In this section, we introduce related areas of *Policy Learning*, aiming to give readers without extra detailed domain knowledge a general understanding of alignment.

## Basic Reinforcement Learning

Reinforcement Learning (RL) enables agents to learn optimal policies by trial and error via interacting with the environment. This paradigm has successfully tackled complex tasks like optimizing algorithms, video game playing, multi-modal generation, and other domains, demonstrating its potential for decision-making and control in complex state spaces.

**Recommended Papers List**

Reinforcement Learning: An Introduction

## Click to have a preview.

The idea that we learn by interacting with our environment is probably the first to occur to us when we think about the nature of learning. When an infant plays, waves its arms, or looks about, it has no explicit teacher, but it does have a direct sensorimotor connection to its environment. Exercising this connection produces a wealth of information about cause and effect, about the consequences of actions, and about what to do in order to achieve goals. Throughout our lives, such interactions are undoubtedly a major source of knowledge about our environment and ourselves. Whether we are learning to drive a car or to hold a conversation, we are acutely aware of how our environment responds to what we do, and we seek to influence what happens through our behavior. Learning from interaction is a foundational idea underlying nearly all theories of learning and intelligence. In this book we explore a computational approach to learning from interaction. Rather than directly theorizing about how people or animals learn, we primarily explore idealized learning situations and evaluate the effectiveness of various learning methods. That is, we adopt the perspective of an artificial intelligence researcher or engineer. We explore designs for machines that are effective in solving learning problems of scientific or economic interest, evaluating the designs through mathematical analysis or computational experiments. The approach we explore, called reinforcement learning, is much more focused on goal-directed learning from interaction than are other approaches to machine learning.

Deep reinforcement learning: An overview

## Click to have a preview.

We give an overview of recent exciting achievements of deep reinforcement learning (RL). We discuss six core elements, six important mechanisms, and twelve applications. We start with background of machine learning, deep learning and reinforcement learning. Next we discuss core RL elements, including value function, in particular, Deep Q-Network (DQN), policy, reward, model, planning, and exploration. After that, we discuss important mechanisms for RL, including attention and memory, unsupervised learning, transfer learning, multi-agent RL, hierarchical RL, and learning to learn. Then we discuss various applications of RL, including games, in particular, AlphaGo, robotics, natural language processing, including dialogue systems, machine translation, and text generation, computer vision, neural architecture design, business management, finance, healthcare, Industry 4.0, smart grid, intelligent transportation systems, and computer systems. We mention topics not reviewed yet, and list a collection of RL resources. After presenting a brief summary, we close with discussions. Please see Deep Reinforcement Learning, arXiv:1810.06339, for a significant update.

Reinforcement learning in healthcare: A survey

## Click to have a preview.

As a subfield of machine learning, reinforcement learning (RL) aims at empowering one’s capabilities in behavioural decision making by using interaction experience with the world and an evaluative feedback. Unlike traditional supervised learning methods that usually rely on one-shot, exhaustive and supervised reward signals, RL tackles with sequential decision making problems with sampled, evaluative and delayed feedback simultaneously. Such distinctive features make RL technique a suitable candidate for developing powerful solutions in a variety of healthcare domains, where diagnosing decisions or treatment regimes are usually characterized by a prolonged and sequential procedure. This survey discusses the broad applications of RL techniques in healthcare domains, in order to provide the research community with systematic understanding of theoretical foundations, enabling methods and techniques, existing challenges, and new insights of this emerging paradigm. By first briefly examining theoretical foundations and key techniques in RL research from efficient and representational directions, we then provide an overview of RL applications in healthcare domains ranging from dynamic treatment regimes in chronic diseases and critical care, automated medical diagnosis from both unstructured and structured clinical data, as well as many other control or scheduling domains that have infiltrated many aspects of a healthcare system. Finally, we summarize the challenges and open issues in current research, and point out some potential solutions and directions for future research.

Reinforcement learning based recommender systems: A survey

## Click to have a preview.

Recommender systems (RSs) have become an inseparable part of our everyday lives. They help us find our favorite items to purchase, our friends on social networks, and our favorite movies to watch. Traditionally, the recommendation problem was considered to be a classification or prediction problem, but it is now widely agreed that formulating it as a sequential decision problem can better reflect the user-system interaction. Therefore, it can be formulated as a Markov decision process (MDP) and be solved by reinforcement learning (RL) algorithms. Unlike traditional recommendation methods, including collaborative filtering and content-based filtering, RL is able to handle the sequential, dynamic user-system interaction and to take into account the long-term user engagement. Although the idea of using RL for recommendation is not new and has been around for about two decades, it was not very practical, mainly because of scalability problems of traditional RL algorithms. However, a new trend has emerged in the field since the introduction of deep reinforcement learning (DRL), which made it possible to apply RL to the recommendation problem with large state and action spaces. In this paper, a survey on reinforcement learning based recommender systems (RLRSs) is presented. Our aim is to present an outlook on the field and to provide the reader with a fairly complete knowledge of key concepts of the field. We first recognize and illustrate that RLRSs can be generally classified into RL- and DRL-based methods. Then, we propose an RLRS framework with four components, i.e., state representation, policy optimization, reward formulation, and environment building, and survey RLRS algorithms accordingly. We highlight emerging topics and depict important trends using various graphs and tables. Finally, we discuss important aspects and challenges that can be addressed in the future.

