Designing Societally Beneficial Reinforcement Learning Systems


Deep reinforcement learning (DRL) is transitioning from a research field focused on game playing to a technology with real-world applications. Notable examples include DeepMind’s work on controlling a nuclear reactor or on improving Youtube video compression, or Tesla attempting to use a method inspired by MuZero for autonomous vehicle behavior planning. But the exciting potential for real world applications of RL should also come with a healthy dose of caution - for example RL policies are well known to be vulnerable to exploitation, and methods for safe and robust policy development are an active area of research.

At the same time as the emergence of powerful RL systems in the real world, the public and researchers are expressing an increased appetite for fair, aligned, and safe machine learning systems. The focus of these research efforts to date has been to account for shortcomings of datasets or supervised learning practices that can harm individuals. However the unique ability of RL systems to leverage temporal feedback in learning complicates the types of risks and safety concerns that can arise.

This post expands on our recent whitepaper and research paper, where we aim to illustrate the different modalities harms can take when augmented with the temporal axis of RL. To combat these novel societal risks, we also propose a new kind of documentation for dynamic Machine Learning systems which aims to assess and monitor these risks both before and after deployment.


Should I Use Offline RL or Imitation Learning?


Figure 1: Summary of our recommendations for when a practitioner should BC and various imitation learning style methods, and when they should use offline RL approaches.

Offline reinforcement learning allows learning policies from previously collected data, which has profound implications for applying RL in domains where running trial-and-error learning is impractical or dangerous, such as safety-critical settings like autonomous driving or medical treatment planning. In such scenarios, online exploration is simply too risky, but offline RL methods can learn effective policies from logged data collected by humans or heuristically designed controllers. Prior learning-based control methods have also approached learning from existing data as imitation learning: if the data is generally “good enough,” simply copying the behavior in the data can lead to good results, and if it’s not good enough, then filtering or reweighting the data and then copying can work well. Several recent works suggest that this is a viable alternative to modern offline RL methods.

This brings about several questions: when should we use offline RL? Are there fundamental limitations to methods that rely on some form of imitation (BC, conditional BC, filtered BC) that offline RL addresses? While it might be clear that offline RL should enjoy a large advantage over imitation learning when learning from diverse datasets that contain a lot of suboptimal behavior, we will also discuss how even cases that might seem BC-friendly can still allow offline RL to attain significantly better results. Our goal is to help explain when and why you should use each method and provide guidance to practitioners on the benefits of each approach. Figure 1 concisely summarizes our findings and we will discuss each component.


Offline RL Made Easier: No TD Learning, Advantage Reweighting, or Transformers


A demonstration of the RvS policy we learn with just supervised learning and a depth-two MLP. It uses no TD learning, advantage reweighting, or Transformers!

Offline reinforcement learning (RL) is conventionally approached using value-based methods based on temporal difference (TD) learning. However, many recent algorithms reframe RL as a supervised learning problem. These algorithms learn conditional policies by conditioning on goal states (Lynch et al., 2019; Ghosh et al., 2021), reward-to-go (Kumar et al., 2019; Chen et al., 2021), or language descriptions of the task (Lynch and Sermanet, 2021).

We find the simplicity of these methods quite appealing. If supervised learning is enough to solve RL problems, then offline RL could become widely accessible and (relatively) easy to implement. Whereas TD learning must delicately balance an actor policy with an ensemble of critics, these supervised learning methods train just one (conditional) policy, and nothing else!


Accelerating Ukraine Intelligence Analysis with Computer Vision on Synthetic Aperture Radar Imagery


Figure 1: Airmass measurements (clouds) over Ukraine from February 18, 2022 - March 01, 2022 from the SEVIRI instrument. Data accessed via the EUMETSAT Viewer.

Satellite imagery is a critical source of information during the current invasion of Ukraine. Military strategists, journalists, and researchers use this imagery to make decisions, unveil violations of international agreements, and inform the public of the stark realities of war. With Ukraine experiencing a large amount of cloud cover and attacks often occuring during night-time, many forms of satellite imagery are hindered from seeing the ground. Synthetic Aperture Radar (SAR) imagery penetrates cloud cover, but requires special training to interpret. Automating this tedious task would enable real-time insights, but current computer vision methods developed on typical RGB imagery do not properly account for the phenomenology of SAR. This leads to suboptimal performance on this critical modality. Improving the access to and availability of SAR-specific methods, codebases, datasets, and pretrained models will benefit intelligence agencies, researchers, and journalists alike during this critical time for Ukraine.

In this post, we present a baseline method and pretrained models that enable the interchangeable use of RGB and SAR for downstream classification, semantic segmentation, and change detection pipelines.


