The application of reinforcement learning in credit scoring has created a unique setting for contextual logistic bandit that does not conform to the usual exploration-exploitation tradeoff but rather favors exploration-free algorithms. Through sufficient randomness in a pool of observable contexts, the reinforcement learning agent can simultaneously exploit an action with the highest reward while still learning more about the structure governing that environment. Thus, it is the case that greedy algorithms consistently outperform algorithms with efficient exploration, such as Thompson sampling. However, in a more pragmatic scenario in credit scoring, lenders can, to a degree, classify each borrower as a separate group, and learning about the characteristics of each group does not infer any information to another group. Through extensive simulations, we show that Thompson sampling dominates over greedy algorithms given enough timesteps which increase with the complexity of underlying features.

Active target sensing is the task of discovering and classifying an unknown number of targets in an environment and is critical in search-and-rescue missions. This paper develops a deep reinforcement learning approach to plan informative trajectories that increase the likelihood for an uncrewed aerial vehicle (UAV) to discover missing targets. Our approach efficiently (1) explores the environment to discover new targets, (2) exploits its current belief of the target states and incorporates inaccurate sensor models for high-fidelity classification, and (3) generates dynamically feasible trajectories for an agile UAV by employing a motion primitive library. Extensive simulations on randomly generated environments show that our approach is more efficient in discovering and classifying targets than several other baselines. A unique characteristic of our approach, in contrast to heuristic informative path planning approaches, is that it is robust to varying amounts of deviations of the prior belief from the true target distribution, thereby alleviating the challenge of designing heuristics specific to the application conditions.

Recent advances in batch (offline) reinforcement learning have shown promising results in learning from available offline data and proved offline reinforcement learning to be an essential toolkit in learning control policies in a model-free setting. An offline reinforcement learning algorithm applied to a dataset collected by a suboptimal non-learning-based algorithm can result in a policy that outperforms the behavior agent used to collect the data. Such a scenario is frequent in robotics, where existing automation is collecting operational data. Although offline learning techniques can learn from data generated by a sub-optimal behavior agent, there is still an opportunity to improve the sample complexity of existing offline reinforcement learning algorithms by strategically introducing human demonstration data into the training process. To this end, we propose a novel approach that uses uncertainty estimation to trigger the injection of human demonstration data and guide policy training towards optimal behavior while reducing overall sample complexity. Our experiments show that this approach is more sample efficient when compared to a naive way of combining expert data with data collected from a sub-optimal agent. We augmented an existing offline reinforcement learning algorithm Conservative Q-Learning with our approach and performed experiments on data collected from MuJoCo and OffWorld Gym learning environments.

Reinforcement learning can enable robots to navigate to distant goals while optimizing user-specified reward functions, including preferences for following lanes, staying on paved paths, or avoiding freshly mowed grass. However, online learning from trial-and-error for real-world robots is logistically challenging, and methods that instead can utilize existing datasets of robotic navigation data could be significantly more scalable and enable broader generalization. In this paper, we present ReViND, the first offline RL system for robotic navigation that can leverage previously collected data to optimize user-specified reward functions in the real-world. We evaluate our system for off-road navigation without any additional data collection or fine-tuning, and show that it can navigate to distant goals using only offline training from this dataset, and exhibit behaviors that qualitatively differ based on the user-specified reward function.

In many real-world applications of control system and robotics, linear temporal logic (LTL) is a widely-used task specification language which has a compositional grammar that naturally induces temporally extended behaviours across tasks, including conditionals and alternative realizations. An important problem in RL with LTL tasks is to learn task-conditioned policies which can zero-shot generalize to new LTL instructions not observed in the training. However, because symbolic observation is often lossy and LTL tasks can have long time horizon, previous works can suffer from issues such as training sampling inefficiency and infeasibility or sub-optimality of the found solutions. In order to tackle these issues, this paper proposes a novel multi-task RL algorithm with improved learning efficiency and optimality. To achieve the global optimality of task completion, we propose to learn options dependent on the future subgoals via a novel off-policy approach. In order to propagate the rewards of satisfying future subgoals back more efficiently, we propose to train a multi-step value function conditioned on the subgoal sequence which is updated with Monte Carlo estimates of multi-step discounted returns. In experiments on three different domains, we evaluate the LTL generalization capability of the agent trained by the proposed method, showing its advantage over previous representative methods.

