The divergence of the Q-value estimation has been a prominent issue in offline RL, where the agent has no access to real dynamics. Traditional beliefs attribute this instability to querying out-of-distribution actions when bootstrapping value targets. Though this issue can be alleviated with policy constraints or conservative Q estimation, a theoretical understanding of the underlying mechanism causing the divergence has been absent. In this work, we aim to thoroughly comprehend this mechanism and attain an improved solution. We first identify a fundamental pattern, self-excitation, as the primary cause of Q-value estimation divergence in offline RL. Then, we propose a novel Self-Excite Eigenvalue Measure (SEEM) metric based on Neural Tangent Kernel (NTK) to measure the evolving property of Q-network at training, which provides an intriguing explanation of the emergence of divergence. For the first time, our theory can reliably decide whether the training will diverge at an early stage, and even predict the order of the growth for the estimated Q-value, the model's norm, and the crashing step when an SGD optimizer is used. The experiments demonstrate perfect alignment with this theoretic analysis. Building on our insights, we propose to resolve divergence from a novel perspective, namely improving the model's architecture for better extrapolating behavior. Through extensive empirical studies, we identify LayerNorm as a good solution to effectively avoid divergence without introducing detrimental bias, leading to superior performance. Experimental results prove that it can still work in some most challenging settings, i.e. using only 1 transitions of the dataset, where all previous methods fail. Moreover, it can be easily plugged into modern offline RL methods and achieve SOTA results on many challenging tasks. We also give unique insights into its effectiveness.

With the increasing computing power, using data-driven approaches to co-design a robot's morphology and controller has become a promising way. However, most existing data-driven methods require training the controller for each morphology to calculate fitness, which is time-consuming. In contrast, the dual-network framework utilizes data collected by individual networks under a specific morphology to train a population network that provides a surrogate function for morphology optimization. This approach replaces the traditional evaluation of a diverse set of candidates, thereby speeding up the training. Despite considerable results, the online training of both networks impedes their performance. To address this issue, we propose a concurrent network framework that combines online and offline reinforcement learning (RL) methods. By leveraging the behavior cloning term in a flexible manner, we achieve an effective combination of both networks. We conducted multiple sets of comparative experiments in the simulator and found that the proposed method effectively addresses issues present in the dual-network framework, leading to overall algorithmic performance improvement. Furthermore, we validated the algorithm on a real robot, demonstrating its feasibility in a practical application.

Exploration in high-dimensional, continuous spaces with sparse rewards is an open problem in reinforcement learning. Artificial curiosity algorithms address this by creating rewards that lead to exploration. Given a reinforcement learning algorithm capable of maximizing rewards, the problem reduces to finding an optimization objective consistent with exploration. Maximum entropy exploration uses the entropy of the state visitation distribution as such an objective. However, efficiently estimating the entropy of the state visitation distribution is challenging in high-dimensional, continuous spaces. We introduce an artificial curiosity algorithm based on lower bounding an approximation to the entropy of the state visitation distribution. The bound relies on a result we prove for non-parametric density estimation in arbitrary dimensions using k-means. We show that our approach is both computationally efficient and competitive on benchmarks for exploration in high-dimensional, continuous spaces, especially on tasks where reinforcement learning algorithms are unable to find rewards.

Motivation: Disease diagnosis oriented dialogue system models the interactive consultation procedure as Markov Decision Process and reinforcement learning algorithms are used to solve the problem. Existing approaches usually employ a flat policy structure that treat all symptoms and diseases equally for action making. This strategy works well in the simple scenario when the action space is small, however, its efficiency will be challenged in the real environment. Inspired by the offline consultation process, we propose to integrate a hierarchical policy structure of two levels into the dialogue systemfor policy learning. The high-level policy consists of amastermodel that is responsible for triggering a low-levelmodel, the lowlevel policy consists of several symptom checkers and a disease classifier. The proposed policy structure is capable to deal with diagnosis problem including large number of diseases and symptoms. Results: Experimental results on three real-world datasets and a synthetic dataset demonstrate that our hierarchical framework achieves higher accuracy and symptom recall in disease diagnosis compared with existing systems. We construct a benchmark including datasets and implementation of existing algorithms to encourage follow-up researches. Availability: The code and data is available from https://github.com/FudanDISC/DISCOpen-MedBox-DialoDiagnosis Contact: 21210980124@m.fudan.edu.cn Supplementary information: Supplementary data are available at Bioinformatics online.

