Search

We present Language-Image Value learning (LIV), a unified objective for vision-language representation and reward learning from action-free videos with text annotations. Exploiting a novel connection between dual reinforcement learning and mutual information contrastive learning, the LIV objective trains a multi-modal representation that implicitly encodes a universal value function for tasks specified as language or image goals. We use LIV to pre-train the first control-centric vision-language representation from large human video datasets such as EpicKitchen. Given only a language or image goal, the pre-trained LIV model can assign dense rewards to each frame in videos of unseen robots or humans attempting that task in unseen environments. Further, when some target domain-specific data is available, the same objective can be used to fine-tune and improve LIV and even other pre-trained representations for robotic control and reward specification in that domain. In our experiments on several simulated and real-world robot environments, LIV models consistently outperform the best prior input state representations for imitation learning, as well as reward specification methods for policy synthesis. Our results validate the advantages of joint vision-language representation and reward learning within the unified, compact LIV framework., Extended version of ICML 2023 camera-ready; Project website: https://penn-pal-lab.github.io/LIV/

Offline reinforcement learning (RL) allows agents to learn effective, return-maximizing policies from a static dataset. Three major paradigms for offline RL are Q-Learning, Imitation Learning, and Sequence Modeling. A key open question is: which paradigm is preferred under what conditions? We study this question empirically by exploring the performance of representative algorithms -- Conservative Q-Learning (CQL), Behavior Cloning (BC), and Decision Transformer (DT) -- across the commonly used D4RL and Robomimic benchmarks. We design targeted experiments to understand their behavior concerning data suboptimality and task complexity. Our key findings are: (1) Sequence Modeling requires more data than Q-Learning to learn competitive policies but is more robust; (2) Sequence Modeling is a substantially better choice than both Q-Learning and Imitation Learning in sparse-reward and low-quality data settings; and (3) Sequence Modeling and Imitation Learning are preferable as task horizon increases, or when data is obtained from human demonstrators. Based on the overall strength of Sequence Modeling, we also investigate architectural choices and scaling trends for DT on Atari and D4RL and make design recommendations. We find that scaling the amount of data for DT by 5x gives a 2.5x average score improvement on Atari.

Social media platforms moderate content for each user by incorporating the outputs of both platform-wide content moderation systems and, in some cases, user-configured personal moderation preferences. However, it is unclear (1) how end users perceive the choices and affordances of different kinds of personal content moderation tools, and (2) how the introduction of personalization impacts user perceptions of platforms' content moderation responsibilities. This paper investigates end users' perspectives on personal content moderation tools by conducting an interview study with a diverse sample of 24 active social media users. We probe interviewees' preferences using simulated personal moderation interfaces, including word filters, sliders for toxicity levels, and boolean toxicity toggles. We also examine the labor involved for users in choosing moderation settings and present users' attitudes about the roles and responsibilities of social media platforms and other stakeholders towards moderation. We discuss how our findings can inform design solutions to improve transparency and controllability in personal content moderation tools., Will appear in ACM CSCW

When groups of people are tasked with making a judgment, the issue of uncertainty often arises. Existing methods to reduce uncertainty typically focus on iteratively improving specificity in the overall task instruction. However, uncertainty can arise from multiple sources, such as ambiguity of the item being judged due to limited context, or disagreements among the participants due to different perspectives and an under-specified task. A one-size-fits-all intervention may be ineffective if it is not targeted to the right source of uncertainty. In this paper we introduce a new workflow, Judgment Sieve, to reduce uncertainty in tasks involving group judgment in a targeted manner. By utilizing measurements that separate different sources of uncertainty during an initial round of judgment elicitation, we can then select a targeted intervention adding context or deliberation to most effectively reduce uncertainty on each item being judged. We test our approach on two tasks: rating word pair similarity and toxicity of online comments, showing that targeted interventions reduced uncertainty for the most uncertain cases. In the top 10% of cases, we saw an ambiguity reduction of 21.4% and 25.7%, and a disagreement reduction of 22.2% and 11.2% for the two tasks respectively. We also found through a simulation that our targeted approach reduced the average uncertainty scores for both sources of uncertainty as opposed to uniform approaches where reductions in average uncertainty from one source came with an increase for the other.

In goal-reaching reinforcement learning (RL), the optimal value function has a particular geometry, called quasimetric structure. This paper introduces Quasimetric Reinforcement Learning (QRL), a new RL method that utilizes quasimetric models to learn optimal value functions. Distinct from prior approaches, the QRL objective is specifically designed for quasimetrics, and provides strong theoretical recovery guarantees. Empirically, we conduct thorough analyses on a discretized MountainCar environment, identifying properties of QRL and its advantages over alternatives. On offline and online goal-reaching benchmarks, QRL also demonstrates improved sample efficiency and performance, across both state-based and image-based observations., Project Page: https://www.tongzhouwang.info/quasimetric_rl/