In heterogeneous networks (HetNets), the overlap of small cells and the macro cell causes severe cross-tier interference. Although there exist some approaches to address this problem, they usually require global channel state information, which is hard to obtain in practice, and get the sub-optimal power allocation policy with high computational complexity. To overcome these limitations, we propose a multi-agent deep reinforcement learning (MADRL) based power control scheme for the HetNet, where each access point makes power control decisions independently based on local information. To promote cooperation among agents, we develop a penalty-based Q learning (PQL) algorithm for MADRL systems. By introducing regularization terms in the loss function, each agent tends to choose an experienced action with high reward when revisiting a state, and thus the policy updating speed slows down. In this way, an agent's policy can be learned by other agents more easily, resulting in a more efficient collaboration process. Simulation results show that our proposed PQL can learn the desired power control policy from a dynamic environment where the locations of users change episodically and outperform existing DTE MADRL algorithms. The authors are with the Department of Electrical and Electronic Engineering, Imperial College London, London SW7 2AZ, U.K. (e-mail: k.xu21@imperial.ac.uk; huynh.nguyen@imperial.ac.uk; geoffrey.li@imperial.ac.uk) In conventional cellular networks, a macro base station (BS) needs to provide access to the core network for all user devices (UDs) in the cell.

As AI systems become more capable, we would like to enlist their help to supervise other AIs. We experiment with methods for training a harmless AI assistant through self-improvement, without any human labels identifying harmful outputs. The only human oversight is provided through a list of rules or principles, and so we refer to the method as 'Constitutional AI'. The process involves both a supervised learning and a reinforcement learning phase. In the supervised phase we sample from an initial model, then generate self-critiques and revisions, and then finetune the original model on revised responses. In the RL phase, we sample from the finetuned model, use a model to evaluate which of the two samples is better, and then train a preference model from this dataset of AI preferences. We then train with RL using the preference model as the reward signal, i.e. we use 'RL from AI Feedback' (RLAIF). As a result we are able to train a harmless but non-evasive AI assistant that engages with harmful queries by explaining its objections to them. Both the SL and RL methods can leverage chain-of-thought style reasoning to improve the human-judged performance and transparency of AI decision making. These methods make it possible to control AI behavior more precisely and with far fewer human labels.

We consider a multi-agent episodic MDP setup where an agent (leader) takes action at each step of the episode followed by another agent (follower). The state evolution and rewards depend on the joint action pair of the leader and the follower. Such type of interactions can find applications in many domains such as smart grids, mechanism design, security, and policymaking. We are interested in how to learn policies for both the players with provable performance guarantee under a bandit feedback setting. We focus on a setup where both the leader and followers are {\em non-myopic}, i.e., they both seek to maximize their rewards over the entire episode and consider a linear MDP which can model continuous state-space which is very common in many RL applications. We propose a {\em model-free} RL algorithm and show that $\tilde{\mathcal{O}}(\sqrt{d^3H^3T})$ regret bounds can be achieved for both the leader and the follower, where $d$ is the dimension of the feature mapping, $H$ is the length of the episode, and $T$ is the total number of steps under the bandit feedback information setup. Thus, our result holds even when the number of states becomes infinite. The algorithm relies on {\em novel} adaptation of the LSVI-UCB algorithm. Specifically, we replace the standard greedy policy (as the best response) with the soft-max policy for both the leader and the follower. This turns out to be key in establishing uniform concentration bound for the value functions. To the best of our knowledge, this is the first sub-linear regret bound guarantee for the Markov games with non-myopic followers with function approximation.

Members of the AIRL lab at Imperial College, and authors of the reported work in this blog post. From left to right: Bryan Lim, Dr Antoine Cully (director of the AIRL lab), Manon Flageat, Luca Grillotti, Dr Simón C Smith, and Maxime Allard. Learning and finding different solutions to the same problem is commonly associated with creativity and adaptation, which are important characteristics of intelligence. In the AIRL lab at Imperial College, we believe in the importance of diversity in learning algorithms. With this focus in mind, we develop learning algorithms known as Quality-Diversity algorithms.

This paper is a technical overview of DeepMind and Google's recent work on reinforcement learning for controlling commercial cooling systems. Building on expertise that began with cooling Google's data centers more efficiently, we recently conducted live experiments on two real-world facilities in partnership with Trane Technologies, a building management system provider. These live experiments had a variety of challenges in areas such as evaluation, learning from offline data, and constraint satisfaction. Our paper describes these challenges in the hope that awareness of them will benefit future applied RL work. We also describe the way we adapted our RL system to deal with these challenges, resulting in energy savings of approximately 9% and 13% respectively at the two live experiment sites.