This panel is inspired by this year ' s SIGCSE readings, especially Whistleblower by Susan Fowler and the Modern Figures podcasts. Broad perspectives from history and current events will drive a conversation about persistence, resistance, empowerment, and maintaining principles. Our panelists will tell stories about individuals and groups who have faced social and financial barriers, persisted, and through resistance changed the world for the better. The goal is to motivate and inspire actions that improve CS education for all.

One of the most common complaints about large language models (LLMs) is their prompt sensitivity – that is, the fact that their ability to perform a task or provide a correct answer to a question can depend unpredictably on the way the question is posed. We investigate this variation by comparing two very different but commonly-used styles of prompting: instruction-based prompts, which describe the task in natural language, and example-based prompts, which provide in-context few-shot demonstration pairs to illustrate the task. We find that, despite large variation in performance as a function of the prompt, the model engages some common underlying mechanisms across different prompts of a task. Specifically, we identify task-specific attention heads whose outputs literally describe the task – which we dub lexical task heads – and show that these heads are shared across prompting styles and trigger subsequent answer production. We further find that behavioral variation between prompts can be explained by the degree to which these heads are activated, and that failures are at least sometimes due to competing task representations that dilute the signal of the target task. Our results together present an increasingly clear picture of how LLMs' internal representations can explain behavior that otherwise seems idiosyncratic to users and developers.

Recent works have explored using language models for planning problems. One approach examines translating natural language descriptions of planning tasks into structured planning languages, such as the planning domain definition language (PDDL). Existing evaluation methods struggle to ensure semantic correctness and rely on simple or unrealistic datasets. To bridge this gap, we introduce Planetarium, a benchmark designed to evaluate language models' ability to generate PDDL code from natural language descriptions of planning tasks. Planetarium features a novel PDDL equivalence algorithm that flexibly evaluates the correctness of generated PDDL, along with a dataset of 145,918 text-to-PDDL pairs across 73 unique state combinations with varying levels of difficulty. Finally, we evaluate several API-access and open-weight language models that reveal this task's complexity. For example, 96.1% of the PDDL problem descriptions generated by GPT-4o are syntactically parseable, 94.4% are solvable, but only 24.8% are semantically correct, highlighting the need for a more rigorous benchmark for this problem.

Reinforcement learning (RL) is a powerful learning technique in robotics, where people can specify rewards that robots learn how to maximize through a process of trial-and-error. Despite the numerous advantages of RL to robot programming, no approaches to our knowledge have sought to enable non-technical users to specify RL programs for robots. In this work, we designed two novel RL-based robot programming paradigms for non-technical users: Full MDP Programming (Full-MDP and Goal-Only MDP Programming (Goal-MDP). To evaluate the efficacy of these two approaches, we ran a between-subjects online user study (N = 409) where participants were asked to program a simulated robot to complete example household tasks (e.g., delivering coffee) using one of our RL programming paradigms or a commonly used baseline: Sequential Programming (Seq), or Trigger-Action Programming (TAP). While users neither performed well nor reported positive experiences with the FullMDP interface, user performance and experience with Goal-MDP was similar to the baselines (Seq and TAP) with significantly shorter programs. These results demonstrate that RL-based paradigms like Goal-MDP are a viable alternative to more traditional approaches and provide a starting point for robot programming interfaces that allow end-users to leverage the myriad benefits of RL for programming robots.

We propose DRAGO, a novel approach for continual model-based reinforcement learning aimed at improving the incremental development of world models across a sequence of tasks that differ in their reward functions but not the state space or dynamics. DRAGO comprises two key components: Synthetic Experience Rehearsal, which leverages generative models to create synthetic experiences from past tasks, allowing the agent to reinforce previously learned dynamics without storing data, and Regaining Memories Through Exploration, which introduces an intrinsic reward mechanism to guide the agent toward revisiting relevant states from prior tasks. Together, these components enable the agent to maintain a comprehensive and continually developing world model, facilitating more effective learning and adaptation across diverse environments. Empirical evaluations demonstrate that DRAGO is able to preserve knowledge across tasks, achieving superior performance in various continual learning scenarios.

Large language models (LLMs) like GPT-4 can convert natural-language descriptions of a task into computer code, making them a promising interface for end-user programming. We undertake a systematic analysis of how people with and without programming experience describe information-processing tasks (IPTs) in natural language, focusing on the characteristics of successful communication. Across two online between-subjects studies, we paired crowdworkers either with one another or with an LLM, asking senders (always humans) to communicate IPTs in natural language to their receiver (either a human or LLM). Both senders and receivers tried to answer test cases, the latter based on their sender’s description. While participants with programming experience tended to communicate IPTs more successfully than non-programmers, this advantage was not overwhelming. Furthermore, a user interface that solicited example test cases from senders often, but not always, improved IPT communication. Allowing receivers to request clarification, though, was less successful at improving communication.

