentially, this proposed drone-charger scheduling problem is a multi-stage decision-making process, in which the drone and the mobile charger act as two agents who cooperate to finish a task. The discrete-continuous hybrid action space of the two agents poses a significant challenge in our problem. To address this issue, we present a hybrid-action deep reinforcement learning framework, called HaDMC, which uses a standard policy learning algorithm to generate latent continuous actions. Motivated by representation learning, we specifically design and train an action decoder. It involves two pipelines to convert the latent continuous actions into original discrete and continuous actions, by which the drone and the charger can directly interact with environment. We embed a mutual learning scheme in model training, emphasizing the collaborative rather than individual actions. We conduct extensive numerical experiments to evaluate HaDMC and compare it with state-of-the-art deep reinforcement learning approaches. The experimental results show the effectiveness and efficiency of our solution.
With the advent of universal function approximators in the domain of reinforcement learning, the number of practical applications leveraging deep reinforcement learning (DRL) has exploded. Decision-making in autonomous vehicles (AVs) has emerged as a chief application among them, taking the sensor data or the higher-order kinematic variables as the input and providing a discrete choice or continuous control output. There has been a continuous effort to understand the black-box nature of the DRL models, but so far, there hasn't been any discussion (to the best of authors' knowledge) about how the models learn the physical process. This presents an overwhelming limitation that restricts the real-world deployment of DRL in AVs. Therefore, in this research work, we try to decode the knowledge learnt by the attention-based DRL framework about the physical process. We use a continuous proximal policy optimization-based DRL algorithm as the baseline model and add a multi-head attention framework in an open-source AV simulation environment. We provide some analytical techniques for discussing the interpretability of the trained models in terms of explainability and causality for spatial and temporal correlations. We show that the weights in the first head encode the positions of the neighboring vehicles while the second head focuses on the leader vehicle exclusively. Also, the ego vehicle's action is causally dependent on the vehicles in the target lane spatially and temporally. Through these findings, we reliably show that these techniques can help practitioners decipher the results of the DRL algorithms.
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.
Mobile robots are being used on a large scale in various crowded situations and become part of our society. The socially acceptable navigation behavior of a mobile robot with individual human consideration is an essential requirement for scalable applications and human acceptance. Deep Reinforcement Learning (DRL) approaches are recently used to learn a robot's navigation policy and to model the complex interactions between robots and humans. We propose to divide existing DRL-based navigation approaches based on the robot's exhibited social behavior and distinguish between social collision avoidance with a lack of social behavior and socially aware approaches with explicit predefined social behavior. In addition, we propose a novel socially integrated navigation approach where the robot's social behavior is adaptive and emerges from the interaction with humans. The formulation of our approach is derived from a sociological definition, which states that social acting is oriented toward the acting of others. The DRL policy is trained in an environment where other agents interact socially integrated and reward the robot's behavior individually. The simulation results indicate that the proposed socially integrated navigation approach outperforms a socially aware approach in terms of ego navigation performance while significantly reducing the negative impact on all agents within the environment.
In software-defined networking (SDN), the implementation of distributed SDN controllers, with each controller responsible for managing a specific sub-network or domain, plays a critical role in achieving a balance between centralized control, scalability, reliability, and network efficiency. These controllers must be synchronized to maintain a logically centralized view of the entire network. While there are various approaches for synchronizing distributed SDN controllers, most tend to prioritize goals such as optimization of communication latency or load balancing, often neglecting to address both the aspects simultaneously. This limitation becomes particularly significant when considering applications like Augmented and Virtual Reality (AR/VR), which demand constrained network latencies and substantial computational resources. Additionally, many existing studies in this field predominantly rely on value-based reinforcement learning (RL) methods, overlooking the potential advantages offered by state-of-the-art policy-based RL algorithms. To bridge this gap, our work focuses on examining deep reinforcement learning (DRL) techniques, encompassing both value-based and policy-based methods, to guarantee an upper latency threshold for AR/VR task offloading within SDN environments, while selecting the most cost-effective servers for AR/VR task offloading. Our evaluation results indicate that while value-based methods excel in optimizing individual network metrics such as latency or load balancing, policy-based approaches exhibit greater robustness in adapting to sudden network changes or reconfiguration.
Recent advancements in the flexible job-shop scheduling problem (FJSSP) are primarily based on deep reinforcement learning (DRL) due to its ability to generate high-quality, real-time solutions. However, DRL approaches often fail to fully harness the strengths of existing techniques such as exact methods or constraint programming (CP), which can excel at finding optimal or near-optimal solutions for smaller instances. This paper aims to integrate CP within a deep learning (DL) based methodology, leveraging the benefits of both. In this paper, we introduce a method that involves training a DL model using optimal solutions generated by CP, ensuring the model learns from high-quality data, thereby eliminating the need for the extensive exploration typical in DRL and enhancing overall performance. Further, we integrate CP into our DL framework to jointly construct solutions, utilizing DL for the initial complex stages and transitioning to CP for optimal resolution as the problem is simplified. Our hybrid approach has been extensively tested on three public FJSSP benchmarks, demonstrating superior performance over five state-of-the-art DRL approaches and a widely-used CP solver. Additionally, with the objective of exploring the application to other combinatorial optimization problems, promising preliminary results are presented on applying our hybrid approach to the traveling salesman problem, combining an exact method with a well-known DRL method.
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.
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.