This paper provides the first formalisation and empirical demonstration of a particular safety concern in reinforcement learning (RL)-based news and social media recommendation algorithms. This safety concern is what we call "user tampering" -- a phenomenon whereby an RL-based recommender system may manipulate a media user's opinions, preferences and beliefs via its recommendations as part of a policy to increase long-term user engagement. We provide a simulation study of a media recommendation problem constrained to the recommendation of political content, and demonstrate that a Q-learning algorithm consistently learns to exploit its opportunities to 'polarise' simulated 'users' with its early recommendations in order to have more consistent success with later recommendations catering to that polarisation. Finally, we argue that given our findings, designing an RL-based recommender system which cannot learn to exploit user tampering requires making the metric for the recommender's success independent of observable signals of user engagement, and thus that a media recommendation system built solely with RL is necessarily either unsafe, or almost certainly commercially unviable.

In light of the emergence of deep reinforcement learning (DRL) in recommender systems research and several fruitful results in recent years, this survey aims to provide a timely and comprehensive overview of the recent trends of deep reinforcement learning in recommender systems. We start with the motivation of applying DRL in recommender systems. Then, we provide a taxonomy of current DRL-based recommender systems and a summary of existing methods. We discuss emerging topics and open issues, and provide our perspective on advancing the domain. This survey serves as introductory material for readers from academia and industry into the topic and identifies notable opportunities for further research.

Meta-learning empowers artificial intelligence to increase its efficiency by learning how to learn. Unlocking this potential involves overcoming a challenging meta-optimisation problem that often exhibits ill-conditioning, and myopic meta-objectives. We propose an algorithm that tackles these issues by letting the meta-learner teach itself. The algorithm first bootstraps a target from the meta-learner, then optimises the meta-learner by minimising the distance to that target under a chosen (pseudo-)metric. Focusing on meta-learning with gradients, we establish conditions that guarantee performance improvements and show that the improvement is related to the target distance. Thus, by controlling curvature, the distance measure can be used to ease meta-optimization, for instance by reducing ill-conditioning. Further, the bootstrapping mechanism can extend the effective meta-learning horizon without requiring backpropagation through all updates. The algorithm is versatile and easy to implement. We achieve a new state-of-the art for model-free agents on the Atari ALE benchmark, improve upon MAML in few-shot learning, and demonstrate how our approach opens up new possibilities by meta-learning efficient exploration in a Q-learning agent.

Mean field control (MFC) is an effective way to mitigate the curse of dimensionality of cooperative multi-agent reinforcement learning (MARL) problems. This work considers a collection of $N_{\mathrm{pop}}$ heterogeneous agents that can be segregated into $K$ classes such that the $k$-th class contains $N_k$ homogeneous agents. We aim to prove approximation guarantees of the MARL problem for this heterogeneous system by its corresponding MFC problem. We consider three scenarios where the reward and transition dynamics of all agents are respectively taken to be functions of $(1)$ joint state and action distributions across all classes, $(2)$ individual distributions of each class, and $(3)$ marginal distributions of the entire population. We show that, in these cases, the $K$-class MARL problem can be approximated by MFC with errors given as $e_1=\mathcal{O}(\frac{\sqrt{|\mathcal{X}||\mathcal{U}|}}{N_{\mathrm{pop}}}\sum_{k}\sqrt{N_k})$, $e_2=\mathcal{O}(\sqrt{|\mathcal{X}||\mathcal{U}|}\sum_{k}\frac{1}{\sqrt{N_k}})$ and $e_3=\mathcal{O}\left(\sqrt{|\mathcal{X}||\mathcal{U}|}\left[\frac{A}{N_{\mathrm{pop}}}\sum_{k\in[K]}\sqrt{N_k}+\frac{B}{\sqrt{N_{\mathrm{pop}}}}\right]\right)$, respectively, where $A, B$ are some constants and $|\mathcal{X}|,|\mathcal{U}|$ are the sizes of state and action spaces of each agent. Finally, we design a Natural Policy Gradient (NPG) based algorithm that, in the three cases stated above, can converge to an optimal MARL policy within $\mathcal{O}(e_j)$ error with a sample complexity of $\mathcal{O}(e_j^{-3})$, $j\in\{1,2,3\}$, respectively.

