In this advanced course on deep reinforcement learning, you will learn how to implement policy gradient, actor critic, deep deterministic policy gradient (DDPG), twin delayed deep deterministic policy gradient (TD3), and soft actor critic (SAC) algorithms in a variety of challenging environments from the Open AI gym. There will be a strong focus on dealing with environments with continuous action spaces, which is of particular interest for those looking to do research into robotic control with deep reinforcement learning. Rather than being a course that spoon feeds the student, here you are going to learn to read deep reinforcement learning research papers on your own, and implement them from scratch. You will learn a repeatable framework for quickly implementing the algorithms in advanced research papers. Mastering the content in this course will be a quantum leap in your capabilities as an artificial intelligence engineer, and will put you in a league of your own among students who are reliant on others to break down complex ideas for them.

In this post I present our recent NeurIPS 2022 paper (co-authored with Yinlam Chow, Mohammad Ghavamzadeh and Shie Mannor) about risk-averse reinforcement learning (RL). I discuss why and how risk aversion is applied to RL, what its limitations are, and how we propose to overcome them. An application to accidents prevention in autonomous driving is demonstrated. Our code is also available on GitHub. Risk-averse RL is crucial when applying RL to risk-sensitive real-world problems.

We study reward-free reinforcement learning (RL) under general non-linear function approximation, and establish sample efficiency and hardness results under various standard structural assumptions. On the positive side, we propose the RFOLIVE (Reward-Free OLIVE) algorithm for sample-efficient reward-free exploration under minimal structural assumptions, which covers the previously studied settings of linear MDPs (Jin et al., 2020b), linear completeness (Zanette et al., 2020b) and low-rank MDPs with unknown representation (Modi et al., 2021). Our analyses indicate that the explorability or reachability assumptions, previously made for the latter two settings, are not necessary statistically for reward-free exploration. On the negative side, we provide a statistical hardness result for both reward-free and reward-aware exploration under linear completeness assumptions when the underlying features are unknown, showing an exponential separation between low-rank and linear completeness settings.

We evaluate the use of original game curricula supported by the Atari 2600 console as a heterogeneous transfer benchmark for deep reinforcement learning agents. Game designers created curricula using combinations of several discrete modifications to the basic versions of games such as Space Invaders, Breakout and Freeway, making them progressively more challenging for human players. By formally organising these modifications into several factors of variation, we are able to show that Analyses of Variance (ANOVA) are a potent tool for studying the effects of human-relevant domain changes on the learning and transfer performance of a deep reinforcement learning agent. Since no manual task engineering is needed on our part, leveraging the original multi-factorial design avoids the pitfalls of unintentionally biasing the experimental setup. We find that game design factors have a large and statistically significant impact on an agent's ability to learn, and so do their combinatorial interactions. Furthermore, we show that zero-shot transfer from the basic games to their respective variations is possible, but the variance in performance is also largely explained by interactions between factors. As such, we argue that Atari game curricula offer a challenging benchmark for transfer learning in RL, that can help the community better understand the generalisation capabilities of RL agents along dimensions which meaningfully impact human generalisation performance. As a start, we report that value-function finetuning of regularly trained agents achieves positive transfer in a majority of cases, but significant headroom for algorithmic innovation remains. We conclude with the observation that selective transfer from multiple variants could further improve performance.

Long-term fairness is an important factor of consideration in designing and deploying learning-based decision systems in high-stake decision-making contexts. Recent work has proposed the use of Markov Decision Processes (MDPs) to formulate decision-making with long-term fairness requirements in dynamically changing environments, and demonstrated major challenges in directly deploying heuristic and rule-based policies that worked well in static environments. We show that policy optimization methods from deep reinforcement learning can be used to find strictly better decision policies that can often achieve both higher overall utility and less violation of the fairness requirements, compared to previously-known strategies. In particular, we propose new methods for imposing fairness requirements in policy optimization by regularizing the advantage evaluation of different actions. Our proposed methods make it easy to impose fairness constraints without reward engineering or sacrificing training efficiency. We perform detailed analyses in three established case studies, including attention allocation in incident monitoring, bank loan approval, and vaccine distribution in population networks.