Reinforcement learning for control: Performance, stability, and deep approximators

## Click to have a preview.

Reinforcement learning (RL) offers powerful algorithms to search for optimal controllers of systems with nonlinear, possibly stochastic dynamics that are unknown or highly uncertain. This review mainly covers artificial-intelligence approaches to RL, from the viewpoint of the control engineer. We explain how approximate representations of the solution make RL feasible for problems with continuous states and control actions. Stability is a central concern in control, and we argue that while the control-theoretic RL subfield called adaptive dynamic programming is dedicated to it, stability of RL largely remains an open question. We also cover in detail the case where deep neural networks are used for approximation, leading to the field of deep RL, which has shown great success in recent years. With the control practitioner in mind, we outline opportunities and pitfalls of deep RL; and we close the survey with an outlook that – among other things – points out some avenues for bridging the gap between control and artificial-intelligence RL techniques.

## Click to have a preview.

We provide a natural gradient method that represents the steepest descent direction based on the underlying structure of the parameter space. Although gradient methods cannot make large changes in the values of the parameters, we show that the natural gradient is moving toward choosing a greedy optimal action rather than just a better action. These greedy optimal actions are those that would be chosen under one improvement step of policy iteration with approximate, compatible value functions, as defined by Sutton et al. [9]. We then show drastic performance improvements in simple MDPs and in the more challenging MDP of Tetris.

Trust Region Policy Optimization

## Click to have a preview.

We describe an iterative procedure for optimizing policies, with guaranteed monotonic improvement. By making several approximations to the theoretically-justified procedure, we develop a practical algorithm, called Trust Region Policy Optimization (TRPO). This algorithm is similar to natural policy gradient methods and is effective for optimizing large nonlinear policies such as neural networks. Our experiments demonstrate its robust performance on a wide variety of tasks: learning simulated robotic swimming, hopping, and walking gaits; and playing Atari games using images of the screen as input. Despite its approximations that deviate from the theory, TRPO tends to give monotonic improvement, with little tuning of hyperparameters.

Playing Atari with Deep Reinforcement Learning

## Click to have a preview.

We present the first deep learning model to successfully learn control policies directly from high-dimensional sensory input using reinforcement learning. The model is a convolutional neural network, trained with a variant of Q-learning, whose input is raw pixels and whose output is a value function estimating future rewards. We apply our method to seven Atari 2600 games from the Arcade Learning Environment, with no adjustment of the architecture or learning algorithm. We find that it outperforms all previous approaches on six of the games and surpasses a human expert on three of them.

Proximal Policy Optimization Algorithms

## Click to have a preview.

We propose a new family of policy gradient methods for reinforcement learning, which alternate between sampling data through interaction with the environment, and optimizing a “surrogate” objective function using stochastic gradient ascent. Whereas standard policy gradient methods perform one gradient update per data sample, we propose a novel objective function that enables multiple epochs of minibatch updates. The new methods, which we call proximal policy optimization (PPO), have some of the benefits of trust region policy optimization (TRPO), but they are much simpler to implement, more general, and have better sample complexity (empirically). Our experiments test PPO on a collection of benchmark tasks, including simulated robotic locomotion and Atari game playing, and we show that PPO outperforms other online policy gradient methods, and overall strikes a favorable balance between sample complexity, simplicity, and wall-time.

Discovering faster matrix multiplication algorithms with reinforcement learning

## Click to have a preview.

Improving the efficiency of algorithms for fundamental computations can have a widespread impact, as it can affect the overall speed of a large amount of computations. Matrix multiplication is one such primitive task, occurring in many systems—from neural networks to scientific computing routines. The automatic discovery of algorithms using machine learning offers the prospect of reaching beyond human intuition and outperforming the current best human-designed algorithms. However, automating the algorithm discovery procedure is intricate, as the space of possible algorithms is enormous. Here we report a deep reinforcement learning approach based on AlphaZero for discovering efficient and provably correct algorithms for the multiplication of arbitrary matrices. Our agent, AlphaTensor, is trained to play a single-player game where the objective is finding tensor decompositions within a finite factor space. AlphaTensor discovered algorithms that outperform the state-of-the-art complexity for many matrix sizes. Particularly relevant is the case of 4 × 4 matrices in a finite field, where AlphaTensor’s algorithm improves on Strassen’s two-level algorithm for the first time, to our knowledge, since its discovery 50 years ago2. We further showcase the flexibility of AlphaTensor through different use-cases: algorithms with state-of-the-art complexity for structured matrix multiplication and improved practical efficiency by optimizing matrix multiplication for runtime on specific hardware. Our results highlight AlphaTensor’s ability to accelerate the process of algorithmic discovery on a range of problems, and to optimize for different criteria.

