-- Data centers are increasingly using more energy due to the rise in Artificial Intelligence (AI) workloads, which negatively impacts the environment and raises operational costs. Reducing operating expenses and carbon emissions while maintaining performance in data centers is a challenging problem. This work introduces a unique approach combining Game Theory (GT) and Deep Reinforcement Learning (DRL) for optimizing the distribution of AI inference workloads in geo-distributed data centers to reduce carbon emissions and cloud operating (energy + data transfer) costs. The proposed technique integrates the principles of non-cooperative Game Theory into a DRL framework, enabling data centers to make intelligent decisions regarding workload allocation while considering the heterogeneity of hardware resources, the dynamic nature of electricity prices, inter-data center data transfer costs, and carbon footprints. We conducted extensive experiments comparing our game-theoretic DRL (GT-DRL) approach with current DRL-based and other optimization techniques. The results demonstrate that our strategy outperforms the state-of-the-art in reducing carbon emissions and minimizing cloud operating costs without compromising computational performance. This work has significant implications for achieving sustainability and cost-efficiency in data centers handling AI inference workloads across diverse geographic locations. The use of data centers to support Internet applications There are several reasons to favor geographically distributed is growing rapidly, with experts predicting increased data centers as they provide advantages in load balancing, reliability reliance on data centers fueled by Artificial Intelligence and redundancy, latency reduction, and compliance (AI)-driven workloads, new cloud services, demand for with data sovereignty rules [11]. Another compelling reason edge applications, expansion of the Internet of Things (IoT) for geographically spreading data centers is to reduce electricity devices, and the proliferation of technologies such as 5G/6G. AI workloads require immense [12], electricity prices fluctuate depending on the time of day. They also necessitate vast amounts of data for costs rise; whereas when the demand is low, they decline [13]. As a result, AI-driven computing cost on non-residential or commercial customers depending is fueling demand for more power-hungry data centers with on the maximum (peak) power utilized at any given moment expanded storage capabilities [2].

Reinforcement learning for control over continuous spaces typically uses high-entropy stochastic policies, such as Gaussian distributions, for local exploration and estimating policy gradient to optimize performance. Many robotic control problems deal with complex unstable dynamics, where applying actions that are off the feasible control manifolds can quickly lead to undesirable divergence. In such cases, most samples taken from the ambient action space generate low-value trajectories that hardly contribute to policy improvement, resulting in slow or failed learning. We propose to improve action selection in this model-free RL setting by introducing additional adaptive control steps based on Extremum-Seeking Control (ESC). On each action sampled from stochastic policies, we apply sinusoidal perturbations and query for estimated Q-values as the response signal. Based on ESC, we then dynamically improve the sampled actions to be closer to nearby optima before applying them to the environment. Our methods can be easily added in standard policy optimization to improve learning efficiency, which we demonstrate in various control learning environments.

Manipulating objects is a hallmark of human intelligence, and an important task in domains such as robotics. In principle, Reinforcement Learning (RL) offers a general approach to learn object manipulation. In practice, however, domains with more than a few objects are difficult for RL agents due to the curse of dimensionality, especially when learning from raw image observations. In this work we propose a structured approach for visual RL that is suitable for representing multiple objects and their interaction, and use it to learn goal-conditioned manipulation of several objects. Key to our method is the ability to handle goals with dependencies between the objects (e.g., moving objects in a certain order). We further relate our architecture to the generalization capability of the trained agent, based on a theoretical result for compositional generalization, and demonstrate agents that learn with 3 objects but generalize to similar tasks with over 10 objects. Videos and code are available on the project website: https://sites.google.com/view/entity-centric-rl

Modern control systems are increasingly turning to machine learning algorithms to augment their performance and adaptability. Within this context, Deep Reinforcement Learning (DRL) has emerged as a promising control framework, particularly in the domain of marine transportation. Its potential for autonomous marine applications lies in its ability to seamlessly combine path-following and collision avoidance with an arbitrary number of obstacles. However, current DRL algorithms require disproportionally large computational resources to find near-optimal policies compared to the posed control problem when the searchable parameter space becomes large. To combat this, our work delves into the application of Variational AutoEncoders (VAEs) to acquire a generalized, low-dimensional latent encoding of a high-fidelity range-finding sensor, which serves as the exteroceptive input to a DRL agent. The agent's performance, encompassing path-following and collision avoidance, is systematically tested and evaluated within a stochastic simulation environment, presenting a comprehensive exploration of our proposed approach in maritime control systems.