Unsupervised Skill Discovery with Contrastive Intrinsic Control


Main Image

Unsupervised Reinforcement Learning (RL), where RL agents pre-train with self-supervised rewards, is an emerging paradigm for developing RL agents that are capable of generalization. Recently, we released the Unsupervised RL Benchmark (URLB) which we covered in a previous post. URLB benchmarked many unsupervised RL algorithms across three categories — competence-based, knowledge-based, and data-based algorithms. A surprising finding was that competence-based algorithms significantly underperformed other categories. In this post we will demystify what has been holding back competence-based methods and introduce Contrastive Intrinsic Control (CIC), a new competence-based algorithm that is the first to achieve leading results on URLB.


imodels: leveraging the unreasonable effectiveness of rules


imodels: A python package with cutting-edge techniques for concise, transparent, and accurate predictive modeling. All sklearn-compatible and easy to use.

Recent machine-learning advances have led to increasingly complex predictive models, often at the cost of interpretability. We often need interpretability, particularly in high-stakes applications such as medicine, biology, and political science (see here and here for an overview). Moreover, interpretable models help with all kinds of things, such as identifying errors, leveraging domain knowledge, and speeding up inference.

Despite new advances in formulating/fitting interpretable models, implementations are often difficult to find, use, and compare. imodels (github, paper) fills this gap by providing a simple unified interface and implementation for many state-of-the-art interpretable modeling techniques, particularly rule-based methods.


The Unsupervised Reinforcement Learning Benchmark



The shortcomings of supervised RL

Reinforcement Learning (RL) is a powerful paradigm for solving many problems of interest in AI, such as controlling autonomous vehicles, digital assistants, and resource allocation to name a few. We’ve seen over the last five years that, when provided with an extrinsic reward function, RL agents can master very complex tasks like playing Go, Starcraft, and dextrous robotic manipulation. While large-scale RL agents can achieve stunning results, even the best RL agents today are narrow. Most RL algorithms today can only solve the single task they were trained on and do not exhibit cross-task or cross-domain generalization capabilities.

A side-effect of the narrowness of today’s RL systems is that today’s RL agents are also very data inefficient. If we were to train AlphaGo-like agents on many tasks each agent would likely require billions of training steps because today’s RL agents don’t have the capabilities to reuse prior knowledge to solve new tasks more efficiently. RL as we know it is supervised - agents overfit to a specific extrinsic reward which limits their ability to generalize.


Sequence Modeling Solutions
for Reinforcement Learning Problems


Sequence Modeling Solutions for Reinforcement Learning Problems

Long-horizon predictions of (top) the Trajectory Transformer compared to those of (bottom) a single-step dynamics model.

Modern machine learning success stories often have one thing in common: they use methods that scale gracefully with ever-increasing amounts of data. This is particularly clear from recent advances in sequence modeling, where simply increasing the size of a stable architecture and its training set leads to qualitatively different capabilities.1

Meanwhile, the situation in reinforcement learning has proven more complicated. While it has been possible to apply reinforcement learning algorithms to large-scale problems, generally there has been much more friction in doing so. In this post, we explore whether we can alleviate these difficulties by tackling the reinforcement learning problem with the toolbox of sequence modeling. The end result is a generative model of trajectories that looks like a large language model and a planning algorithm that looks like beam search. Code for the approach can be found here.


Which Mutual Information Representation Learning Objectives are Sufficient for Control?


Processing raw sensory inputs is crucial for applying deep RL algorithms to real-world problems. For example, autonomous vehicles must make decisions about how to drive safely given information flowing from cameras, radar, and microphones about the conditions of the road, traffic signals, and other cars and pedestrians. However, direct “end-to-end” RL that maps sensor data to actions (Figure 1, left) can be very difficult because the inputs are high-dimensional, noisy, and contain redundant information. Instead, the challenge is often broken down into two problems (Figure 1, right): (1) extract a representation of the sensory inputs that retains only the relevant information, and (2) perform RL with these representations of the inputs as the system state.

Figure 1. Representation learning can extract compact representations of states for RL.

A wide variety of algorithms have been proposed to learn lossy state representations in an unsupervised fashion (see this recent tutorial for an overview). Recently, contrastive learning methods have proven effective on RL benchmarks such as Atari and DMControl (Oord et al. 2018, Stooke et al. 2020, Schwarzer et al. 2021), as well as for real-world robotic learning (Zhan et al.). While we could ask which objectives are better in which circumstances, there is an even more basic question at hand: are the representations learned via these methods guaranteed to be sufficient for control? In other words, do they suffice to learn the optimal policy, or might they discard some important information, making it impossible to solve the control problem? For example, in the self-driving car scenario, if the representation discards the state of stoplights, the vehicle would be unable to drive safely. Surprisingly, we find that some widely used objectives are not sufficient, and in fact do discard information that may be needed for downstream tasks.