Designing reinforcement learning (RL) agents is typically a difficult process that requires numerous design iterations. Learning can fail for a multitude of reasons and standard RL methods provide too few tools to provide insight into the exact cause. In this paper, we show how to integrate \textit{value decomposition} into a broad class of actor-critic algorithms and use it to assist in the iterative agent-design process. Value decomposition separates a reward function into distinct components and learns value estimates for each. These value estimates provide insight into an agent's learning and decision-making process and enable new training methods to mitigate common problems. As a demonstration, we introduce SAC-D, a variant of soft actor-critic (SAC) adapted for value decomposition. SAC-D maintains similar performance to SAC, while learning a larger set of value predictions. We also introduce decomposition-based tools that exploit this information, including a new reward \textit{influence} metric, which measures each reward component's effect on agent decision-making. Using these tools, we provide several demonstrations of decomposition's use in identifying and addressing problems in the design of both environments and agents. Value decomposition is broadly applicable and easy to incorporate into existing algorithms and workflows, making it a powerful tool in an RL practitioner's toolbox.

It has been recently demonstrated that Temporal Convolution Networks (TCNs) provide state-of-the-art results in many problem domains where the input data is a time-series. TCNs typically incorporate information from a long history of inputs (the receptive field) into a single output using many convolution layers. Real-time inference using a trained TCN can be challenging on devices with limited compute and memory, especially if the receptive field is large. This paper introduces the RT-TCN algorithm that reuses the output of prior convolution operations to minimize the computational requirements and persistent memory footprint of a TCN during real-time inference. We also show that when a TCN is trained using time slices of the input time-series, it can be executed in realtime continually using RT-TCN. In addition, we provide TCN architecture guidelines that ensure that real-time inference can be performed within memory and computational constraints.

Temporal abstraction is a key requirement for agents making decisions over long time horizons—a fundamental challenge in reinforcement learning. There are many reasons why value estimates at multiple timescales might be useful; recent work has shown that value estimates at different time scales can be the basis for creating more advanced discounting functions and for driving representation learning. Further, predictions at many different timescales serve to broaden an agent's model of its environment. One predictive approach of interest within an online learning setting is general value function (GVFs), which represent models of an agent's world as a collection of predictive questions each defined by a policy, a signal to be predicted, and a prediction timescale. In this paper we present Γ-nets, a method for generalizing value function estimation over timescale, allowing a given GVF to be trained and queried for arbitrary timescales so as to greatly increase the predictive ability and scalability of a GVF-based model. The key to our approach is to use timescale as one of the value estimator's inputs. As a result, the prediction target for any timescale is available at every timestep and we are free to train on any number of timescales. We first provide two demonstrations by 1) predicting a square wave and 2) predicting sensorimotor signals on a robot arm using a linear function approximator. Next, we empirically evaluate Γ-nets in the deep reinforcement learning setting using policy evaluation on a set of Atari video games. Our results show that Γ-nets can be effective for predicting arbitrary timescales, with only a small cost in accuracy as compared to learning estimators for fixed timescales. Γ-nets provide a method for accurately and compactly making predictions at many timescales without requiring a priori knowledge of the task, making it a valuable contribution to ongoing work on model-based planning, representation learning, and lifelong learning algorithms.

This paper investigates the problem of interactively learning behaviors communicated by a human teacher using positive and negative feedback. Much previous work on this problem has made the assumption that people provide feedback for decisions that is dependent on the behavior they are teaching and is independent from the learner’s current policy. We present empirical results that show this assumption to be false—whether human trainers give a positive or negative feedback for a decision is influenced by the learner’s current policy. Based on this insight, we introduce Convergent Actor-Critic by Humans (COACH), an algorithm for learning from policy-dependent feedback that converges to a local optimum. Finally, we demonstrate that COACH can successfully learn multiple behaviors on a physical robot.