Faster sorting algorithms discovered using deep reinforcement learning

## Click to have a preview.

Fundamental algorithms such as sorting or hashing are used trillions of times on any given day. As demand for computation grows, it has become critical for these algorithms to be as performant as possible. Whereas remarkable progress has been achieved in the past, making further improvements on the efficiency of these routines has proved challenging for both human scientists and computational approaches. Here we show how artificial intelligence can go beyond the current state of the art by discovering hitherto unknown routines. To realize this, we formulated the task of finding a better sorting routine as a single-player game. We then trained a new deep reinforcement learning agent, AlphaDev, to play this game. AlphaDev discovered small sorting algorithms from scratch that outperformed previously known human benchmarks. These algorithms have been integrated into the LLVM standard C++ sort library3. This change to this part of the sort library represents the replacement of a component with an algorithm that has been automatically discovered using reinforcement learning. We also present results in extra domains, showcasing the generality of the approach.

Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videos

## Click to have a preview.

Pretraining on noisy, internet-scale datasets has been heavily studied as a technique for training models with broad, general capabilities for text, images, and other modalities. However, for many sequential decision domains such as robotics, video games, and computer use, publicly available data does not contain the labels required to train behavioral priors in the same way. We extend the internet-scale pretraining paradigm to sequential decision domains through semi-supervised imitation learning wherein agents learn to act by watching online unlabeled videos. Specifically, we show that with a small amount of labeled data we can train an inverse dynamics model accurate enough to label a huge unlabeled source of online data – here, online videos of people playing Minecraft – from which we can then train a general behavioral prior. Despite using the native human interface (mouse and keyboard at 20Hz), we show that this behavioral prior has nontrivial zero-shot capabilities and that it can be fine-tuned, with both imitation learning and reinforcement learning, to hard-exploration tasks that are impossible to learn from scratch via reinforcement learning. For many tasks our models exhibit human-level performance, and we are the first to report computer agents that can craft diamond tools, which can take proficient humans upwards of 20 minutes (24,000 environment actions) of gameplay to accomplish.

## Preference-based Reinforcement Learning

Preference-based Reinforcement Learning (PbRL) seeks to facilitate training RL agents using preference feedback instead of explicit reward signals.

**Recommended Papers List**

A survey of preference-based reinforcement learning methods

## Click to have a preview.

Reinforcement learning (RL) techniques optimize the accumulated long-term reward of a suitably chosen reward function. However, designing such a reward function often requires a lot of task-specific prior knowledge. The designer needs to consider different objectives that do not only influence the learned behavior but also the learning progress. To alleviate these issues, preference-based reinforcement learning algorithms (PbRL) have been proposed that can directly learn from an expert’s preferences instead of a hand-designed numeric reward. PbRL has gained traction in recent years due to its ability to resolve the reward shaping problem, its ability to learn from non numeric rewards and the possibility to reduce the dependence on expert knowledge. We provide a unified framework for PbRL that describes the task formally and points out the different design principles that affect the evaluation task for the human as well as the computational complexity. The design principles include the type of feedback that is assumed, the representation that is learned to capture the preferences, the optimization problem that has to be solved as well as how the exploration/exploitation problem is tackled. Furthermore, we point out shortcomings of current algorithms, propose open research questions and briefly survey practical tasks that have been solved using PbRL.

Deep Reinforcement Learning from Human Preferences

## Click to have a preview.

For sophisticated reinforcement learning (RL) systems to interact usefully with real-world environments, we need to communicate complex goals to these systems. In this work, we explore goals defined in terms of (non-expert) human preferences between pairs of trajectory segments. Our approach separates learning the goal from learning the behavior to achieve it. We show that this approach can effectively solve complex RL tasks without access to the reward function, including Atari games and simulated robot locomotion, while providing feedback on about 0.1% of our agent’s interactions with the environment. This reduces the cost of human oversight far enough that it can be practically applied to state-of-the-art RL systems. To demonstrate the flexibility of our approach, we show that we can successfully train complex novel behaviors with about an hour of human time. These behaviors and environments are considerably more complex than any which have been previously learned from human feedback.

Active preference-based learning of reward functions

## Click to have a preview.

Our goal is to efficiently learn reward functions encoding a human’s preferences for how a dynamical system should act. There are two challenges with this. First, in many problems it is difficult for people to provide demonstrations of the desired system trajectory (like a high-DOF robot arm motion or an aggressive driving maneuver), or to even assign how much numerical reward an action or trajectory should get. We build on work in label ranking and propose to learn from preferences (or comparisons) instead: the person provides the system a relative preference between two trajectories. Second, the learned reward function strongly depends on what environments and trajectories were experienced during the training phase. We thus take an active learning approach, in which the system decides on what preference queries to make. A novel aspect of our work is the complexity and continuous nature of the queries: continuous trajectories of a dynamical system in environments with other moving agents (humans or robots). We contribute a method for actively synthesizing queries that satisfy the dynamics of the system. Further, we learn the reward function from a continuous hypothesis space by maximizing the volume removed from the hypothesis space by each query. We assign weights to the hypothesis space in the form of a log-concave distribution and provide a bound on the number of iterations required to converge. We show that our algorithm converges faster to the desired reward compared to approaches that are not active or that do not synthesize queries in an autonomous driving domain. We then run a user study to put our method to the test with real people.

