In this paper, we place deep Q-learning into a control-oriented perspective and study its learning dynamics with well-established techniques from robust control. We formulate an uncertain linear time-invariant model by means of the neural tangent kernel to describe learning. We show the instability of learning and analyze the agent's behavior in frequency-domain. Then, we ensure convergence via robust controllers acting as dynamical rewards in the loss function. We synthesize three controllers: state-feedback gain scheduling H2, dynamic Hinf, and constant gain Hinf controllers. Setting up the learning agent with a control-oriented tuning methodology is more transparent and has well-established literature compared to the heuristics in reinforcement learning. In addition, our approach does not use a target network and randomized replay memory. The role of the target network is overtaken by the control input, which also exploits the temporal dependency of samples (opposed to a randomized memory buffer). Numerical simulations in different OpenAI Gym environments suggest that the Hinf controlled learning performs slightly better than Double deep Q-learning.

Deep reinforcement learning (DRL) has empowered a variety of artificial intelligence fields, including pattern recognition, robotics, recommendation-systems, and gaming. Similarly, graph neural networks (GNN) have also demonstrated their superior performance in supervised learning for graph-structured data. In recent times, the fusion of GNN with DRL for graph-structured environments has attracted a lot of attention. This paper provides a comprehensive review of these hybrid works. These works can be classified into two categories: (1) algorithmic enhancement, where DRL and GNN complement each other for better utility; (2) application-specific enhancement, where DRL and GNN support each other. This fusion effectively addresses various complex problems in engineering and life sciences. Based on the review, we further analyze the applicability and benefits of fusing these two domains, especially in terms of increasing generalizability and reducing computational complexity. Finally, the key challenges in integrating DRL and GNN, and potential future research directions are highlighted, which will be of interest to the broader machine learning community.

Decision making systems trained with real-world data are deployed ubiquitously in our daily life, for example, in regard to credit, education, and medical care. However, those decision systems may demonstrate discrimination against disadvantaged groups due to the biases in the data [16]. In order to mitigate this issue, many have proposed to impose fairness constraints [16, 20] on the decision, such that certain statistical parity properties are achieved. Despite the fact that fair learning has been extensively studied, most of this work is in the static setting without considering the sequential feedback effects of decisions. At the same time, in many scenarios, algorithmic decisions may incur changes in the underlying features or qualification status of individuals, which further feeds back to the decision making process; for example, banks' decision may induce borrowers to react, for example changing their FICO score by closing credit cards.

Hybrid FSO/RF system requires an efficient FSO and RF link switching mechanism to improve the system capacity by realizing the complementary benefits of both the links. The dynamics of network conditions, such as fog, dust, and sand storms compound the link switching problem and control complexity. To address this problem, we initiate the study of deep reinforcement learning (DRL) for link switching of hybrid FSO/RF systems. Specifically, in this work, we focus on actor-critic called Actor/Critic-FSO/RF and Deep-Q network (DQN) called DQN-FSO/RF for FSO/RF link switching under atmospheric turbulences. To formulate the problem, we define the state, action, and reward function of a hybrid FSO/RF system. DQN-FSO/RF frequently updates the deployed policy that interacts with the environment in a hybrid FSO/RF system, resulting in high switching costs. To overcome this, we lift this problem to ensemble consensus-based representation learning for deep reinforcement called DQNEnsemble-FSO/RF. The proposed novel DQNEnsemble-FSO/RF DRL approach uses consensus learned features representations based on an ensemble of asynchronous threads to update the deployed policy. Experimental results corroborate that the proposed DQNEnsemble-FSO/RF's consensus-learned features switching achieves better performance than Actor/Critic-FSO/RF, DQN-FSO/RF, and MyOpic for FSO/RF link switching while keeping the switching cost significantly low.

We introduce CAMRL, the first curriculum-based asymmetric multi-task learning (AMTL) algorithm for dealing with multiple reinforcement learning (RL) tasks altogether. To mitigate the negative influence of customizing the one-off training order in curriculum-based AMTL, CAMRL switches its training mode between parallel single-task RL and asymmetric multi-task RL (MTRL), according to an indicator regarding the training time, the overall performance, and the performance gap among tasks. To leverage the multi-sourced prior knowledge flexibly and to reduce negative transfer in AMTL, we customize a composite loss with multiple differentiable ranking functions and optimize the loss through alternating optimization and the Frank-Wolfe algorithm. The uncertainty-based automatic adjustment of hyper-parameters is also applied to eliminate the need of laborious hyper-parameter analysis during optimization. By optimizing the composite loss, CAMRL predicts the next training task and continuously revisits the transfer matrix and network weights. We have conducted experiments on a wide range of benchmarks in multi-task RL, covering Gym-minigrid, Meta-world, Atari video games, vision-based PyBullet tasks, and RLBench, to show the improvements of CAMRL over the corresponding single-task RL algorithm and state-of-the-art MTRL algorithms. The code is available at: https://github.com/huanghanchi/CAMRL