Scholarly publications are key to the transfer of knowledge from scholars to others. However, research papers are information-dense, and as the volume of the scientific literature grows, the need for new technology to support the reading process grows. In contrast to the process of finding papers, which has been transformed by Internet technology, the experience of reading research papers has changed little in decades. The PDF format for sharing research papers is widely used due to its portability, but it has significant downsides including: static content, poor accessibility for low-vision readers, and difficulty reading on mobile devices. This paper explores the question "Can recent advances in AI and HCI power intelligent, interactive, and accessible reading interfaces -- even for legacy PDFs?" We describe the Semantic Reader Project, a collaborative effort across multiple institutions to explore automatic creation of dynamic reading interfaces for research papers. Through this project, we've developed ten research prototype interfaces and conducted usability studies with more than 300 participants and real-world users showing improved reading experiences for scholars. We've also released a production reading interface for research papers that will incorporate the best features as they mature. We structure this paper around challenges scholars and the public face when reading research papers -- Discovery, Efficiency, Comprehension, Synthesis, and Accessibility -- and present an overview of our progress and remaining open challenges.

Deep learning appearance-based 3D gaze estimation is gaining popularity due to its minimal hardware requirements and being free of constraint. Unreliable and overconfident inferences, however, still limit the adoption of this gaze estimation method. To address the unreliable and overconfident issues, we introduce a confidence-aware model that predicts uncertainties together with gaze angle estimations. We also introduce a novel effectiveness evaluation method based on the causality between eye feature degradation and the rise in inference uncertainty to assess the uncertainty estimation. Our confidence-aware model demonstrates reliable uncertainty estimations while providing angular estimation accuracies on par with the state-of-the-art. Compared with the existing statistical uncertainty-angular-error evaluation metric, the proposed effectiveness evaluation approach can more effectively judge inferred uncertainties' performance at each prediction., 9 pages 12 figures

A brief idea

The goal of reinforcement learning (RL) is to maximize the expected cumulative return. It has been shown that this objective can be represented by an optimization problem of the state-action visitation distribution under linear constraints. The dual problem of this formulation, which we refer to as dual RL, is unconstrained and easier to optimize. We show that several state-of-the-art off-policy deep reinforcement learning (RL) algorithms, under both online and offline, RL and imitation learning (IL) settings, can be viewed as dual RL approaches in a unified framework. This unification provides a common ground to study and identify the components that contribute to the success of these methods and also reveals the common shortcomings across methods with new insights for improvement. Our analysis shows that prior off-policy imitation learning methods are based on an unrealistic coverage assumption and are minimizing a particular f-divergence between the visitation distributions of the learned policy and the expert policy. We propose a new method using a simple modification to the dual RL framework that allows for performant imitation learning with arbitrary off-policy data to obtain near-expert performance, without learning a discriminator. Further, by framing a recent SOTA offline RL method XQL in the dual RL framework, we propose alternative choices to replace the Gumbel regression loss, which achieve improved performance and resolve the training instability issue of XQL. Project code and details can be found at this https://hari-sikchi.github.io/dual-rl., 46 pages. Under review

Emotions, and emotional responses, are fundamental to human growth and development. Considering how feeling both positively and negatively can lead to increased risk-taking behavior, it is important to consider the effects of friendship on emotional arousal. This study uses a novel technique of continuously measured electrodermal activity (EDA) signals, a sympathetic nervous system measure of emotional arousal, in conjunction with self-reported survey data to examine how social friendships impact emotional arousal. Furthermore, the study examines how socializing impacts positive and negative affect in self-reported social situations. Results show that socializing, in particular with groups of friends, results in more positive affect and is associated with increased emotional arousal, when compared to not socializing. The association is particularly salient when respondents also report feeling positive. This work makes a contribution to the emerging literature on how biosignal collection can be a way to distinctly measure the effects of social experiences on physical health.

Goal-conditioned reinforcement learning (GCRL) refers to learning general-purpose skills which aim to reach diverse goals. In particular, offline GCRL only requires purely pre-collected datasets to perform training tasks without additional interactions with the environment. Although offline GCRL has become increasingly prevalent and many previous works have demonstrated its empirical success, the theoretical understanding of efficient offline GCRL algorithms is not well established, especially when the state space is huge and the offline dataset only covers the policy we aim to learn. In this paper, we propose a novel provably efficient algorithm (the sample complexity is $\tilde{O}({\rm poly}(1/\epsilon))$ where $\epsilon$ is the desired suboptimality of the learned policy) with general function approximation. Our algorithm only requires nearly minimal assumptions of the dataset (single-policy concentrability) and the function class (realizability). Moreover, our algorithm consists of two uninterleaved optimization steps, which we refer to as $V$-learning and policy learning, and is computationally stable since it does not involve minimax optimization. To the best of our knowledge, this is the first algorithm with general function approximation and single-policy concentrability that is both statistically efficient and computationally stable., Comment: 26 pages