Reinforcement-learning agents seek to maximize a reward signal through environmental interactions. As humans, our job in the learning process is to design reward functions to express desired behavior and enable the agent to learn such behavior swiftly. However, designing good reward functions to induce the desired behavior is generally hard, let alone the question of which rewards make learning fast. In this work, we introduce a family of a reward structures we call Tiered Reward that addresses both of these questions. We consider the reward-design problem in tasks formulated as reaching desirable states and avoiding undesirable states. To start, we propose a strict partial ordering of the policy space to resolve trade-offs in behavior preference. We prefer policies that reach the good states faster and with higher probability while avoiding the bad states longer. Next, we introduce Tiered Reward, a class of environment-independent reward functions and show it is guaranteed to induce policies that are Pareto-optimal according to our preference relation. Finally, we demonstrate that Tiered Reward leads to fast learning with multiple tabular and deep reinforcement-learning algorithms.

We consider the reinforcement learning problem under partial observability, where observations in the decision process lack the Markov property. To cope with partial observability, first we must detect it. We introduce the $\lambda$-discrepancy: a measure of the degree of non-Markovianity of system dynamics. The $\lambda$-discrepancy is the difference between TD($\lambda$) value functions for two different values of $\lambda$; for example, between 1-step temporal difference learning (TD($0$)), which makes an implicit Markov assumption, and Monte Carlo value estimation (TD($1$)), which does not. We prove that this observable and scalable value-based measure is a reliable signal of partial observability. We then use it as an optimization target for resolving partial observability by searching for memory functions---functions over the agent's history---to augment the agent's observations and reduce $\lambda$-discrepancy. We empirically demonstrate that our approach produces memory-augmented observations that resolve partial observability and improve decision making.

Reinforcement learning algorithms typically rely on the assumption that the environment dynamics and value function can be expressed in terms of a Markovian state representation. However, when state information is only partially observable, how can an agent learn such a state representation, and how can it detect when it has found one? We introduce a metric that can accomplish both objectives, without requiring access to---or knowledge of---an underlying, unobservable state space. Our metric, the λ-discrepancy, is the difference between two distinct temporal difference (TD) value estimates, each computed using TD(λ) with a different value of λ. Since TD(λ=0) makes an implicit Markov assumption and TD(λ=1) does not, a discrepancy between these estimates is a potential indicator of a non-Markovian state representation. Indeed, we prove that the λ-discrepancy is exactly zero for all Markov decision processes and almost always non-zero for a broad class of partially observable environments. We also demonstrate empirically that, once detected, minimizing the λ-discrepancy can help with learning a memory function to mitigate the corresponding partial observability. We then train a reinforcement learning agent that simultaneously constructs two recurrent value networks with different λ parameters and minimizes the difference between them as an auxiliary loss. The approach scales to challenging partially observable domains, where the resulting agent frequently performs significantly better (and never performs worse) than a baseline recurrent agent with only a single value network.

Seeking to make machine learning more dependable.

In reinforcement learning, the classic objectives of maximizing discounted and finite-horizon cumulative rewards are PAC-learnable: There are algorithms that learn a near-optimal policy with high probability using a finite amount of samples and computation. In recent years, researchers have introduced objectives and corresponding reinforcement-learning algorithms beyond the classic cumulative rewards, such as objectives specified as linear temporal logic formulas. However, questions about the PAC-learnability of these new objectives have remained open. This work demonstrates the PAC-learnability of general reinforcement-learning objectives through sufficient conditions for PAC-learnability in two analysis settings. In particular, for the analysis that considers only sample complexity, we prove that if an objective given as an oracle is uniformly continuous, then it is PAC-learnable. Further, for the analysis that considers computational complexity, we prove that if an objective is computable, then it is PAC-learnable. In other words, if a procedure computes successive approximations of the objective's value, then the objective is PAC-learnable. We give three applications of our condition on objectives from the literature with previously unknown PAC-learnability and prove that these objectives are PAC-learnable. Overall, our result helps verify existing objectives' PAC-learnability. Also, as some studied objectives that are not uniformly continuous have been shown to be not PAC-learnable, our results could guide the design of new PAC-learnable objectives.