Reinforcement learning has been successful across several applications in which agents have to learn to act in environments with sparse feedback. However, despite this empirical success there is still a lack of theoretical understanding of how the parameters of reinforcement learning models and the features used to represent states interact to control the dynamics of learning. In this work, we use concepts from statistical physics, to study the typical case learning curves for temporal difference learning of a value function with linear function approximators. Our theory is derived under a Gaussian equivalence hypothesis where averages over the random trajectories are replaced with temporally correlated Gaussian feature averages and we validate our assumptions on small scale Markov Decision Processes. We find that the stochastic semi-gradient noise due to subsampling the space of possible episodes leads to significant plateaus in the value error, unlike in traditional gradient descent dynamics. We study how learning dynamics and plateaus depend on feature structure, learning rate, discount factor, and reward function. We then analyze how strategies like learning rate annealing and reward shaping can favorably alter learning dynamics and plateaus. To conclude, our work introduces new tools to open a new direction towards developing a theory of learning dynamics in reinforcement learning.

Robotics has dramatically increased our ability to gather data about our environments, creating an opportunity for the robotics and algorithms communities to collaborate on novel solutions to environmental monitoring problems. To understand a taxonomy of problems and methods in this realm, we present the first comprehensive survey of decision-theoretic approaches that enable efficient sampling of various environmental processes. We investigate representations for different environments, followed by a discussion of using these presentations to solve tasks of interest, such as learning, localization, and monitoring. To efficiently implement the tasks, decision-theoretic optimization algorithms consider: (1) where to take measurements from, (2) which tasks to be assigned, (3) what samples to collect, (4) when to collect samples, (5) how to learn environment; and (6) who to communicate. Finally, we summarize our study and present the challenges and opportunities in robotic environmental monitoring.

We focus on developing efficient and reliable policy optimization strategies for robot learning with real-world data. In recent years, policy gradient methods have emerged as a promising paradigm for training control policies in simulation. However, these approaches often remain too data inefficient or unreliable to train on real robotic hardware. In this paper we introduce a novel policy gradient-based policy optimization framework which systematically leverages a (possibly highly simplified) first-principles model and enables learning precise control policies with limited amounts of real-world data. Our approach $1)$ uses the derivatives of the model to produce sample-efficient estimates of the policy gradient and $2)$ uses the model to design a low-level tracking controller, which is embedded in the policy class. Theoretical analysis provides insight into how the presence of this feedback controller overcomes key limitations of stand-alone policy gradient methods, while hardware experiments with a small car and quadruped demonstrate that our approach can learn precise control strategies reliably and with only minutes of real-world data.

This work studies a central extremal graph theory problem inspired by a 1975 conjecture of Erd\H{o}s, which aims to find graphs with a given size (number of nodes) that maximize the number of edges without having 3- or 4-cycles. We formulate this problem as a sequential decision-making problem and compare AlphaZero, a neural network-guided tree search, with tabu search, a heuristic local search method. Using either method, by introducing a curriculum -- jump-starting the search for larger graphs using good graphs found at smaller sizes -- we improve the state-of-the-art lower bounds for several sizes. We also propose a flexible graph-generation environment and a permutation-invariant network architecture for learning to search in the space of graphs.

The requirement of wireless data demands is increasingly high as the sixth-generation (6G) technology evolves. Reconfigurable intelligent surface (RIS) is promisingly deemed to be one of 6G techniques for extending service coverage, reducing power consumption, and enhancing spectral efficiency. In this article, we have provided some fundamentals of RIS deployment in theory and hardware perspectives as well as utilization of artificial intelligence (AI) and machine learning. We conducted an intelligent deployment of RIS (i-Dris) prototype, including dual-band auto-guided vehicle (AGV) assisted RISs associated with an mmWave base station (BS) and a receiver. The RISs are deployed on the AGV with configured incident/reflection angles. While, both the mmWave BS and receiver are associated with an edge server monitoring downlink packets for obtaining system throughput. We have designed a federated multi-agent reinforcement learning scheme associated with several AGV-RIS agents and sub-agents per AGV-RIS consisting of the deployment of position, height, orientation and elevation angles. The experimental results presented the stationary measurement in different aspects and scenarios. The i-Dris can reach up to 980 Mbps transmission throughput under a bandwidth of 100 MHz with comparably low complexity as well as rapid deployment, which outperforms the other existing works. At last, we highlight some opportunities and future issues in leveraging RIS-empowered wireless communication networks.

To promote cooperation and strengthen the individual impact on the collective outcome in social dilemmas, we propose the Environmental-impact Multi-Agent Reinforcement Learning (EMuReL) method where each agent estimates the "environmental impact" of every other agent, that is, the difference in the current environment state compared to the hypothetical environment in the absence of that other agent. Inspired by the Inequity Aversion model, the agent then compares its own reward with those of its fellows multiplied by their environmental impacts. If its reward exceeds the scaled reward of one of its fellows, the agent takes "social responsibility" toward that fellow by reducing its own reward. Therefore, the less influential an agent is in reaching the current state, the more social responsibility is taken by other agents. Experiments in the Cleanup (resp. Harvest) test environment demonstrate that agents trained based on EMuReL learn to cooperate more effectively and obtain $54\%$ ($39\%$) and $20\%$ ($44\%$) more total rewards while preserving the same cooperation levels compared to when they are trained based on the two state-of-the-art reward reshaping methods inequity aversion and social influence.