Robots assist humans in various activities, from daily living public service (e.g., airports and restaurants), and to collaborative manufacturing. However, it is risky to assume that the knowledge and strategies robots learned from one group of people can apply to other groups. The discriminatory performance of robots will undermine their service quality for some people, ignore their service requests, and even offend them. Therefore, it is critically important to mitigate bias in robot decision-making for more fair services. In this paper, we designed a self-reflective mechanism -- Fairness-Sensitive Policy Gradient Reinforcement Learning (FSPGRL), to help robots to self-identify biased behaviors during interactions with humans. FSPGRL identifies bias by examining the abnormal update along particular gradients and updates the policy network to support fair decision-making for robots. To validate FSPGRL's effectiveness, a human-centered service scenario, "A robot is serving people in a restaurant," was designed. A user study was conducted; 24 human subjects participated in generating 1,000 service demonstrations. Four commonly-seen issues "Willingness Issue," "Priority Issue," "Quality Issue," "Risk Issue" were observed from robot behaviors. By using FSPGRL to improve robot decisions, robots were proven to have a self-bias detection capability for a more fair service. We have achieved the suppression of bias and improved the quality during the process of robot learning to realize a relatively fair model.

Open-source Software (OSS) has become a valuable resource in both industry and academia over the last few decades. Despite the innovative structures they develop to support the projects, OSS projects and their communities have complex needs and face risks such as getting abandoned. To manage the internal social dynamics and community evolution, OSS developer communities have started relying on written governance documents that assign roles and responsibilities to different community actors. To facilitate the study of the impact and effectiveness of formal governance documents on OSS projects and communities, we present a longitudinal dataset of 710 GitHub-hosted OSS projects with \path{GOVERNANCE.MD} governance files. This dataset includes all commits made to the repository, all issues and comments created on GitHub, and all revisions made to the governance file. We hope its availability will foster more research interest in studying how OSS communities govern their projects and the impact of governance files on communities., Comment: 5 pages, 1 figure, 1 table, to be published in MSR 2023 Data and Tool Showcase Track

To investigate the well-observed racial disparities in computer vision systems that analyze images of humans, researchers have turned to skin tone as more objective annotation than race metadata for fairness performance evaluations. However, the current state of skin tone annotation procedures is highly varied. For instance, researchers use a range of untested scales and skin tone categories, have unclear annotation procedures, and provide inadequate analyses of uncertainty. In addition, little attention is paid to the positionality of the humans involved in the annotation process--both designers and annotators alike--and the historical and sociological context of skin tone in the United States. Our work is the first to investigate the skin tone annotation process as a sociotechnical project. We surveyed recent skin tone annotation procedures and conducted annotation experiments to examine how subjective understandings of skin tone are embedded in skin tone annotation procedures. Our systematic literature review revealed the uninterrogated association between skin tone and race and the limited effort to analyze annotator uncertainty in current procedures for skin tone annotation in computer vision evaluation. Our experiments demonstrated that design decisions in the annotation procedure such as the order in which the skin tone scale is presented or additional context in the image (i.e., presence of a face) significantly affected the resulting inter-annotator agreement and individual uncertainty of skin tone annotations. We call for greater reflexivity in the design, analysis, and documentation of procedures for evaluation using skin tone., Comment: To appear in FAcct '23

Training multiple agents to coordinate is an important problem with applications in robotics, game theory, economics, and social sciences. However, most existing Multi-Agent Reinforcement Learning (MARL) methods are online and thus impractical for real-world applications in which collecting new interactions is costly or dangerous. While these algorithms should leverage offline data when available, doing so gives rise to the offline coordination problem. Specifically, we identify and formalize the strategy agreement (SA) and the strategy fine-tuning (SFT) challenges, two coordination issues at which current offline MARL algorithms fail. To address this setback, we propose a simple model-based approach that generates synthetic interaction data and enables agents to converge on a strategy while fine-tuning their policies accordingly. Our resulting method, Model-based Offline Multi-Agent Proximal Policy Optimization (MOMA-PPO), outperforms the prevalent learning methods in challenging offline multi-agent MuJoCo tasks even under severe partial observability and with learned world models.

User reporting is an essential component of content moderation on many online platforms -- in particular, on end-to-end encrypted (E2EE) messaging platforms where platform operators cannot proactively inspect message contents. However, users' privacy concerns when considering reporting may impede the effectiveness of this strategy in regulating online harassment. In this paper, we conduct interviews with 16 users of E2EE platforms to understand users' mental models of how reporting works and their resultant privacy concerns and considerations surrounding reporting. We find that users expect platforms to store rich longitudinal reporting datasets, recognizing both their promise for better abuse mitigation and the privacy risk that platforms may exploit or fail to protect them. We also find that users have preconceptions about the respective capabilities and risks of moderators at the platform versus community level -- for instance, users trust platform moderators more to not abuse their power but think community moderators have more time to attend to reports. These considerations, along with perceived effectiveness of reporting and how to provide sufficient evidence while maintaining privacy, shape how users decide whether, to whom, and how much to report. We conclude with design implications for a more privacy-preserving reporting system on E2EE messaging platforms., Comment: accepted to SOUPS 2023