Preference Transformer: Modeling Human Preferences using Transformers for RL

## Click to have a preview.

Preference-based reinforcement learning (RL) provides a framework to train agents using human preferences between two behaviors. However, preference-based RL has been challenging to scale since it requires a large amount of human feedback to learn a reward function aligned with human intent. In this paper, we present Preference Transformer, a neural architecture that models human preferences using transformers. Unlike prior approaches assuming human judgment is based on the Markovian rewards which contribute to the decision equally, we introduce a new preference model based on the weighted sum of non-Markovian rewards. We then design the proposed preference model using a transformer architecture that stacks causal and bidirectional self-attention layers. We demonstrate that Preference Transformer can solve a variety of control tasks using real human preferences, while prior approaches fail to work. We also show that Preference Transformer can induce a well-specified reward and attend to critical events in the trajectory by automatically capturing the temporal dependencies in human decision-making. Code is available on the project website: this https URL.

Benchmarks and Algorithms for Offline Preference-Based Reward Learning

## Click to have a preview.

Learning a reward function from human preferences is challenging as it typically requires having a high-fidelity simulator or using expensive and potentially unsafe actual physical rollouts in the environment. However, in many tasks the agent might have access to offline data from related tasks in the same target environment. While offline data is increasingly being used to aid policy optimization via offline RL, our observation is that it can be a surprisingly rich source of information for preference learning as well. We propose an approach that uses an offline dataset to craft preference queries via pool-based active learning, learns a distribution over reward functions, and optimizes a corresponding policy via offline RL. Crucially, our proposed approach does not require actual physical rollouts or an accurate simulator for either the reward learning or policy optimization steps. To test our approach, we first evaluate existing offline RL benchmarks for their suitability for offline reward learning. Surprisingly, for many offline RL domains, we find that simply using a trivial reward function results good policy performance, making these domains ill-suited for evaluating learned rewards. To address this, we identify a subset of existing offline RL benchmarks that are well suited for offline reward learning and also propose new offline apprenticeship learning benchmarks which allow for more open-ended behaviors. When evaluated on this curated set of domains, our empirical results suggest that combining offline RL with learned human preferences can enable an agent to learn to perform novel tasks that were not explicitly shown in the offline data.

## Imitation Learning

Imitation Learning, or apprenticeship learning, focuses on emulating human behaviors within specific tasks. Within this framework, the agent learns a mapping between observations and actions and refines its policy by observing demonstrations in a collection of teacher demonstration data.

**Recommended Papers List**

A Survey of Imitation Learning: Algorithms, Recent Developments, and Challenges

## Click to have a preview.

In recent years, the development of robotics and artificial intelligence (AI) systems has been nothing short of remarkable. As these systems continue to evolve, they are being utilized in increasingly complex and unstructured environments, such as autonomous driving, aerial robotics, and natural language processing. As a consequence, programming their behaviors manually or defining their behavior through reward functions (as done in reinforcement learning (RL)) has become exceedingly difficult. This is because such environments require a high degree of flexibility and adaptability, making it challenging to specify an optimal set of rules or reward signals that can account for all possible situations. In such environments, learning from an expert’s behavior through imitation is often more appealing. This is where imitation learning (IL) comes into play - a process where desired behavior is learned by imitating an expert’s behavior, which is provided through demonstrations. This paper aims to provide an introduction to IL and an overview of its underlying assumptions and approaches. It also offers a detailed description of recent advances and emerging areas of research in the field. Additionally, the paper discusses how researchers have addressed common challenges associated with IL and provides potential directions for future research. Overall, the goal of the paper is to provide a comprehensive guide to the growing field of IL in robotics and AI.

Imitation learning: A survey of learning methods

## Click to have a preview.

Imitation learning techniques aim to mimic human behavior in a given task. An agent (a learning machine) is trained to perform a task from demonstrations by learning a mapping between observations and actions. The idea of teaching by imitation has been around for many years; however, the field is gaining attention recently due to advances in computing and sensing as well as rising demand for intelligent applications. The paradigm of learning by imitation is gaining popularity because it facilitates teaching complex tasks with minimal expert knowledge of the tasks. Generic imitation learning methods could potentially reduce the problem of teaching a task to that of providing demonstrations, without the need for explicit programming or designing reward functions specific to the task. Modern sensors are able to collect and transmit high volumes of data rapidly, and processors with high computational power allow fast processing that maps the sensory data to actions in a timely manner. This opens the door for many potential AI applications that require real-time perception and reaction such as humanoid robots, self-driving vehicles, human computer interaction, and computer games, to name a few. However, specialized algorithms are needed to effectively and robustly learn models as learning by imitation poses its own set of challenges. In this article, we survey imitation learning methods and present design options in different steps of the learning process. We introduce a background and motivation for the field as well as highlight challenges specific to the imitation problem. Methods for designing and evaluating imitation learning tasks are categorized and reviewed. Special attention is given to learning methods in robotics and games as these domains are the most popular in the literature and provide a wide array of problems and methodologies. We extensively discuss combining imitation learning approaches using different sources and methods, as well as incorporating other motion learning methods to enhance imitation. We also discuss the potential impact on industry, present major applications, and highlight current and future research directions.

