Emotion recognition in conversations is increasingly tackled with language models, but these models can be unstable and expensive to fine-tune or to prompt with long dialogue histories. We propose EmoLASP, a framework that combines a language model with declarative reasoning via Answer Set Programming (ASP) to predict VAD scores (Valence-Arousal-Dominance) in conversations. Experiments on a widely used benchmark dataset (IEMOCAP) across six open-source LLMs (3B-120B) and two PLMs (BERT, RoBERTa) show that EmoLASP improves prediction performance compared to using the language model alone, even when the LLMs/PLMs are given no dialogue history in their prompts or input vectors. The gains are largest for prompt-only LLMs, which EmoLASP uses without any fine-tuning. However, for fine-tuned PLMs, the reasoner adds little once dialogue history is available. EmoLASP's LLM pipeline demonstrates the potential advantages of using a reasoning approach to ensure emotion prediction consistency and to reduce both the cost of fine-tuning and the cost of prompting with long dialogue histories.

We investigate the belief revision problem in epistemic planning, i.e., what will be the beliefs of all agents in a multi-agent system after an agent gains the belief in some state property. Based on the standard representation in epistemic planning of agents' beliefs via a single multi-agent Kripke model, we generalize the classical AGM belief revision postulates to the multi-agent setting, with the aim to provide a formal framework for evaluating dynamic epistemic reasoning frameworks in which the beliefs of all agents as the result of actions are computed. As an example of a simple operator that satisfies all of the generalized AGM postulates, we present generalized full-meet multi-agent belief revision. We moreover define a generalization of the standard postulates for iterated revision, present a more sophisticated, event model based revision operator, and discuss the potential issues in defining an epistemic operator on Kripke models that can satisfy all of the generalized postulates for iterated multi-agent belief revision.

The Game Description Language (GDL) is a widely used formalism for specifying general games. Due to their similar syntax and semantics, Answer Set Programming (ASP) and its extensions have been applied to single- and two-player deterministic turn-taking GDL games. This paper presents the first ASP-based approach for solving two-player turn-taking GDL games with uncertainty. We introduce Stochastic An- swer Set Programming (SQASP) to encode the maximally achievable winning probability for a given player in stochastic GDL games, and develop a translation-based solver that evaluates SQASP programs by converting them to Extended Stochastic Satisfiability. Our empirical results show that the proposed approach is competitive with forward search on small stochastic games and can potentially support general game players in endgame evaluation.

Expert-based approaches to Continual Learning employ a dynamically growing set of experts where each is trained on a separate task. During test time, the most relevant expert is selected and used. The Hierarchically Gated Experts (HGE) algorithm improves the efficiency of the expert selection process by organising the experts into a tree. But whilst the efficiency is improved, the expert selection accuracy drops due to the Masking phenomenon, where an expert low on the hierarchy is hidden by another closer to the root of the tree. We introduce a Masking measure which we show can predict the effectiveness of hierarchical organisation and explain the structure of the generated trees. Further, we explore an extension to the HGE algorithm where multiple autoencoders are employed to improve the expert selection accuracy. Whilst multiple autoencoders are superior to just one, training them under the Continual Learning constraint remains a challenge.

We investigate the belief revision problem in epistemic planning, i.e., what will be the beliefs of all agents in a multi-agent system after an agent gains the belief in some state property. Based on the standard representation in epistemic planning of agents' beliefs via a single multi-agent Kripke model, we generalize the classical AGM belief revision postulates to the multi-agent setting, with the aim to provide a formal framework for evaluating dynamic epistemic reasoning frameworks in which the beliefs of all agents as the result of actions are computed. As an example of a simple operator that satisfies all of the generalized AGM postulates, we present generalized full-meet multi-agent belief revision. We moreover define a generalization of the standard postulates for iterated revision, present a more sophisticated, event model based revision operator, and discuss the potential issues in defining an epistemic operator on Kripke models that can satisfy all of the generalized postulates for iterated multi-agent belief revision.