Fetching items is an important problem for a social robot. It requires a robot to interpret a person's language and gesture and use these noisy observations to infer what item to deliver. If the robot could ask questions, it would help the robot be faster and more accurate in its task. Existing approaches either do not ask questions, or rely on fixed question-asking policies. To address this problem, we propose a model that makes assumptions about cooperation between agents to perform richer signal extraction from observations. This work defines a mathematical framework for an item-fetching domain that allows a robot to increase the speed and accuracy of its ability to interpret a person's requests by reasoning about its own uncertainty as well as processing implicit information (implicatures). We formalize the item-delivery domain as a Partially Observable Markov Decision Process (POMDP), and approximately solve this POMDP in real time. Our model improves speed and accuracy of fetching tasks by asking relevant clarifying questions only when necessary. To measure our model's improvements, we conducted a real world user study with 16 participants. Our method achieved greater accuracy and a faster interaction time compared to state-of-the-art baselines. Our model is 2.17 seconds faster (25% faster) than a state-of-the-art baseline, while being 2.1% more accurate.

People often learn from others' demonstrations, and classic inverse reinforcement learning (IRL) algorithms have brought us closer to realizing this capacity in machines. In contrast, teaching by demonstration has been less well studied computationally. Here, we develop a novel Bayesian model for teaching by demonstration. Stark differences arise when demonstrators are intentionally teaching a task versus simply performing a task. In two experiments, we show that human participants systematically modify their teaching behavior consistent with the predictions of our model. Further, we show that even standard IRL algorithms benefit when learning from behaviors that are intentionally pedagogical. We conclude by discussing IRL algorithms that can take advantage of intentional pedagogy.

This paper introduces two novel algorithms for learning behaviors from human-provided rewards. The primary novelty of these algorithms is that instead of treating the feedback as a numeric reward signal, they interpret feedback as a form of discrete communication that depends on both the behavior the trainer is trying to teach and the teaching strategy used by the trainer. For example, some human trainers use a lack of feedback to indicate whether actions are correct or incorrect, and interpreting this lack of feedback accurately can significantly improve learning speed. Results from user studies show that humans use a variety of training strategies in practice and both algorithms can learn a contextual bandit task faster than algorithms that treat the feedback as numeric. Simulated trainers are also employed to evaluate the algorithms in both contextual bandit and sequential decision-making tasks with similar results.

In order to be useful in real-world situations, it is critical to allow non-technical users to train robots. Existing work has considered the problem of a robot or virtual agent learning behaviors from evaluative feedback provided by a human trainer. That work, however, has treated feedback as a numeric reward that the agent seeks to maximize, and has assumed that all trainers will provide feedback in the same way when teaching the same behavior. We report the results of a series of user studies that indicate human trainers use a variety of approaches to providing feedback in practice, which we describe as different “training strategies.” For example, 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. If the trainer is consistent in their strategy, then it may be possible to infer knowledge about the desired behavior from cases where no explicit feedback is provided. We discuss a probabilistic model of human-provided feedback that can be used to classify these different training strategies based on when the trainer chooses to provide explicit reward and/or explicit punishment, and when they choose to provide no feedback. Additionally, we investigate how training strategies may change in response to the appearance of the learning agent. Ultimately, based on this work, we argue that learning agents designed to understand and adapt to different users' training strategies will allow more efficient and intuitive learning experiences.

We present Confidence-based Feature Acquisition (CFA), a novel supervised learning method for acquiring missing feature values when there is missing data at both training and test time. Previous work has considered the cases of missing data at training time (e.g., Active Feature Acquisition, AFA [8]), or at test time (e.g., Cost-Sensitive Naive Bayes, CSNB [2]), but not both. At training time, CFA constructs a cascaded ensemble of classifiers, starting with the zero-cost features and adding a single feature for each successive model. For each model, CFA selects a subset of training instances for which the added feature should be acquired. At test time, the set of models is applied sequentially (as a cascade), stopping when a user-supplied confidence threshold is met. We compare CFA to AFA, CSNB, and several other baselines, and find that CFA's accuracy is at least as high as the other methods, while incurring significantly lower feature acquisition costs.

Interactive Visual Clustering (IVC) is a novel method that allows a user to explore relational data sets interactively, in order to produce a clustering that satisfies their objectives. IVC combines spring-embedded graph layout with user interaction and constrained clustering. Experimental results on several synthetic and real-world data sets show that IVC yields better clustering performance than alternative methods.