In recent years, industry leaders and researchers have proposed to use technical provenance standards to address visual misinformation spread through digitally altered media. By adding immutable and secure provenance information such as authorship and edit date to media metadata, social media users could potentially better assess the validity of the media they encounter. However, it is unclear how end users would respond to provenance information, or how to best design provenance indicators to be understandable to laypeople. We conducted an online experiment with 595 participants from the US and UK to investigate how provenance information altered users' accuracy perceptions and trust in visual content shared on social media. We found that provenance information often lowered trust and caused users to doubt deceptive media, particularly when it revealed that the media was composited. We additionally tested conditions where the provenance information itself was shown to be incomplete or invalid, and found that these states have a significant impact on participants' accuracy perceptions and trust in media, leading them, in some cases, to disbelieve honest media. Our findings show that provenance, although enlightening, is still not a concept well-understood by users, who confuse media credibility with the orthogonal (albeit related) concept of provenance credibility. We discuss how design choices may contribute to provenance (mis)understanding, and conclude with implications for usable provenance systems, including clearer interfaces and user education.

This study examines social media users' preferences for the use of platform-wide moderation in comparison to user-controlled, personalized moderation tools to regulate three categories of norm-violating content - hate speech, sexually explicit content, and violent content. Via a nationally representative survey of 984 US adults, we explore the influence of third-person effects and support for freedom of expression on this choice. We find that perceived negative effects on others negatively predict while free speech support positively predicts a preference for having personal moderation settings over platform-directed moderation for regulating each speech category. Our findings show that platform governance initiatives need to account for both actual and perceived media effects of norm-violating speech categories to increase user satisfaction. Our analysis suggests that users do not view personal moderation tools as an infringement on others' free speech but as a means to assert greater agency over their social media feeds.

Reinforcement Learning (RL), bolstered by the expressive capabilities of Deep Neural Networks (DNNs) for function approximation, has demonstrated considerable success in numerous applications. However, its practicality in addressing a wide range of real-world scenarios, characterized by diverse and unpredictable dynamics, noisy signals, and large state and action spaces, remains limited. This limitation stems from issues such as poor data efficiency, limited generalization capabilities, a lack of safety guarantees, and the absence of interpretability, among other factors. To overcome these challenges and improve performance across these crucial metrics, one promising avenue is to incorporate additional structural information about the problem into the RL learning process. Various sub-fields of RL have proposed methods for incorporating such inductive biases. We amalgamate these diverse methodologies under a unified framework, shedding light on the role of structure in the learning problem, and classify these methods into distinct patterns of incorporating structure. By leveraging this comprehensive framework, we provide valuable insights into the challenges associated with structured RL and lay the groundwork for a design pattern perspective on RL research. This novel perspective paves the way for future advancements and aids in the development of more effective and efficient RL algorithms that can potentially handle real-world scenarios better.

Learning policies which are robust to changes in the environment are critical for real world deployment of Reinforcement Learning agents. They are also necessary for achieving good generalization across environment shifts. We focus on bisimulation metrics, which provide a powerful means for abstracting task relevant components of the observation and learning a succinct representation space for training the agent using reinforcement learning. In this work, we extend the bisimulation framework to also account for context dependent observation shifts. Specifically, we focus on the simulator based learning setting and use alternate observations to learn a representation space which is invariant to observation shifts using a novel bisimulation based objective. This allows us to deploy the agent to varying observation settings during test time and generalize to unseen scenarios. We further provide novel theoretical bounds for simulator fidelity and performance transfer guarantees for using a learnt policy to unseen shifts. Empirical analysis on the high-dimensional image based control domains demonstrates the efficacy of our method.

How well do reward functions learned with inverse reinforcement learning (IRL) generalize? We illustrate that state-of-the-art IRL algorithms, which maximize a maximum-entropy objective, learn rewards that overfit to the demonstrations. Such rewards struggle to provide meaningful rewards for states not covered by the demonstrations, a major detriment when using the reward to learn policies in new situations. We introduce BC-IRL a new inverse reinforcement learning method that learns reward functions that generalize better when compared to maximum-entropy IRL approaches. In contrast to the MaxEnt framework, which learns to maximize rewards around demonstrations, BC-IRL updates reward parameters such that the policy trained with the new reward matches the expert demonstrations better. We show that BC-IRL learns rewards that generalize better on an illustrative simple task and two continuous robotic control tasks, achieving over twice the success rate of baselines in challenging generalization settings.

Training generalist agents is difficult across several axes, requiring us to deal with high-dimensional inputs (space), long horizons (time), and multiple and new tasks. Recent advances with architectures have allowed for improved scaling along one or two of these dimensions, but are still prohibitive computationally. In this paper, we propose to address all three axes by leveraging Language to Control Diffusion models as a hierarchical planner conditioned on language (LCD). We effectively and efficiently scale diffusion models for planning in extended temporal, state, and task dimensions to tackle long horizon control problems conditioned on natural language instructions. We compare LCD with other state-of-the-art models on the CALVIN language robotics benchmark and find that LCD outperforms other SOTA methods in multi task success rates while dramatically improving computational efficiency with a single task success rate (SR) of 88.7% against the previous best of 82.6%. We show that LCD can successfully leverage the unique strength of diffusion models to produce coherent long range plans while addressing their weakness at generating low-level details and control. We release our code and models at https://github.com/ezhang7423/language-control-diffusion., Project and code available at https://github.com/ezhang7423/language-control-diffusion