The Game Description Language (GDL) is a widely used formalism for specifying the rules of general games. Writing correct GDL descriptions can be challenging, especially for non-experts. Automated theorem proving has been proposed to assist game design by verifying if a GDL description satisfies desirable logical properties. However, when a description is proved to be faulty, the repair task itself can only be done manually. Motivated by the work on repairing unsolvable planning domain descriptions, we define a more general problem of finding minimal repairs for GDL descriptions that violate formal requirements, and we provide complexity results for various computational problems related to minimal repair. Moreover, we present an Answer Set Programming-based encoding for solving the minimal repair problem and demonstrate its application for automatically repairing ill-defined game descriptions.

The Game Description Language (GDL) is a widely used formalism for specifying the rules of general games. Writing correct GDL descriptions can be challenging, especially for non-experts. Automated theorem proving has been proposed to assist game design by verifying if a GDL description satisfies desirable logical properties. However, when a description is proved to be faulty, the repair task itself can only be done manually. Motivated by the work on repairing unsolvable planning domain descriptions, we define a more general problem of finding minimal repairs for GDL descriptions that violate formal requirements, and we provide complexity results for various computational problems related to minimal repair. Moreover, we present an Answer Set Programming-based encoding for solving the minimal repair problem and demonstrate its application for automatically repairing ill-defined game descriptions.

Hybrid models combining Monte-Carlo Tree Search (MCTS) with fixed depth minimax searches have shown great success as the brute force search allow the model to navigate highly tactical domains. However, minimax is computationally expensive and unnecessary in positions that do not require precise calculations. Ideally, we can adjust the depth to efficiently rely on minimax only when needed. In this paper, we build up the motivation for augmenting MCTS with dynamic depth minimax searches. We analyse the nature of different domains to create some simple dynamic depth adjustment functions which we then benchmark to reinforce our hypothesis that dynamic adjustments of the search depth in MCTS-Minimax hybrids result in stronger play. For this paper we assume that heuristics or evaluator functions are not available to the player, e.g. as in the context of General Game Playing.

The Game Description Language with Imperfect Information (GDL-II) is a lightweight formalism for representing the rules of arbitrary games, including those where players have private information. Its purpose is to build general game-playing systems, that is, automated players that can understand the rules of games and learn how to play them without human intervention. Epistemic Strategy Logic (SLK), on the other hand, is a rich logical framework for reasoning about multi-agent systems and the strategic behavior of agents with partial observability. To enable a general game-playing system to take advantage of this rich formalism for the automatic verification of properties of games, we present a formal translation from GDL-II to SLK models. We prove the correctness of this translation and show how crucial properties of general games, including playability and the existence of Nash equilibria, can be expressed as formulas in SLK. Finally, we demonstrate the application of an existing model-checking system for SLK to verify the properties of GDL-II games.

Game solving is a relatively less explored area in general game playing. This paper introduces a translation from the Game Description Language GDL to Quantified Boolean Formulas (QBF) that lets us leverage QBF solvers to compute winning strategies in two-player games described in GDL. We implement this approach and measure the computation time needed by state-of-the-art QBF solvers on a range of two-player zero-sum turn-taking games. We introduce a variety of optimizations to the translation and evaluate them experimentally. Our empirical analysis establishes that our proposed approach is suitable for solving small games and can potentially help general game players evaluate endgame positions.

The recent popularity of Wordle has revived interest in guessing games. We develop a general method for finding optimal strategies for guessing games while avoiding an exhaustive search. Our main contribution are several theorems that build towards a general theory to prove optimality of a strategy for a guessing game. This work is developed to apply to any guessing game, but we use Wordle as an example to present concrete results.

Communication actions in games are usually given meaning by either the effect they have on the game state or a possible reduction in the size of the information set of an agent. However, this precludes analysis of games involving state-independent communication , in which players are given the ability to communicate with each other but are not required to be truthful. As such, these communication actions cannot be used to reduce the size of the information set or update beliefs in states without considering the intent of the communicating agent. In this paper, we introduce a language to describe the rules of such games as an extension of the Game Description Language (GDL) . We also identify a set of scenarios involving state-independent communication actions in which an effective agent should be able to derive information, and propose and evaluate strategies for reasoning about such actions in these scenarios.