The connectivity-aware path design is crucial in the effective deployment of autonomous Unmanned Aerial Vehicles (UAVs). Recently, Reinforcement Learning (RL) algorithms have become the popular approach to solving this type of complex problem, but RL algorithms suffer slow convergence. In this paper, we propose a Transfer Learning (TL) approach, where we use a teacher policy previously trained in an old domain to boost the path learning of the agent in the new domain. As the exploration processes and the training continue, the agent refines the path design in the new domain based on the subsequent interactions with the environment. We evaluate our approach considering an old domain at sub-6 GHz and a new domain at millimeter Wave (mmWave). The teacher path policy, previously trained at sub-6 GHz path, is the solution to a connectivity-aware path problem that we formulate as a constrained Markov Decision Process (CMDP). We employ a Lyapunov-based model-free Deep Q-Network (DQN) to solve the path design at sub-6 GHz that guarantees connectivity constraint satisfaction. We empirically demonstrate the effectiveness of our approach for different urban environment scenarios. The results demonstrate that our proposed approach is capable of reducing the training time considerably at mmWave.

We study non-parametric estimation of the value function of an infinite-horizon $\gamma$-discounted Markov reward process (MRP) using observations from a single trajectory. We provide non-asymptotic guarantees for a general family of kernel-based multi-step temporal difference (TD) estimates, including canonical $K$-step look-ahead TD for $K = 1, 2, \ldots$ and the TD$(\lambda)$ family for $\lambda \in [0,1)$ as special cases. Our bounds capture its dependence on Bellman fluctuations, mixing time of the Markov chain, any mis-specification in the model, as well as the choice of weight function defining the estimator itself, and reveal some delicate interactions between mixing time and model mis-specification. For a given TD method applied to a well-specified model, its statistical error under trajectory data is similar to that of i.i.d. sample transition pairs, whereas under mis-specification, temporal dependence in data inflates the statistical error. However, any such deterioration can be mitigated by increased look-ahead. We complement our upper bounds by proving minimax lower bounds that establish optimality of TD-based methods with appropriately chosen look-ahead and weighting, and reveal some fundamental differences between value function estimation and ordinary non-parametric regression.

The past few years have seen rapid progress in combining reinforcement learning (RL) with deep learning. Various breakthroughs ranging from games to robotics have spurred the interest in designing sophisticated RL algorithms and systems. However, the prevailing workflow in RL is to learn tabula rasa, which may incur computational inefficiency. This precludes continuous deployment of RL algorithms and potentially excludes researchers without large-scale computing resources. In many other areas of machine learning, the pretraining paradigm has shown to be effective in acquiring transferable knowledge, which can be utilized for a variety of downstream tasks. Recently, we saw a surge of interest in Pretraining for Deep RL with promising results. However, much of the research has been based on different experimental settings. Due to the nature of RL, pretraining in this field is faced with unique challenges and hence requires new design principles. In this survey, we seek to systematically review existing works in pretraining for deep reinforcement learning, provide a taxonomy of these methods, discuss each sub-field, and bring attention to open problems and future directions.

Policies produced by deep reinforcement learning are typically characterised by their learning curves, but they remain poorly understood in many other respects. ReLU-based policies result in a partitioning of the input space into piecewise linear regions. We seek to understand how observed region counts and their densities evolve during deep reinforcement learning using empirical results that span a range of continuous control tasks and policy network dimensions. Intuitively, we may expect that during training, the region density increases in the areas that are frequently visited by the policy, thereby affording fine-grained control. We use recent theoretical and empirical results for the linear regions induced by neural networks in supervised learning settings for grounding and comparison of our results. Empirically, we find that the region density increases only moderately throughout training, as measured along fixed trajectories coming from the final policy. However, the trajectories themselves also increase in length during training, and thus the region densities decrease as seen from the perspective of the current trajectory. Our findings suggest that the complexity of deep reinforcement learning policies does not principally emerge from a significant growth in the complexity of functions observed on-and-around trajectories of the policy.

Given a dataset of expert agent interactions with an environment of interest, a viable method to extract an effective agent policy is to estimate the maximum likelihood policy indicated by this data. This approach is commonly referred to as behavioral cloning (BC). In this work, we describe a key disadvantage of BC that arises due to the maximum likelihood objective function; namely that BC is mean-seeking with respect to the state-conditional expert action distribution when the learner's policy is represented with a Gaussian. To address this issue, we introduce a modified version of BC, Adversarial Behavioral Cloning (ABC), that exhibits mode-seeking behavior by incorporating elements of GAN (generative adversarial network) training. We evaluate ABC on toy domains and a domain based on Hopper from the DeepMind Control suite, and show that it outperforms standard BC by being mode-seeking in nature.