While the maximum entropy (MaxEnt) reinforcement learning (RL) framework -- often touted for its exploration and robustness capabilities -- is usually motivated from a probabilistic perspective, the use of deep probabilistic models has not gained much traction in practice due to their inherent complexity. In this work, we propose the adoption of latent variable policies within the MaxEnt framework, which we show can provably approximate any policy distribution, and additionally, naturally emerges under the use of world models with a latent belief state. We discuss why latent variable policies are difficult to train, how naive approaches can fail, then subsequently introduce a series of improvements centered around low-cost marginalization of the latent state, allowing us to make full use of the latent state at minimal additional cost. We instantiate our method under the actor-critic framework, marginalizing both the actor and critic. The resulting algorithm, referred to as Stochastic Marginal Actor-Critic (SMAC), is simple yet effective. We experimentally validate our method on continuous control tasks, showing that effective marginalization can lead to better exploration and more robust training. Our implementation is open sourced at https://github.com/zdhNarsil/Stochastic-Marginal-Actor-Critic., Accepted by ICLR 2023

Reward and representation learning are two long-standing challenges for learning an expanding set of robot manipulation skills from sensory observations. Given the inherent cost and scarcity of in-domain, task-specific robot data, learning from large, diverse, offline human videos has emerged as a promising path towards acquiring a generally useful visual representation for control; however, how these human videos can be used for general-purpose reward learning remains an open question. We introduce $\textbf{V}$alue-$\textbf{I}$mplicit $\textbf{P}$re-training (VIP), a self-supervised pre-trained visual representation capable of generating dense and smooth reward functions for unseen robotic tasks. VIP casts representation learning from human videos as an offline goal-conditioned reinforcement learning problem and derives a self-supervised dual goal-conditioned value-function objective that does not depend on actions, enabling pre-training on unlabeled human videos. Theoretically, VIP can be understood as a novel implicit time contrastive objective that generates a temporally smooth embedding, enabling the value function to be implicitly defined via the embedding distance, which can then be used to construct the reward for any goal-image specified downstream task. Trained on large-scale Ego4D human videos and without any fine-tuning on in-domain, task-specific data, VIP's frozen representation can provide dense visual reward for an extensive set of simulated and $\textbf{real-robot}$ tasks, enabling diverse reward-based visual control methods and significantly outperforming all prior pre-trained representations. Notably, VIP can enable simple, $\textbf{few-shot}$ offline RL on a suite of real-world robot tasks with as few as 20 trajectories., ICLR 2023, Notable-Top-25% (Spotlight). Project website: https://sites.google.com/view/vip-rl

BackgroundPrevious research has revealed links between air pollution exposure and metabolic syndrome in adults; however, these associations are less explored in children.ObjectiveThis study aims to investigate the association between traffic-related air pollutants (TRAP) and biomarkers of metabolic dysregulation, oxidative stress, and lung epithelial damage in children.MethodsWe conducted cross-sectional analyses in a sample of predominantly Latinx, low-income children (n = 218) to examine associations between air pollutants (nitrogen dioxide (NO2), nitrogen oxides (NOx), elemental carbon, polycyclic aromatic hydrocarbons, carbon monoxide (CO), fine particulates (PM2.5)) and biomarkers of metabolic function (high-density lipoprotein (HDL), hemoglobin A1c (HbA1c), oxidative stress (8-isoprostane), and lung epithelial damage (club cell protein 16 (CC16)).ResultsHDL cholesterol showed an inverse association with NO2 and NOx, with the strongest relationship between HDL and 3-month exposure to NO2 (-15.4 mg/dL per IQR increase in 3-month NO2, 95% CI = -27.4, -3.4). 8-isoprostane showed a consistent pattern of increasing values with 1-day and 1-week exposure across all pollutants. Non-significant increases in % HbA1c were found during 1-month time frames and decreasing CC16 in 3-month exposure time frames.ConclusionOur results suggest that TRAP is significantly associated with decreased HDL cholesterol in longer-term time frames and elevated 8-isoprostane in shorter-term time frames. TRAP could have the potential to influence lifelong metabolic patterns, through metabolic effects in childhood.

The ability to separate signal from noise, and reason with clean abstractions, is critical to intelligence. With this ability, humans can efficiently perform real world tasks without considering all possible nuisance factors.How can artificial agents do the same? What kind of information can agents safely discard as noises? In this work, we categorize information out in the wild into four types based on controllability and relation with reward, and formulate useful information as that which is both controllable and reward-relevant. This framework clarifies the kinds information removed by various prior work on representation learning in reinforcement learning (RL), and leads to our proposed approach of learning a Denoised MDP that explicitly factors out certain noise distractors. Extensive experiments on variants of DeepMind Control Suite and RoboDesk demonstrate superior performance of our denoised world model over using raw observations alone, and over prior works, across policy optimization control tasks as well as the non-control task of joint position regression., Project page: https://ssnl.github.io/denoised_mdp/ Code: https://github.com/facebookresearch/denoised_mdp