Multiplication is a fundamental operation in many applications, and multipliers are widely adopted in various circuits. However, optimizing multipliers is challenging and non-trivial due to the huge design space. In this paper, we propose RL-MUL, a multiplier design optimization framework based on reinforcement learning. Specifically, we utilize matrix and tensor representations for the compressor tree of a multiplier, based on which the convolutional neural networks can be seamlessly incorporated as the agent network. The agent can learn to optimize the multiplier structure based on a Pareto-driven reward which is customized to accommodate the trade-off between area and delay. Additionally, the capability of RL-MUL is extended to optimize the fused multiply-accumulator (MAC) designs. Experiments are conducted on different bit widths of multipliers. The results demonstrate that the multipliers produced by RL-MUL can dominate all baseline designs in terms of area and delay. The performance gain of RL-MUL is further validated by comparing the area and delay of processing element arrays using multipliers from RL-MUL and baseline approaches.

Mine fleet management algorithms can significantly reduce operational costs and enhance productivity in mining systems. Most current fleet management algorithms are evaluated based on self-implemented or proprietary simulation environments, posing challenges for replication and comparison. This paper models the simulation environment for mine fleet management from a complex systems perspective. Building upon previous work, we introduce probabilistic, user-defined events for random event simulation and implement various evaluation metrics and baselines, effectively reflecting the robustness of fleet management algorithms against unforeseen incidents. We present ``OpenMines'', an open-source framework encompassing the entire process of mine system modeling, algorithm development, and evaluation, facilitating future algorithm comparison and replication in the field. Code is available in https://github.com/370025263/openmines.

Traffic signal control has a great impact on alleviating traffic congestion in modern cities. Deep reinforcement learning (RL) has been widely used for this task in recent years, demonstrating promising performance but also facing many challenges such as limited performances and sample inefficiency. To handle these challenges, MTLight is proposed to enhance the agent observation with a latent state, which is learned from numerous traffic indicators. Meanwhile, multiple auxiliary and supervisory tasks are constructed to learn the latent state, and two types of embedding latent features, the task-specific feature and task-shared feature, are used to make the latent state more abundant. Extensive experiments conducted on CityFlow demonstrate that MTLight has leading convergence speed and asymptotic performance. We further simulate under peak-hour pattern in all scenarios with increasing control difficulty and the results indicate that MTLight is highly adaptable.

In data-driven decision-making in marketing, healthcare, and education, it is desirable to utilize a large amount of data from existing ventures to navigate high-dimensional feature spaces and address data scarcity in new ventures. We explore knowledge transfer in dynamic decision-making by concentrating on batch stationary environments and formally defining task discrepancies through the lens of Markov decision processes (MDPs). We propose a framework of Transferred Fitted $Q$-Iteration algorithm with general function approximation, enabling the direct estimation of the optimal action-state function $Q^*$ using both target and source data. We establish the relationship between statistical performance and MDP task discrepancy under sieve approximation, shedding light on the impact of source and target sample sizes and task discrepancy on the effectiveness of knowledge transfer. We show that the final learning error of the $Q^*$ function is significantly improved from the single task rate both theoretically and empirically.

Reinforcement Learning (RL) has shown remarkable success in solving relatively complex tasks, yet the deployment of RL systems in real-world scenarios poses significant challenges related to safety and robustness. This paper aims to identify and further understand those challenges thorough the exploration of the main dimensions of the safe and robust RL landscape, encompassing algorithmic, ethical, and practical considerations. We conduct a comprehensive review of methodologies and open problems that summarizes the efforts in recent years to address the inherent risks associated with RL applications. After discussing and proposing definitions for both safe and robust RL, the paper categorizes existing research works into different algorithmic approaches that enhance the safety and robustness of RL agents. We examine techniques such as uncertainty estimation, optimisation methodologies, exploration-exploitation trade-offs, and adversarial training. Environmental factors, including sim-to-real transfer and domain adaptation, are also scrutinized to understand how RL systems can adapt to diverse and dynamic surroundings. Moreover, human involvement is an integral ingredient of the analysis, acknowledging the broad set of roles that humans can take in this context. Importantly, to aid practitioners in navigating the complexities of safe and robust RL implementation, this paper introduces a practical checklist derived from the synthesized literature. The checklist encompasses critical aspects of algorithm design, training environment considerations, and ethical guidelines. It will serve as a resource for developers and policymakers alike to ensure the responsible deployment of RL systems in many application domains.

With extensive pre-trained knowledge and high-level general capabilities, large language models (LLMs) emerge as a promising avenue to augment reinforcement learning (RL) in aspects such as multi-task learning, sample efficiency, and task planning. In this survey, we provide a comprehensive review of the existing literature in $\textit{LLM-enhanced RL}$ and summarize its characteristics compared to conventional RL methods, aiming to clarify the research scope and directions for future studies. Utilizing the classical agent-environment interaction paradigm, we propose a structured taxonomy to systematically categorize LLMs' functionalities in RL, including four roles: information processor, reward designer, decision-maker, and generator. Additionally, for each role, we summarize the methodologies, analyze the specific RL challenges that are mitigated, and provide insights into future directions. Lastly, potential applications, prospective opportunities and challenges of the $\textit{LLM-enhanced RL}$ are discussed.