Abstract:To effectively study complex causal systems, it is often useful to construct representations that simplify parts of the system by discarding irrelevant details while preserving key features. The Information Bottleneck (IB) method is a widely used approach in representation learning that compresses random variables while retaining information about a target variable. Traditional methods like IB are purely statistical and ignore underlying causal structures, making them ill-suited for causal tasks. We propose the Causal Information Bottleneck (CIB), a causal extension of the IB, which compresses a set of chosen variables while maintaining causal control over a target variable. This method produces representations which are causally interpretable, and which can be used when reasoning about interventions. We present experimental results demonstrating that the learned representations accurately capture causality as intended.
Abstract:Reward machines allow the definition of rewards for temporally extended tasks and behaviors. Specifying "informative" reward machines can be challenging. One way to address this is to generate reward machines from a high-level abstract description of the learning environment, using techniques such as AI planning. However, previous planning-based approaches generate a reward machine based on a single (sequential or partial-order) plan, and do not allow maximum flexibility to the learning agent. In this paper we propose a new approach to synthesising reward machines which is based on the set of partial order plans for a goal. We prove that learning using such "maximally permissive" reward machines results in higher rewards than learning using RMs based on a single plan. We present experimental results which support our theoretical claims by showing that our approach obtains higher rewards than the single-plan approach in practice.
Abstract:Automated testing of computer games is a challenging problem, especially when lengthy scenarios have to be tested. Automating such a scenario boils down to finding the right sequence of interactions given an abstract description of the scenario. Recent works have shown that an agent-based approach works well for the purpose, e.g. due to agents' reactivity, hence enabling a test agent to immediately react to game events and changing state. Many games nowadays are multi-player. This opens up an interesting possibility to deploy multiple cooperative test agents to test such a game, for example to speed up the execution of multiple testing tasks. This paper offers a cooperative multi-agent testing approach and a study of its performance based on a case study on a 3D game called Lab Recruits.
Abstract:Recent developments enable the quantification of causal control given a structural causal model (SCM). This has been accomplished by introducing quantities which encode changes in the entropy of one variable when intervening on another. These measures, named causal entropy and causal information gain, aim to address limitations in existing information theoretical approaches for machine learning tasks where causality plays a crucial role. They have not yet been properly mathematically studied. Our research contributes to the formal understanding of the notions of causal entropy and causal information gain by establishing and analyzing fundamental properties of these concepts, including bounds and chain rules. Furthermore, we elucidate the relationship between causal entropy and stochastic interventions. We also propose definitions for causal conditional entropy and causal conditional information gain. Overall, this exploration paves the way for enhancing causal machine learning tasks through the study of recently-proposed information theoretic quantities grounded in considerations about causality.
Abstract:Improving sample efficiency is central to Reinforcement Learning (RL), especially in environments where the rewards are sparse. Some recent approaches have proposed to specify reward functions as manually designed or learned reward structures whose integrations in the RL algorithms are claimed to significantly improve the learning efficiency. Manually designed reward structures can suffer from inaccuracy and existing automatically learning methods are often computationally intractable for complex tasks. The integration of inaccurate or partial reward structures in RL algorithms fail to learn optimal policies. In this work, we propose an RL algorithm that can automatically structure the reward function for sample efficiency, given a set of labels that signify subtasks. Given such minimal knowledge about the task, we train a high-level policy that selects optimal sub-tasks in each state together with a low-level policy that efficiently learns to complete each sub-task. We evaluate our algorithm in a variety of sparse-reward environments. The experiment results show that our approach significantly outperforms the state-of-art baselines as the difficulty of the task increases.
Abstract:Artificial intelligence models and methods commonly lack causal interpretability. Despite the advancements in interpretable machine learning (IML) methods, they frequently assign importance to features which lack causal influence on the outcome variable. Selecting causally relevant features among those identified as relevant by these methods, or even before model training, would offer a solution. Feature selection methods utilizing information theoretical quantities have been successful in identifying statistically relevant features. However, the information theoretical quantities they are based on do not incorporate causality, rendering them unsuitable for such scenarios. To address this challenge, this article proposes information theoretical quantities that incorporate the causal structure of the system, which can be used to evaluate causal importance of features for some given outcome variable. Specifically, we introduce causal versions of entropy and mutual information, termed causal entropy and causal information gain, which are designed to assess how much control a feature provides over the outcome variable. These newly defined quantities capture changes in the entropy of a variable resulting from interventions on other variables. Fundamental results connecting these quantities to the existence of causal effects are derived. The use of causal information gain in feature selection is demonstrated, highlighting its superiority over standard mutual information in revealing which features provide control over a chosen outcome variable. Our investigation paves the way for the development of methods with improved interpretability in domains involving causation.
Abstract:The increasing applications of AI systems require personalized explanations for their behaviors to various stakeholders since the stakeholders may have various knowledge and backgrounds. In general, a conversation between explainers and explainees not only allows explainers to obtain the explainees' background, but also allows explainees to better understand the explanations. In this paper, we propose an approach for an explainer to communicate personalized explanations to an explainee through having consecutive conversations with the explainee. We prove that the conversation terminates due to the explainee's justification of the initial claim as long as there exists an explanation for the initial claim that the explainee understands and the explainer is aware of.
Abstract:Training a dialogue policy using deep reinforcement learning requires a lot of exploration of the environment. The amount of wasted invalid exploration makes their learning inefficient. In this paper, we find and define an important reason for the invalid exploration: dead-ends. When a conversation enters a dead-end state, regardless of the actions taken afterward, it will continue in a dead-end trajectory until the agent reaches a termination state or maximum turn. We propose a dead-end resurrection (DDR) algorithm that detects the initial dead-end state in a timely and efficient manner and provides a rescue action to guide and correct the exploration direction. To prevent dialogue policies from repeatedly making the same mistake, DDR also performs dialogue data augmentation by adding relevant experiences containing dead-end states. We first validate the dead-end detection reliability and then demonstrate the effectiveness and generality of the method by reporting experimental results on several dialogue datasets from different domains.
Abstract:Communication is an effective mechanism for coordinating the behavior of multiple agents. In the field of multi-agent reinforcement learning, agents can improve the overall learning performance and achieve their objectives by communication. Moreover, agents can communicate various types of messages, either to all agents or to specific agent groups, and through specific channels. With the growing body of research work in MARL with communication (Comm-MARL), there is lack of a systematic and structural approach to distinguish and classify existing Comm-MARL systems. In this paper, we survey recent works in the Comm-MARL field and consider various aspects of communication that can play a role in the design and development of multi-agent reinforcement learning systems. With these aspects in mind, we propose several dimensions along which Comm-MARL systems can be analyzed, developed, and compared.
Abstract:Norms have been widely proposed as a way of coordinating and controlling the activities of agents in a multi-agent system (MAS). A norm specifies the behaviour an agent should follow in order to achieve the objective of the MAS. However, designing norms to achieve a particular system objective can be difficult, particularly when there is no direct link between the language in which the system objective is stated and the language in which the norms can be expressed. In this paper, we consider the problem of synthesising a norm from traces of agent behaviour, where each trace is labelled with whether the behaviour satisfies the system objective. We show that the norm synthesis problem is NP-complete.