Online social platforms centered around content creators often allow comments on content, where creators moderate the comments they receive. As creators can face overwhelming numbers of comments, with some of them harassing or hateful, platforms typically provide tools such as word filters for creators to automate aspects of moderation. From needfinding interviews with 19 creators about how they use existing tools, we found that they struggled with writing good filters as well as organizing and revisiting their filters, due to the difficulty of determining what the filters actually catch. To address these issues, we present FilterBuddy, a system that supports creators in authoring new filters or building from existing filter lists, as well as organizing their filters and visualizing what comments are captured over time. We conducted an early-stage evaluation of FilterBuddy with YouTube creators, finding that participants see FilterBuddy not just as a moderation tool, but also a means to organize their comments to better understand their audiences., to be published in CHI Conference on Human Factors in Computing Systems (CHI '22)

Building generalizable goal-conditioned agents from rich observations is a key to reinforcement learning (RL) solving real world problems. Traditionally in goal-conditioned RL, an agent is provided with the exact goal they intend to reach. However, it is often not realistic to know the configuration of the goal before performing a task. A more scalable framework would allow us to provide the agent with an example of an analogous task, and have the agent then infer what the goal should be for its current state. We propose a new form of state abstraction called goal-conditioned bisimulation that captures functional equivariance, allowing for the reuse of skills to achieve new goals. We learn this representation using a metric form of this abstraction, and show its ability to generalize to new goals in simulation manipulation tasks. Further, we prove that this learned representation is sufficient not only for goal conditioned tasks, but is amenable to any downstream task described by a state-only reward function. Videos can be found at https://sites.google.com/view/gc-bisimulation., ICML 2022. 20 Pages, 15 Figures, 4 Tables. Website at https://sites.google.com/view/gc-bisimulation

Recent work has shown that offline reinforcement learning (RL) can be formulated as a sequence modeling problem (Chen et al., 2021; Janner et al., 2021) and solved via approaches similar to large-scale language modeling. However, any practical instantiation of RL also involves an online component, where policies pretrained on passive offline datasets are finetuned via taskspecific interactions with the environment. We propose Online Decision Transformers (ODT), an RL algorithm based on sequence modeling that blends offline pretraining with online finetuning in a unified framework. Our framework uses sequence-level entropy regularizers in conjunction with autoregressive modeling objectives for sample-efficient exploration and finetuning. Empirically, we show that ODT is competitive with the state-of-the-art in absolute performance on the D4RL benchmark but shows much more significant gains during the finetuning procedure., Accepted to ICML 2022 (Long Oral Presentation)

Model development often takes data structure, subject matter considerations, model assumptions, and goodness of fit into consideration. To diagnose issues with any of these factors, it can be helpful to understand regression model estimates at a more granular level. We propose a new method for decomposing point estimates from a regression model via weights placed on data clusters. The weights are informed only by the model specification and data availability and thus can be used to explicitly link the effects of data imbalance and model assumptions to actual model estimates. The weight matrix has been understood in linear models as the hat matrix in the existing literature. We extend it to Bayesian hierarchical regression models that incorporate prior information and complicated dependence structures through the covariance among random effects. We show that the model weights, which we call borrowing factors, generalize shrinkage and information borrowing to all regression models. In contrast, the focus of the hat matrix has been mainly on the diagonal elements indicating the amount of leverage. We also provide metrics that summarize the borrowing factors and are practically useful. We present the theoretical properties of the borrowing factors and associated metrics and demonstrate their usage in two examples. By explicitly quantifying borrowing and shrinkage, researchers can better incorporate domain knowledge and evaluate model performance and the impacts of data properties such as data imbalance or influential points.

Reinforcement learning (RL) agents need to be robust to variations in safety-critical environments. While system identification methods provide a way to infer the variation from online experience, they can fail in settings where fast identification is not possible. Another dominant approach is robust RL which produces a policy that can handle worst-case scenarios, but these methods are generally designed to achieve robustness to a single uncertainty set that must be specified at train time. Towards a more general solution, we formulate the multi-set robustness problem to learn a policy robust to different perturbation sets. We then design an algorithm that enjoys the benefits of both system identification and robust RL: it reduces uncertainty where possible given a few interactions, but can still act robustly with respect to the remaining uncertainty. On a diverse set of control tasks, our approach demonstrates improved worst-case performance on new environments compared to prior methods based on system identification and on robust RL alone., Comment: Project webpage at https://sites.google.com/view/sirsa-public/home

While research continues to investigate and improve the accuracy, fairness, and normative appropriateness of content moderation processes on large social media platforms, even the best process cannot be effective if users reject its authority as illegitimate. We present a survey experiment comparing the perceived institutional legitimacy of four popular content moderation processes. We conducted a within-subjects experiment in which we showed US Facebook users moderation decisions and randomized the description of whether those decisions were made by paid contractors, algorithms, expert panels, or juries of users. Prior work suggests that juries will have the highest perceived legitimacy due to the benefits of judicial independence and democratic representation. However, expert panels had greater perceived legitimacy than algorithms or juries. Moreover, outcome alignment - agreement with the decision - played a larger role than process in determining perceived legitimacy. These results suggest benefits to incorporating expert oversight in content moderation and underscore that any process will face legitimacy challenges derived from disagreement about outcomes., Comment: This paper will appear at CSCW 2022

