Welcome to Deep Reinforcement Learning 2.0! In this course, we will learn and implement a new incredibly smart AI model, called the Twin-Delayed DDPG, which combines state of the art techniques in Artificial Intelligence including continuous Double Deep Q-Learning, Policy Gradient, and Actor Critic. The model is so strong that for the first time in our courses, we are able to solve the most challenging virtual AI applications (training an ant/spider and a half humanoid to walk and run across a field). In this part we will study all the fundamentals of Artificial Intelligence which will allow you to understand and master the AI of this course. These include Q-Learning, Deep Q-Learning, Policy Gradient, Actor-Critic and more.

An important challenge in reinforcement learning is training agents that can solve a wide variety of tasks. If tasks depend on each other (e.g. needing to learn to walk before learning to run), curriculum learning can speed up learning by focusing on the next best task to learn. We explore curriculum learning in a complex, visual domain with many hard exploration challenges: Minecraft. We find that learning progress (defined as a change in success probability of a task) is a reliable measure of learnability for automatically constructing an effective curriculum. We introduce a learning-progress based curriculum and test it on a complex reinforcement learning problem (called "Simon Says") where an agent is instructed to obtain a desired goal item. Many of the required skills depend on each other. Experiments demonstrate that: (1) a within-episode exploration bonus for obtaining new items improves performance, (2) dynamically adjusting this bonus across training such that it only applies to items the agent cannot reliably obtain yet further increases performance, (3) the learning-progress based curriculum elegantly follows the learning curve of the agent, and (4) when the learning-progress based curriculum is combined with the dynamic exploration bonus it learns much more efficiently and obtains far higher performance than uniform baselines. These results suggest that combining intra-episode and across-training exploration bonuses with learning progress creates a promising method for automated curriculum generation, which may substantially increase our ability to train more capable, generally intelligent agents.

We introduce the "inverse bandit" problem of estimating the rewards of a multi-armed bandit instance from observing the learning process of a low-regret demonstrator. Existing approaches to the related problem of inverse reinforcement learning assume the execution of an optimal policy, and thereby suffer from an identifiability issue. In contrast, our paradigm leverages the demonstrator's behavior en route to optimality, and in particular, the exploration phase, to obtain consistent reward estimates. We develop simple and efficient reward estimation procedures for demonstrations within a class of upper-confidence-based algorithms, showing that reward estimation gets progressively easier as the regret of the algorithm increases. We match these upper bounds with information-theoretic lower bounds that apply to any demonstrator algorithm, thereby characterizing the optimal tradeoff between exploration and reward estimation. Extensive empirical evaluations on both synthetic data and simulated experimental design data from the natural sciences corroborate our theoretical results.

One of the types of learning that we hear about in Machine Learning is Reinforcement Learning, where an agent learns a goal in an environment, known or unknown, through reward and punishment. Unlike learning methods such as Supervised and Unsupervised learning, Reinforcement learning does not require data at all. In my class CS4100, the course briefly touched on the practices of this learning method so I wanted to explore a bit further. A lot of the applications of Reinforcement learning are in games or complex and computationally expensive real-world problems, so it is hard to find something to "meaningfully" apply reinforcement learning to. Nonetheless, this is still a super interesting topic where an agent can build a policy that maximizes the reward function without knowing its environment.

This article is part of "the philosophy of artificial intelligence," a series of posts that explore the ethical, moral, and social implications of AI today and in the future Last week, I wrote an analysis of "Reward Is Enough," a paper by scientists at DeepMind. As the title suggests, the researchers hypothesize that the right reward is all you need to create the abilities associated with intelligence, such as perception, motor functions, and language. This is in contrast with AI systems that try to replicate specific functions of natural intelligence such as classifying images, navigating physical environments, or completing sentences. The researchers go as far as suggesting that with well-defined reward, a complex environment, and the right reinforcement learning algorithm, we will be able to reach artificial general intelligence, the kind of problem-solving and cognitive abilities found in humans and, to a lesser degree, in animals. The article and the paper triggered a heated debate on social media, with reactions going from full support of the idea to outright rejection. Of course, both sides make valid claims.