Generative Adversarial Imitation Learning

## Click to have a preview.

Consider learning a policy from example expert behavior, without interaction with the expert or access to reinforcement signal. One approach is to recover the expert’s cost function with inverse reinforcement learning, then extract a policy from that cost function with reinforcement learning. This approach is indirect and can be slow. We propose a new general framework for directly extracting a policy from data, as if it were obtained by reinforcement learning following inverse reinforcement learning. We show that a certain instantiation of our framework draws an analogy between imitation learning and generative adversarial networks, from which we derive a model-free imitation learning algorithm that obtains significant performance gains over existing model-free methods in imitating complex behaviors in large, high-dimensional environments.

Imitation learning by coaching

## Click to have a preview.

Imitation Learning has been shown to be successful in solving many challenging real-world problems. Some recent approaches give strong performance guarantees by training the policy iteratively. However, it is important to note that these guarantees depend on how well the policy we found can imitate the oracle on the training data. When there is a substantial difference between the oracle’s ability and the learner’s policy space, we may fail to find a policy that has low error on the training set. In such cases, we propose to use a coach that demonstrates easy-to-learn actions for the learner and gradually approaches the oracle. By a reduction of learning by demonstration to online learning, we prove that coaching can yield a lower regret bound than using the oracle. We apply our algorithm to a novel cost-sensitive dynamic feature selection problem, a hard decision problem that considers a user-specified accuracy-cost trade-off. Experimental results on UCI datasets show that our method outperforms state-of-the-art imitation learning methods in dynamic features selection and two static feature selection methods.

Apprenticeship learning using linear programming

## Click to have a preview.

In apprenticeship learning, the goal is to learn a policy in a Markov decision process that is at least as good as a policy demonstrated by an expert. The difficulty arises in that the MDP’s true reward function is assumed to be unknown. We show how to frame apprenticeship learning as a linear programming problem, and show that using an off-the-shelf LP solver to solve this problem results in a substantial improvement in running time over existing methods—up to two orders of magnitude faster in our experiments. Additionally, our approach produces stationary policies, while all existing methods for apprenticeship learning output policies that are “mixed”, i.e. randomized combinations of stationary policies. The technique used is general enough to convert any mixed policy to a stationary policy.

## Inverse Reinforcement Learning

Inverse Reinforcement Learning (IRL), unlike the paradigm of behavioral cloning, focuses on deriving a reward function from observed behavior.

**Recommended Papers List**

A survey of inverse reinforcement learning: Challenges, methods and progress

## Click to have a preview.

Inverse reinforcement learning (IRL) is the problem of inferring the reward function of an agent, given its policy or observed behavior. Analogous to RL, IRL is perceived both as a problem and as a class of methods. By categorically surveying the current literature in IRL, this article serves as a reference for researchers and practitioners of machine learning and beyond to understand the challenges of IRL and select the approaches best suited for the problem on hand. The survey formally introduces the IRL problem along with its central challenges such as the difficulty in performing accurate inference and its generalizability, its sensitivity to prior knowledge, and the disproportionate growth in solution complexity with problem size. The article elaborates how the current methods mitigate these challenges. We further discuss the extensions to traditional IRL methods for handling: inaccurate and incomplete perception, an incomplete model, multiple reward functions, and nonlinear reward functions. This survey concludes the discussion with some broad advances in the research area and currently open research questions.

A survey of inverse reinforcement learning

## Click to have a preview.

Learning from demonstration, or imitation learning, is the process of learning to act in an environment from examples provided by a teacher. Inverse reinforcement learning (IRL) is a specific form of learning from demonstration that attempts to estimate the reward function of a Markov decision process from examples provided by the teacher. The reward function is often considered the most succinct description of a task. In simple applications, the reward function may be known or easily derived from properties of the system and hard coded into the learning process. However, in complex applications, this may not be possible, and it may be easier to learn the reward function by observing the actions of the teacher. This paper provides a comprehensive survey of the literature on IRL. This survey outlines the differences between IRL and two similar methods - apprenticeship learning and inverse optimal control. Further, this survey organizes the IRL literature based on the principal method, describes applications of IRL algorithms, and provides areas of future research.

Algorithms for inverse reinforcement learning

## Click to have a preview.

This paper addresses the problem of inverse reinforcement learning (IRL) in Markov decision processes, that is, the problem of extracting a reward function given observed, optimal behavior. IRL may be useful for apprenticeship learning to acquire skilled behavior, and for ascertaining the reward function being optimized by a natural system. We rst characterize the set of all reward functions for which a given policy is optimal. We then derive three algorithms for IRL. The rst two deal with the case where the entire policy is known; we handle tabulated reward functions on a nite state space and linear functional approximation of the reward function over a potentially innite state space. The third algorithm deals with the more realistic case in which the policy is known only through a nite set of observed trajectories. In all cases, a key issue is degeneracy – the existence of a large set of reward functions for which the observed policy is optimal. To remove degeneracy, we suggest some natural heuristics that attempt to pick a reward function that maximally differentiates the observed policy from other, sub-optimal policies. The result is an efficiently solvable linear programming formulation of the IRL problem. We demonstrate our algorithms on simple discrete/finite and continous/infinite state problems.