As online shopping prevails and e-commerce platforms emerge, there is a tremendous number of parcels being transported every day. Thus, it is crucial for the logistics industry on how to assign a candidate logistics route for each shipping parcel properly as it leaves a significant impact on the total logistics cost optimization and business constraints satisfaction such as transit hub capacity and delivery proportion of delivery providers. This online route-assignment problem can be viewed as a constrained online decision-making problem. Notably, the large amount (beyond ${10^5}$) of daily parcels, the variability and non-Markovian characteristics of parcel information impose difficulties on attaining (near-) optimal solution without violating constraints excessively. In this paper, we develop a model-free DRL approach named PPO-RA, in which Proximal Policy Optimization (PPO) is improved with dedicated techniques to address the challenges for route assignment (RA). The actor and critic networks use attention mechanism and parameter sharing to accommodate each incoming parcel with varying numbers and identities of candidate routes, without modeling non-Markovian parcel arriving dynamics since we make assumption of i.i.d. parcel arrival. We use recorded delivery parcel data to evaluate the performance of PPO-RA by comparing it with widely-used baselines via simulation. The results show the capability of the proposed approach to achieve considerable cost savings while satisfying most constraints.

The field of artificial intelligence (AI), regarded as one of the most enigmatic areas of science, has witnessed exponential growth in the past decade including a remarkably wide array of applications, having already impacted our everyday lives. Advances in computing power and the design of sophisticated AI algorithms have enabled computers to outperform humans in a variety of tasks, especially in the areas of computer vision and speech recognition. Yet, AI's path has never been smooth, having essentially fallen apart twice in its lifetime ('winters' of AI), both after periods of popular success ('summers' of AI). We provide a brief rundown of AI's evolution over the course of decades, highlighting its crucial moments and major turning points from inception to the present. In doing so, we attempt to learn, anticipate the future, and discuss what steps may be taken to prevent another 'winter'.

The stochastic approximation (SA) algorithm is a widely used probabilistic method for finding a solution to an equation of the form $\mathbf{f}(\boldsymbol{\theta}) = \mathbf{0}$ where $\mathbf{f} : \mathbb{R}^d \rightarrow \mathbb{R}^d$, when only noisy measurements of $\mathbf{f}(\cdot)$ are available. In the literature to date, one can make a distinction between "synchronous" updating, whereby the entire vector of the current guess $\boldsymbol{\theta}_t$ is updated at each time, and "asynchronous" updating, whereby ony one component of $\boldsymbol{\theta}_t$ is updated. In convex and nonconvex optimization, there is also the notion of "batch" updating, whereby some but not all components of $\boldsymbol{\theta}_t$ are updated at each time $t$. In addition, there is also a distinction between using a "local" clock versus a "global" clock. In the literature to date, convergence proofs when a local clock is used make the assumption that the measurement noise is an i.i.d\ sequence, an assumption that does not hold in Reinforcement Learning (RL). In this note, we provide a general theory of convergence for batch asymchronous stochastic approximation (BASA), that works whether the updates use a local clock or a global clock, for the case where the measurement noises form a martingale difference sequence. This is the most general result to date and encompasses all others.

Given the success of reinforcement learning (RL) in various domains, it is promising to explore the application of its methods to the development of intelligent and autonomous cyber agents. Enabling this development requires a representative RL training environment. To that end, this work presents CyGIL: an experimental testbed of an emulated RL training environment for network cyber operations. CyGIL uses a stateless environment architecture and incorporates the MITRE ATT&CK framework to establish a high fidelity training environment, while presenting a sufficiently abstracted interface to enable RL training. Its comprehensive action space and flexible game design allow the agent training to focus on particular advanced persistent threat (APT) profiles, and to incorporate a broad range of potential threats and vulnerabilities. By striking a balance between fidelity and simplicity, it aims to leverage state of the art RL algorithms for application to real-world cyber defence.

Many of the challenges facing today's reinforcement learning (RL) algorithms, such as robustness, generalization, transfer, and computational efficiency are closely related to compression. Prior work has convincingly argued why minimizing information is useful in the supervised learning setting, but standard RL algorithms lack an explicit mechanism for compression. The RL setting is unique because (1) its sequential nature allows an agent to use past information to avoid looking at future observations and (2) the agent can optimize its behavior to prefer states where decision making requires few bits. We take advantage of these properties to propose a method (RPC) for learning simple policies. This method brings together ideas from information bottlenecks, model-based RL, and bits-back coding into a simple and theoretically-justified algorithm. Our method jointly optimizes a latent-space model and policy to be self-consistent, such that the policy avoids states where the model is inaccurate. We demonstrate that our method achieves much tighter compression than prior methods, achieving up to 5x higher reward than a standard information bottleneck. We also demonstrate that our method learns policies that are more robust and generalize better to new tasks.

The recent progress in Reinforcement Learning applications to Resource Management presents MDPs without a deeper analysis of the impacts of design decisions on agent performance. In this paper, we compare and contrast four different MDP variations, discussing their computational requirements and impacts on agent performance by means of an empirical analysis. We conclude by showing that, in our experiments, when using Multi-Layer Perceptrons as approximation function, a compact state representation allows transfer of agents between environments, and that transferred agents have good performance and outperform specialized agents in 80\% of the tested scenarios, even without retraining.