Complex robot behaviour typically requires the integration of multiple robotic and Artificial Intelligence (AI) techniques and components. Integrating such disparate components into a coherent system, while also ensuring global properties and behaviours, is a significant challenge for cognitive robotics. Using a formal framework to model the interactions between components can be an important step in dealing with this challenge. In this paper we extend an existing formal framework [Clark et al., 2016] to model complex integrated reasoning behaviours of robotic systems; from symbolic planning through to online learning of policies and transition systems. Furthermore the new framework allows for a more flexible modelling of the interactions between different reasoning components.

General Game Playing agents are capable of learning to play games they have never seen before, merely by looking at a formal description of the rules of a game. Recent developments in deep learning have influenced the way state-of-the-art AI systems can learn to play games with perfect information like Chess and Go. This development is popularised by the success of AlphaZero and was subsequently generalised to arbitrary games describable in the general Game Description Language, GDL. Many real-world problems, however, are non-deterministic and involve actors with concealed information, or events with probabilistic outcomes. We describe a framework and system for General Game Playing with self-play reinforcement learning and search for hidden-information games, which can be applied to any game describable in the extended Game Description Language for imperfect-information games, GDL-II.

Learning to master new games with nothing but the rules given is a hallmark of human intelligence. This ability has recently been successfully replicated in AI systems through a combination of Knowledge Representation, Monte Carlo Tree Search and Deep Reinforcement Learning: Generalised AlphaZero [7] provides a method for building general game-playing agents that can learn any game describable in a formal specification language. We investigate how to boost the ability of deep reinforcement agents for general game playing by applying transfer learning for new game variants. Experiments show that transfer learning can significantly reduce the training time on variations of games that were previously learned, and our results further suggest that the most successful method is to train a source network that uses the guidance of multiple expert networks.

The standard representation formalism for multi-agent epistemic planning has one central disadvantage: When you use event models in dynamic epistemic logic (DEL) to describe the action of one agent, the model must specify not only the actual change and the change of that agent's knowledge. Also required is the epistemic change of any agents that may be observing the first agent performing the action, plus the epistemic change for any further agents that failed to observe that anything had taken place. To overcome the gap between this complex DEL notion of events and a more commonsense notion of actions, we propose a simple high-level action description language for multi-agent epistemic planning domains with just one type of effect laws: a causes x if y. Effect x can either be a physical effect, or an observation from an independent set that is specific to individual agents. We formally prove that any DEL event model can be described in this way. We show how this language provides a framework for expressing a variety of executability and action models; such as describing actions that are both ontic and epistemic, partially observable, or nondeterministic. We further combine our representation of event models with a description language for finitary initial epistemic theories, and we show how this allows us to reason about the effects of a sequence of actions in a multi-agent epistemic domain by updating a single multi-pointed epistemic model.

General Game Playing (GGP) is a platform for developing general Artificial Intelligence algorithms to play a large variety of games that are unknown to players in advance. This paper describes and analyses GGPZero, a learning architecture for GGP, inspired by the success of AlphaGo and AlphaZero. GGPZero takes as input a previously unknown game description and constructs a deep neural network to be trained using self-play together with Monte-Carlo Tree Search. The general architecture of GGPZero is similar to that of Goldwaser and Thielscher (2020) [4] with the main differences in the choice of the GGP reasoner and the neural network construction; furthermore, we explore additional experimental evaluation strategies. Our main contributions are: confirming the feasibility of deep reinforcement for GGP, analysing the impact of the type and depth of the underlying neural network, and investigating simulation vs. time limitations on training.

General Game Playing agents are required to play games they have never seen before simply by looking at a formal description of the rules of the game at runtime. Previous successful agents have been based on search with generic heuristics, with almost no work done into using machine learning. Recent advances in deep reinforcement learning have shown it to be successful in some two-player zero-sum board games such as Chess and Go. This work applies deep reinforcement learning to General Game Playing, extending the AlphaZero algorithm and finds that it can provide competitive results.