Various algorithms in reinforcement learning exhibit dramatic variability in their convergence rates and ultimate accuracy as a function of the problem structure. Such instance-specific behavior is not captured by existing global minimax bounds, which are worst-case in nature. We analyze the problem of estimating optimal $Q$-value functions for a discounted Markov decision process with discrete states and actions and identify an instance-dependent functional that controls the difficulty of estimation in the $\ell_\infty$-norm. Using a local minimax framework, we show that this functional arises in lower bounds on the accuracy on any estimation procedure. In the other direction, we establish the sharpness of our lower bounds, up to factors logarithmic in the state and action spaces, by analyzing a variance-reduced version of $Q$-learning. Our theory provides a precise way of distinguishing "easy" problems from "hard" ones in the context of $Q$-learning, as illustrated by an ensemble with a continuum of difficulty.

If you walk down the street shouting out the names of every object you see -- garbage truck! But if you go through an obstacle course, and you show them how to navigate a series of challenges to get to the end unscathed, they would. Most machine learning algorithms are shouting names in the street. They perform perceptive tasks that a person can do in under a second. But another kind of AI -- deep reinforcement learning -- is strategic.

Maritime autonomous transportation has played a crucial role in the globalization of the world economy. Deep Reinforcement Learning (DRL) has been applied to automatic path planning to simulate vessel collision avoidance situations in open seas. End-to-end approaches that learn complex mappings directly from the input have poor generalization to reach the targets in different environments. In this work, we present a new strategy called state-action rotation to improve agent's performance in unseen situations by rotating the obtained experience (state-action-state) and preserving them in the replay buffer. We designed our model based on Deep Deterministic Policy Gradient, local view maker, and planner. Our agent uses two deep Convolutional Neural Networks to estimate the policy and action-value functions. The proposed model was exhaustively trained and tested in maritime scenarios with real maps from cities such as Montreal and Halifax. Experimental results show that the state-action rotation on top of the CVN consistently improves the rate of arrival to a destination (RATD) by up 11.96% with respect to the Vessel Navigator with Planner and Local View (VNPLV), as well as it achieves superior performance in unseen mappings by up 30.82%. Our proposed approach exhibits advantages in terms of robustness when tested in a new environment, supporting the idea that generalization can be achieved by using state-action rotation.

Solving partially-observable Markov decision processes (POMDPs) is critical when applying deep reinforcement learning (DRL) to real-world robotics problems, where agents have an incomplete view of the world. We present graph convolutional memory (GCM) for solving POMDPs using deep reinforcement learning. Unlike recurrent neural networks (RNNs) or transformers, GCM embeds domain-specific priors into the memory recall process via a knowledge graph. By encapsulating priors in the graph, GCM adapts to specific tasks but remains applicable to any DRL task. Using graph convolutions, GCM extracts hierarchical graph features, analogous to image features in a convolutional neural network (CNN). We show GCM outperforms long short-term memory (LSTM), gated transformers for reinforcement learning (GTrXL), and differentiable neural computers (DNCs) on control, long-term non-sequential recall, and 3D navigation tasks while using significantly fewer parameters.

Model-based Reinforcement Learning (MBRL) algorithms have been traditionally designed with the goal of learning accurate dynamics of the environment. This introduces a mismatch between the objectives of model-learning and the overall learning problem of finding an optimal policy. Value-aware model learning, an alternative model-learning paradigm to maximum likelihood, proposes to inform model-learning through the value function of the learnt policy. While this paradigm is theoretically sound, it does not scale beyond toy settings. In this work, we propose a novel value-aware objective that is an upper bound on the absolute performance difference of a policy across two models. Further, we propose a general purpose algorithm that modifies the standard MBRL pipeline -- enabling learning with value aware objectives. Our proposed objective, in conjunction with this algorithm, is the first successful instantiation of value-aware MBRL on challenging continuous control environments, outperforming previous value-aware objectives and with competitive performance w.r.t. MLE-based MBRL approaches.