Apprenticeship Learning via Inverse Reinforcement Learning

## Click to have a preview.

We consider learning in a Markov decision process where we are not explicitly given a reward function, but where instead we can observe an expert demonstrating the task that we want to learn to perform. This setting is useful in applications (such as the task of driving) where it may be difficult to write down an explicit reward function specifying exactly how different desiderata should be traded off. We think of the expert as trying to maximize a reward function that is expressible as a linear combination of known features, and give an algorithm for learning the task demonstrated by the expert. Our algorithm is based on using “inverse reinforcement learning” to try to recover the unknown reward function. We show that our algorithm terminates in a small number of iterations, and that even though we may never recover the expert’s reward function, the policy output by the algorithm will attain performance close to that of the expert, where here performance is measured with respect to the expert’s unknown reward function.

Maximum entropy inverse reinforcement learning

## Click to have a preview.

Recent research has shown the benefit of framing problems of imitation learning as solutions to Markov Decision Problems. This approach reduces learning to the problem of recovering a utility function that makes the behavior induced by a near-optimal policy closely mimic demonstrated behavior. In this work, we develop a probabilistic approach based on the principle of maximum entropy. Our approach provides a well-defined, globally normalized distribution over decision sequences, while providing the same performance guarantees as existing methods. We develop our technique in the context of modeling realworld navigation and driving behaviors where collected data is inherently noisy and imperfect. Our probabilistic approach enables modeling of route preferences as well as a powerful new approach to inferring destinations and routes based on partial trajectories.

Modeling pedestrian-cyclist interactions in shared space using inverse reinforcement learning

## Click to have a preview.

The objective of this study is to model the microscopic behaviour of mixed traffic (cyclist-pedestrian) interactions in non-motorized shared spaces. Video data were collected at two locations of Robson Square non-motorized shared space in downtown Vancouver, British Columbia. Trajectories of cyclists and pedestrians involved in interactions were extracted using computer vision algorithms. The extracted trajectories were used to obtain several variables that describe elements of road users’ behaviour including longitudinal and lateral distances, speed and speed differences, interaction angle, and cyclist acceleration and yaw rate. The road users behaviour was modeled as utility-based intelligent rational agents using the finite-state Markov Decision Process (MDP) framework with unknown reward functions. The study implemented Inverse Reinforcement Learning (IRL) using two algorithms: the Maximum Entropy (ME) algorithm, and the Feature Matching (FM) algorithm to recover/estimate the reward function weights of cyclists in two types of interactions with pedestrians: following and overtaking interactions. Reward function weights infer cyclist preferences during their interactions with pedestrians in non-motorized shared spaces, and can form the key component in developing agent based microsimulation model for road users. Furthermore, the estimated reward functions were used to estimate cyclists’ optimal policy for such interactions. A simulation platform was developed using the estimated reward functions and the cyclist optimal policies to simulate cyclist trajectories for the validation dataset. Results show that the Maximum Entropy (ME) IRL algorithm outperformed the Feature Matching (FM) IRL algorithm, and generally provided reasonable results for modeling such interactions in non-motorized shared spaces, considering the high degrees of freedom in movement and the more-complex road users interactions in such facilities. This research is considered an important step toward developing a full Agent-Based Model (ABM) for road users in shared space facilities to evaluate the safety and efficiency of such facilities.

Bayesian Inverse Reinforcement Learning

## Click to have a preview.

Inverse Reinforcement Learning (IRL) is the problem of learning the reward function underlying a Markov Decision Process given the dynamics of the system and the behaviour of an expert. IRL is motivated by situations where knowledge of the rewards is a goal by itself (as in preference elicitation) and by the task of apprenticeship learning (learning policies from an expert). In this paper we show how to combine prior knowledge and evidence from the expert’s actions to derive a probability distribution over the space of reward functions. We present efficient algorithms that find solutions for the reward learning and apprenticeship learning tasks that generalize well over these distributions. Experimental results show strong improvement for our methods over previous heuristic-based approaches.

Generative adversarial imitation learning

## Click to have a preview.

Consider learning a policy from example expert behavior, without interaction with the expert or access to reinforcement signal. One approach is to recover the expert’s cost function with inverse reinforcement learning, then extract a policy from that cost function with reinforcement learning. This approach is indirect and can be slow. We propose a new general framework for directly extracting a policy from data, as if it were obtained by reinforcement learning following inverse reinforcement learning. We show that a certain instantiation of our framework draws an analogy between imitation learning and generative adversarial networks, from which we derive a model-free imitation learning algorithm that obtains significant performance gains over existing model-free methods in imitating complex behaviors in large, high-dimensional environments.