Traditional approaches to RL have focused on learning decision policies directly from episodic decisions, while slowly and implicitly learning the semantics of compositional representations needed for generalization. While some approaches have been adopted to refine representations via auxiliary self-supervised losses while simultaneously learning decision policies, learning compositional representations from hand-designed and context-independent self-supervised losses (multi-view) still adapts relatively slowly to the real world, which contains many non-IID subspaces requiring rapid distribution shift in both time and spatial attention patterns at varying levels of abstraction. In contrast, supervised language model cascades have shown the flexibility to adapt to many diverse manifolds, and hints of self-learning needed for autonomous task transfer. However, to date, transfer methods for language models like few-shot learning and fine-tuning still require human supervision and transfer learning using self-learning methods has been underexplored. We propose a self-supervised loss policy called contrastive distillation which manifests latent variables with high mutual information with both source and target tasks from weights to tokens. We show how this outperforms common methods of transfer learning and suggests a useful design axis of trading off compute for generalizability for online transfer. Contrastive distillation is improved through sampling from memory and suggests a simple algorithm for more efficiently sampling negative examples for contrastive losses than random sampling.

Making online social communities 'better' is a challenging undertaking, as online communities are extraordinarily varied in their size, topical focus, and governance. As such, what is valued by one community may not be valued by another. However, community values are challenging to measure as they are rarely explicitly stated. In this work, we measure community values through the first large-scale survey of community values, including 2,769 reddit users in 2,151 unique subreddits. Through a combination of survey responses and a quantitative analysis of public reddit data, we characterize how these values vary within and across communities. Amongst other findings, we show that community members disagree about how safe their communities are, that longstanding communities place 30.1% more importance on trustworthiness than newer communities, and that community moderators want their communities to be 56.7% less democratic than non-moderator community members. These findings have important implications, including suggesting that care must be taken to protect vulnerable community members, and that participatory governance strategies may be difficult to implement. Accurate and scalable modeling of community values enables research and governance which is tuned to each community's different values. To this end, we demonstrate that a small number of automatically quantifiable features capture a significant yet limited amount of the variation in values between communities with a ROC AUC of 0.667 on a binary classification task. However, substantial variation remains, and modeling community values remains an important topic for future work. We make our models and data public to inform community design and governance., 12 pages, 8 figures, 4 tables; to appear at ICWSM 2022

Accuracy and generalization of dynamics models is key to the success of model-based reinforcement learning (MBRL). As the complexity of tasks increases, so does the sample inefficiency of learning accurate dynamics models. However, many complex tasks also exhibit sparsity in the dynamics, i.e., actions have only a local effect on the system dynamics. In this paper, we exploit this property with a causal invariance perspective in the single-task setting, introducing a new type of state abstraction called \textit{model-invariance}. Unlike previous forms of state abstractions, a model-invariance state abstraction leverages causal sparsity over state variables. This allows for compositional generalization to unseen states, something that non-factored forms of state abstractions cannot do. We prove that an optimal policy can be learned over this model-invariance state abstraction and show improved generalization in a simple toy domain. Next, we propose a practical method to approximately learn a model-invariant representation for complex domains and validate our approach by showing improved modelling performance over standard maximum likelihood approaches on challenging tasks, such as the MuJoCo-based Humanoid. Finally, within the MBRL setting we show strong performance gains with respect to sample efficiency across a host of other continuous control tasks.

Society is showing signs of strong ideological polarization. When pushed to seek perspectives different from their own, people often reject diverse ideas or find them unfathomable. Work has shown that framing controversial issues using the values of the audience can improve understanding of opposing views. In this paper, we present our work designing systems for addressing ideological division through educating U.S. news consumers to engage using a framework of fundamental human values known as Moral Foundations. We design and implement a series of new features that encourage users to challenge their understanding of opposing views, including annotation of moral frames in news articles, discussion of those frames via inline comments, and recommendations based on relevant moral frames. We describe two versions of features -- the first covering a suite of ways to interact with moral framing in news, and the second tailored towards collaborative annotation and discussion. We conduct a field evaluation of each design iteration with 71 participants in total over a period of 6-8 days, finding evidence suggesting users learned to re-frame their discourse in moral values of the opposing side. Our work provides several design considerations for building systems to engage with moral framing., Accepted to ACM GROUP 2022