We study video-grounded dialogue generation, where a response is generated based on the dialogue context and the associated video. The primary challenges of this task lie in (1) the difficulty of integrating video data into pre-trained language models (PLMs) which presents obstacles to exploiting the power of large-scale pre-training; and (2) the necessity of taking into account the complementarity of various modalities throughout the reasoning process. Although having made remarkable progress in video-grounded dialogue generation, existing methods still fall short when it comes to integrating with PLMs in a way that allows information from different modalities to complement each other. To alleviate these issues, we first propose extracting pertinent information from videos and turning it into reasoning paths that are acceptable to PLMs. Additionally, we propose a multi-agent reinforcement learning method to collaboratively perform reasoning on different modalities (i.e., video and dialogue context). Empirical experiment results on two public datasets indicate that the proposed model can significantly outperform state-of-the-art models by large margins on both automatic and human evaluations.

Current methods for 3D object reconstruction from a set of planar cross-sections still struggle to capture detailed topology or require a considerable number of cross-sections. In this paper, we present, to the best of our knowledge the first 3D shape reconstruction network to solve this task which additionally uses orthographic projections of the shape. Our method is based on applying a Reinforcement Learning algorithm to learn how to effectively parse the shape using a trial-and-error scheme relying on scalar rewards. This method cuts a part of a 3D shape in each step which is then approximated as a polygon mesh. The agent aims to maximize the reward that depends on the accuracy of surface reconstruction for the approximated parts. We also consider pre-training of the network for faster learning using demonstrations generated by a heuristic approach. Experiments show that our training algorithm which benefits from both imitation learning and also self exploration, learns efficient policies faster, which results the agent to produce visually compelling results.

The building sector has been recognized as one of the primary sectors for worldwide energy consumption. Improving the energy efficiency of the building sector can help reduce the operation cost and reduce the greenhouse gas emission. The energy management system (EMS) can monitor and control the operations of built-in appliances in buildings, so an efficient EMS is of crucial importance to improve the building operation efficiency and maintain safe operations. With the growing penetration of renewable energy and electrical appliances, increasing attention has been paid to the development of intelligent building EMS. Recently, reinforcement learning (RL) has been applied for building EMS and has shown promising potential. However, most of the current RL-based EMS solutions would need a large amount of data to learn a reliable control policy, which limits the applicability of these solutions in the real world. In this work, we propose MetaEMS, which can help achieve better energy management performance with the benefits of RL and meta-learning. Experiment results showcase that our proposed MetaEMS can adapt faster to environment changes and perform better in most situations compared with other baselines.

Prompting has shown impressive success in enabling large pretrained language models (LMs) to perform diverse NLP tasks, especially when only few downstream data are available. Automatically finding the optimal prompt for each task, however, is challenging. Most existing work resorts to tuning soft prompt (e.g., embeddings) which falls short of interpretability, reusability across LMs, and applicability when gradients are not accessible. Discrete prompt, on the other hand, is difficult to optimize, and is often created by "enumeration (e.g., paraphrasing)-then-selection" heuristics that do not explore the prompt space systematically. This paper proposes RLPrompt, an efficient discrete prompt optimization approach with reinforcement learning (RL). RLPrompt formulates a parameter-efficient policy network that generates the desired discrete prompt after training with reward. To overcome the complexity and stochasticity of reward signals by the large LM environment, we incorporate effective reward stabilization that substantially enhances the training efficiency. RLPrompt is flexibly applicable to different types of LMs, such as masked (e.g., BERT) and left-to-right models (e.g., GPTs), for both classification and generation tasks. Experiments on few-shot classification and unsupervised text style transfer show superior performance over a wide range of existing finetuning or prompting methods. Interestingly, the resulting optimized prompts are often ungrammatical gibberish text; and surprisingly, those gibberish prompts are transferrable between different LMs to retain significant performance, indicating LM prompting may not follow human language patterns.

Maximum likelihood estimation (MLE) is the predominant algorithm for training text generation models. This paradigm relies on direct supervision examples, which is not applicable to many emerging applications, such as generating adversarial attacks or generating prompts to control language models. Reinforcement learning (RL) on the other hand offers a more flexible solution by allowing users to plug in arbitrary task metrics as reward. Yet previous RL algorithms for text generation, such as policy gradient (on-policy RL) and Q-learning (off-policy RL), are often notoriously inefficient or unstable to train due to the large sequence space and the sparse reward received only at the end of sequences. In this paper, we introduce a new RL formulation for text generation from the soft Q-learning (SQL) perspective. It enables us to draw from the latest RL advances, such as path consistency learning, to combine the best of on-/off-policy updates, and learn effectively from sparse reward. We apply the approach to a wide range of novel text generation tasks, including learning from noisy/negative examples, adversarial attacks, and prompt generation. Experiments show our approach consistently outperforms both task-specialized algorithms and the previous RL methods.