Variational inverse control with events: A general framework for data-driven reward definition

## Click to have a preview.

The design of a reward function often poses a major practical challenge to real-world applications of reinforcement learning. Approaches such as inverse reinforcement learning attempt to overcome this challenge, but require expert demonstrations, which can be difficult or expensive to obtain in practice. We propose variational inverse control with events (VICE), which generalizes inverse reinforcement learning methods to cases where full demonstrations are not needed, such as when only samples of desired goal states are available. Our method is grounded in an alternative perspective on control and reinforcement learning, where an agent’s goal is to maximize the probability that one or more events will happen at some point in the future, rather than maximizing cumulative rewards. We demonstrate the effectiveness of our methods on continuous control tasks, with a focus on high-dimensional observations like images where rewards are hard or even impossible to specify.

## Reinforcement Learning from Human Feedback

Reinforcement Learning from Human Feedback (RLHF) uses reward modeling to replace handcrafted reward function, facilitating the refinement of policies to align with human intentions.

**Recommended Papers List**

Thoughts on the impact of RLHF research

## Click to have a preview.

In this post I’m going to describe my basic justification for working on RLHF in 2017-2020, which I still stand behind. I’ll discuss various arguments that RLHF research had an overall negative impact and explain why I don’t find them persuasive.

I’ll also clarify that I don’t think research on RLHF is automatically net positive; alignment research should address real alignment problems, and we should reject a vague association between “RLHF progress” and “alignment progress.”

Deep reinforcement learning from human preferences

## Click to have a preview.

For sophisticated reinforcement learning (RL) systems to interact usefully with real-world environments, we need to communicate complex goals to these systems. In this work, we explore goals defined in terms of (non-expert) human preferences between pairs of trajectory segments. Our approach separates learning the goal from learning the behavior to achieve it. We show that this approach can effectively solve complex RL tasks without access to the reward function, including Atari games and simulated robot locomotion, while providing feedback on about 0.1% of our agent’s interactions with the environment. This reduces the cost of human oversight far enough that it can be practically applied to state-of-the-art RL systems. To demonstrate the flexibility of our approach, we show that we can successfully train complex novel behaviors with about an hour of human time. These behaviors and environments are considerably more complex than any which have been previously learned from human feedback.

Fine-tuning language models from human preferences)

## Click to have a preview.

Reward learning enables the application of reinforcement learning (RL) to tasks where reward is defined by human judgment, building a model of reward by asking humans questions. Most work on reward learning has used simulated environments, but complex information about values is often expressed in natural language, and we believe reward learning for language is a key to making RL practical and safe for real-world tasks. In this paper, we build on advances in generative pretraining of language models to apply reward learning to four natural language tasks: continuing text with positive sentiment or physically descriptive language, and summarization tasks on the TL;DR and CNN/Daily Mail datasets. For stylistic continuation we achieve good results with only 5,000 comparisons evaluated by humans. For summarization, models trained with 60,000 comparisons copy whole sentences from the input but skip irrelevant preamble; this leads to reasonable ROUGE scores and very good performance according to our human labelers, but may be exploiting the fact that labelers rely on simple heuristics.

## Click to have a preview.

We report the development of GPT-4, a large-scale, multimodal model which can accept image and text inputs and produce text outputs. While less capable than humans in many real-world scenarios, GPT-4 exhibits human-level performance on various professional and academic benchmarks, including passing a simulated bar exam with a score around the top 10% of test takers. GPT-4 is a Transformer-based model pre-trained to predict the next token in a document. The post-training alignment process results in improved performance on measures of factuality and adherence to desired behavior. A core component of this project was developing infrastructure and optimization methods that behave predictably across a wide range of scales. This allowed us to accurately predict some aspects of GPT-4’s performance based on models trained with no more than 1/1,000th the compute of GPT-4.

## Click to have a preview.

GPT-4 with vision (GPT-4V) enables users to instruct GPT-4 to analyze image inputs provided by the user, and is the latest capability we are making broadly available. Incorporating additional modalities (such as image inputs) into large language models (LLMs) is viewed by some as a key frontier in artificial intelligence research and development. Multimodal LLMs offer the possibility of expanding the impact of language-only systems with novel interfaces and capabilities, enabling them to solve new tasks and provide novel experiences for their users. In this system card, we analyze the safety properties of GPT-4V. Our work on safety for GPT-4V builds on the work done for GPT-4 and here we dive deeper into the evaluations, preparation, and mitigation work done specifically for image inputs.

## Click to have a preview.

For real-world applications, virtual agents must be able to learn new behaviors from non-technical users. Positive and negative feedback are an intuitive way to train new behaviors, and existing work has presented algorithms for learning from such feedback. That work, however, treats feedback as numeric reward to be maximized, and assumes that all trainers provide feedback in the same way. In this work, we show that users can provide feedback in many different ways, which we describe as “training strategies.” Specifically, users may not always give explicit feedback in response to an action, and may be more likely to provide explicit reward than explicit punishment, or vice versa, such that the lack of feedback itself conveys information about the behavior. We present a probabilistic model of trainer feedback that describes how a trainer chooses to provide explicit reward and/or explicit punishment and, based on this model, develop two novel learning algorithms (SABL and I-SABL) which take trainer strategy into account, and can therefore learn from cases where no feedback is provided. Through online user studies we demonstrate that these algorithms can learn with less feedback than algorithms based on a numerical interpretation of feedback. Furthermore, we conduct an empirical analysis of the training strategies employed by users, and of factors that can affect their choice of strategy.