The study of zero-shot generalisation (ZSG) in deep Reinforcement Learning (RL) aims to produce RL algorithms whose policies generalise well to novel unseen situations at deployment time, avoiding overfitting to their training environments. Tackling this is vital if we are to deploy reinforcement learning algorithms in real world scenarios, where the environment will be diverse, dynamic and unpredictable. This survey is an overview of this nascent field. We rely on a unifying formalism and terminology for discussing different ZSG problems, building upon previous works. We go on to categorise existing benchmarks for ZSG, as well as current methods for tackling these problems. Finally, we provide a critical discussion of the current state of the field, including recommendations for future work. Among other conclusions, we argue that taking a purely procedural content generation approach to benchmark design is not conducive to progress in ZSG, we suggest fast online adaptation and tackling RL-specific problems as some areas for future work on methods for ZSG, and we recommend building benchmarks in underexplored problem settings such as offline RL ZSG and reward-function variation., Comment: JAIR version. Added formal definitions of ZSPT and related concepts, JAIR formatting, other small rewrites; https://www.jair.org/index.php/jair/article/view/14174

Generalization is a central challenge for the deployment of reinforcement learning (RL) systems in the real world. In this paper, we show that the sequential structure of the RL problem necessitates new approaches to generalization beyond the well-studied techniques used in supervised learning. While supervised learning methods can generalize effectively without explicitly accounting for epistemic uncertainty, we show that, perhaps surprisingly, this is not the case in RL. We show that generalization to unseen test conditions from a limited number of training conditions induces implicit partial observability, effectively turning even fully-observed MDPs into POMDPs. Informed by this observation, we recast the problem of generalization in RL as solving the induced partially observed Markov decision process, which we call the epistemic POMDP. We demonstrate the failure modes of algorithms that do not appropriately handle this partial observability, and suggest a simple ensemble-based technique for approximately solving the partially observed problem. Empirically, we demonstrate that our simple algorithm derived from the epistemic POMDP achieves significant gains in generalization over current methods on the Procgen benchmark suite., Comment: First two authors contributed equally

Model-based reinforcement learning is a compelling framework for data-efficient learning of agents that interact with the world. This family of algorithms has many subcomponents that need to be carefully selected and tuned. As a result the entry-bar for researchers to approach the field and to deploy it in real-world tasks can be daunting. In this paper, we present MBRL-Lib -- a machine learning library for model-based reinforcement learning in continuous state-action spaces based on PyTorch. MBRL-Lib is designed as a platform for both researchers, to easily develop, debug and compare new algorithms, and non-expert user, to lower the entry-bar of deploying state-of-the-art algorithms. MBRL-Lib is open-source at https://github.com/facebookresearch/mbrl-lib.

In reinforcement learning (RL), when defining a Markov Decision Process (MDP), the environment dynamics is implicitly assumed to be stationary. This assumption of stationarity, while simplifying, can be unrealistic in many scenarios. In the continual reinforcement learning scenario, the sequence of tasks is another source of nonstationarity. In this work, we propose to examine this continual reinforcement learning setting through the block contextual MDP (BC-MDP) framework, which enables us to relax the assumption of stationarity. This framework challenges RL algorithms to handle both nonstationarity and rich observation settings and, by additionally leveraging smoothness properties, enables us to study generalization bounds for this setting. Finally, we take inspiration from adaptive control to propose a novel algorithm that addresses the challenges introduced by this more realistic BC-MDP setting, allows for zero-shot adaptation at evaluation time, and achieves strong performance on several nonstationary environments., Comment: 26pages, Under Review

Learning representations for pixel-based control has garnered significant attention recently in reinforcement learning. A wide range of methods have been proposed to enable efficient learning, leading to sample complexities similar to those in the full state setting. However, moving beyond carefully curated pixel data sets (centered crop, appropriate lighting, clear background, etc.) remains challenging. In this paper, we adopt a more difficult setting, incorporating background distractors, as a first step towards addressing this challenge. We present a simple baseline approach that can learn meaningful representations with no metric-based learning, no data augmentations, no world-model learning, and no contrastive learning. We then analyze when and why previously proposed methods are likely to fail or reduce to the same performance as the baseline in this harder setting and why we should think carefully about extending such methods beyond the well curated environments. Our results show that finer categorization of benchmarks on the basis of characteristics like density of reward, planning horizon of the problem, presence of task-irrelevant components, etc., is crucial in evaluating algorithms. Based on these observations, we propose different metrics to consider when evaluating an algorithm on benchmark tasks. We hope such a data-centric view can motivate researchers to rethink representation learning when investigating how to best apply RL to real-world tasks.

Many researchers studying online social communities seek to make such communities better. However, understanding what 'better' means is challenging, due to the divergent opinions of community members, and the multitude of possible community values which often conflict with one another. Community members' own values for their communities are not well understood, and how these values align with one another is an open question. Previous research has mostly focused on specific and comparatively well-defined harms within online communities, such as harassment, rule-breaking, and misinformation. In this work, we ask 39 community members on reddit to describe their values for their communities. We gather 301 responses in members' own words, spanning 125 unique communities, and use iterative categorization to produce a taxonomy of 29 different community values across 9 major categories. We find that members value a broad range of topics ranging from technical features to the diversity of the community, and most frequently prioritize content quality. We identify important understudied topics such as content quality and community size, highlight where values conflict with one another, and call for research into governance methods for communities that protect vulnerable members., Comment: 26 pages, 3 figures