I M A G E B Y E L E N Y S K A conditions; ˲ Communication networks resilient to natural disasters; ˲ Advanced computing infrastructure for efficient storage and aggregation of the data, and high-speed, heterogenous computing resources that can handle enormous volumes of climaterelated data and large complex climate models; ˲ State-of-the-art, data-driven computational modeling and high-precision simulation for enabling deeper T H E N A T I O N A L S C I E N C E Foundation’s (NSF) Directorate for Computer and Information Science and Engineering (CISE) advances understanding of the principles and uses of next-generation computing, communications, and information systems in service to society; supports advanced cyberinfrastructure that enables and accelerates discovery and innovation across all science and engineering disciplines; and contributes to universal, transparent, and affordable participation in an information-based society. CISE provides approximately 80% of the federal funding for fundamental computer science research at U.S. academic institutions. Andrew A. Chien’s Communications editorial “Computing’s Grand Challenge for Sustainability”a (Oct. 2022) lays out a compelling and timely challenge for environmental sustainability: “The computing community should embrace a grand challenge to reduce the carbon-emissions and environmental impact of computing in absolute terms dramatically, and if possible, to zero.” The role of computing and environmental sustainability, as noted in Chien’s editorial, can be viewed in two different ways: “Computing in Sustainability” and “Sustainability in Computing.” This Viewpoint focuses on ways the computing community can contrib-

We propose a novel parameterized skill-learning algorithm that aims to learn transferable parameterized skills and synthesize them into a new action space that supports efficient learning in long-horizon tasks. We propose to leverage off-policy Meta-RL combined with a trajectory-centric smoothness term to learn a set of parameterized skills. Our agent can use these learned skills to construct a three-level hierarchical framework that models a Temporally-extended Parameterized Action Markov Decision Process. We empirically demonstrate that the proposed algorithms enable an agent to solve a set of difficult long-horizon (obstacle-course and robot manipulation) tasks.

Despite the advancement of machine learning techniques in recent years, state-of-the-art systems lack robustness to "real world" events, where the input distributions and tasks encountered by the deployed systems will not be limited to the original training context, and systems will instead need to adapt to novel distributions and tasks while deployed. This critical gap may be addressed through the development of "Lifelong Learning" systems that are capable of 1) Continuous Learning, 2) Transfer and Adaptation, and 3) Scalability. Unfortunately, efforts to improve these capabilities are typically treated as distinct areas of research that are assessed independently, without regard to the impact of each separate capability on other aspects of the system. We instead propose a holistic approach, using a suite of metrics and an evaluation framework to assess Lifelong Learning in a principled way that is agnostic to specific domains or system techniques. Through five case studies, we show that this suite of metrics can inform the development of varied and complex Lifelong Learning systems. We highlight how the proposed suite of metrics quantifies performance trade-offs present during Lifelong Learning system development - both the widely discussed Stability-Plasticity dilemma and the newly proposed relationship between Sample Efficient and Robust Learning. Further, we make recommendations for the formulation and use of metrics to guide the continuing development of Lifelong Learning systems and assess their progress in the future.

Principled decision-making in continuous state--action spaces is impossible without some assumptions. A common approach is to assume Lipschitz continuity of the Q-function. We show that, unfortunately, this property fails to hold in many typical domains. We propose a new coarse-grained smoothness definition that generalizes the notion of Lipschitz continuity, is more widely applicable, and allows us to compute significantly tighter bounds on Q-functions, leading to improved learning. We provide a theoretical analysis of our new smoothness definition, and discuss its implications and impact on control and exploration in continuous domains.

Trigger-action programming (TAP) empowers a wide array of users to automate Internet of Things (IoT) devices. However, it can be challenging for users to create completely correct trigger-action programs (TAPs) on the first try, necessitating debugging. While TAP has received substantial research attention, TAP debugging has not. In this paper, we present the first empirical study of users' end-to-end TAP debugging process, focusing on obstacles users face in debugging TAPs and how well users ultimately fix incorrect automations. To enable this study, we added TAP capabilities to an existing 3-D smart home simulator. Thirty remote participants spent a total of 84 hours debugging TAPs using this simulator. Without additional support, participants were often unable to fix buggy TAPs due to a series of obstacles we document. However, we also found that two novel tools we developed helped participants overcome many of these obstacles and more successfully debug TAPs. These tools collect either implicit or explicit feedback from users about automations that should or should not have happened in the past, using a SAT-solving-based algorithm we developed to automatically modify the TAPs to account for this feedback.

Recent advances in reinforcement-learning research have demonstrated impressive results in building algorithms that can out-perform humans in complex tasks. Nevertheless, creating reinforcement-learning systems that can build abstractions of their experience to accelerate learning in new contexts still remains an active area of research. Previous work showed that reward-predictive state abstractions fulfill this goal, but have only be applied to tabular settings. Here, we provide a clustering algorithm that enables the application of such state abstractions to deep learning settings, providing compressed representations of an agent's inputs that preserve the ability to predict sequences of reward. A convergence theorem and simulations show that the resulting reward-predictive deep network maximally compresses the agent's inputs, significantly speeding up learning in high dimensional visual control tasks. Furthermore, we present different generalization experiments and analyze under which conditions a pre-trained reward-predictive representation network can be re-used without re-training to accelerate learning – a form of systematic out-of-distribution transfer.