Learning to summarize with human feedback

## Click to have a preview.

As language models become more powerful, training and evaluation are increasingly bottlenecked by the data and metrics used for a particular task. For example, summarization models are often trained to predict human reference summaries and evaluated using ROUGE, but both of these metrics are rough proxies for what we really care about—summary quality. In this work, we show that it is possible to significantly improve summary quality by training a model to optimize for human preferences. We collect a large, high-quality dataset of human comparisons between summaries, train a model to predict the human-preferred summary, and use that model as a reward function to fine-tune a summarization policy using reinforcement learning. We apply our method to a version of the TL; DR dataset of Reddit posts and find that our models significantly outperform both human reference summaries and much larger models fine-tuned with supervised learning alone. Our models also transfer to CNN/DM news articles, producing summaries nearly as good as the human reference without any news-specific fine-tuning. We conduct extensive analyses to understand our human feedback dataset and fine-tuned models. We establish that our reward model generalizes to new datasets, and that optimizing our reward model results in better summaries than optimizing ROUGE according to humans. We hope the evidence from our paper motivates machine learning researchers to pay closer attention to how their training loss affects the model behavior they actually want.

Llama 2: Open foundation and fine-tuned chat models

## Click to have a preview.

In this work, we develop and release Llama 2, a collection of pretrained and fine-tuned large language models (LLMs) ranging in scale from 7 billion to 70 billion parameters. Our fine-tuned LLMs, called Llama 2-Chat, are optimized for dialogue use cases. Our models outperform open-source chat models on most benchmarks we tested, and based on our human evaluations for helpfulness and safety, may be a suitable substitute for closed-source models. We provide a detailed description of our approach to fine-tuning and safety improvements of Llama 2-Chat in order to enable the community to build on our work and contribute to the responsible development of LLMs.

## Click to have a preview.

We present a study on reinforcement learning (RL) from human bandit feedback for sequence-to-sequence learning, exemplified by the task of bandit neural machine translation (NMT). We investigate the reliability of human bandit feedback, and analyze the influence of reliability on the learnability of a reward estimator, and the effect of the quality of reward estimates on the overall RL task. Our analysis of cardinal (5-point ratings) and ordinal (pairwise preferences) feedback shows that their intra- and inter-annotator -agreement is comparable. Best reliability is obtained for standardized cardinal feedback, and cardinal feedback is also easiest to learn and generalize from. Finally, improvements of over 1 BLEU can be obtained by integrating a regression-based reward estimator trained on cardinal feedback for 800 translations into RL for NMT. This shows that RL is possible even from small amounts of fairly reliable human feedback, pointing to a great potential for applications at larger scale.

Training a helpful and harmless assistant with reinforcement learning from human feedback

## Click to have a preview.

We apply preference modeling and reinforcement learning from human feedback (RLHF) to finetune language models to act as helpful and harmless assistants. We find this alignment training improves performance on almost all NLP evaluations, and is fully compatible with training for specialized skills such as python coding and summarization. We explore an iterated online mode of training, where preference models and RL policies are updated on a weekly cadence with fresh human feedback data, efficiently improving our datasets and models. Finally, we investigate the robustness of RLHF training, and identify a roughly linear relation between the RL reward and the square root of the KL divergence between the policy and its initialization. Alongside our main results, we perform peripheral analyses on calibration, competing objectives, and the use of OOD detection, compare our models with human writers, and provide samples from our models using prompts appearing in recent related work.

Training language models to follow instructions with human feedback

## Click to have a preview.

Making language models bigger does not inherently make them better at following a user’s intent. For example, large language models can generate outputs that are untruthful, toxic, or simply not helpful to the user. In other words, these models are not aligned with their users. In this paper, we show an avenue for aligning language models with user intent on a wide range of tasks by fine-tuning with human feedback. Starting with a set of labeler-written prompts and prompts submitted through the OpenAI API, we collect a dataset of labeler demonstrations of the desired model behavior, which we use to fine-tune GPT-3 using supervised learning. We then collect a dataset of rankings of model outputs, which we use to further fine-tune this supervised model using reinforcement learning from human feedback. We call the resulting models InstructGPT. In human evaluations on our prompt distribution, outputs from the 1.3B parameter InstructGPT model are preferred to outputs from the 175B GPT-3, despite having 100x fewer parameters. Moreover, InstructGPT models show improvements in truthfulness and reductions in toxic output generation while having minimal performance regressions on public NLP datasets. Even though InstructGPT still makes simple mistakes, our results show that fine-tuning with human feedback is a promising direction for aligning language models with human intent.