"Knowledge representation and reasoning is a core research area in artificial intelligence. This research has provided deep insight into computational aspects of knowledge and formal aspects of commonsense reasoning. Besides its impact on artificial intelligence, knowledge representation has also contributed to the wider field of computer science. While the central issues have changed over the years, new open problems arise to challenge the best minds." Norman Foo, KR Conventicle 2002
The aim of the Australian Knowledge Representation and Multiagent Systems Conventicle is to:
Examine the state-of-the-art in knowledge representation and multiagent systems
Identify the focal issues at hand
Set new directions for future research.
The conventicle is two-days long and will be held at UNSW Australia on May 13–14.
Location: CSE Innovation Hub (K17_B09), Basement, CSE Building (K17). Click here for a campus map.
Mehdi Dastani, Utrecht University
Logics for Safe and Efficient Reinforcement Learning
In this talk, I present our recent work in using logics to improve safety and efficiency of reinformcement learning algorithms. First, I present the use of logic to synthesize reward machines that improve the sample efficiency of reinforcement learning by exposing reward structure to the learning agent. We show how reward machines for team tasks can be automatically synthesised from an Alternating-Time Temporal Logic specification of the desired team behaviour. Second, I present the use of logic for action masking in safe reinforcement learning. Unsafe actions are disallowed ("masked") according to specifications expressed in Pure-Past Linear Temporal Logic. Our approach can enforce non-Markovian constraints, i.e., constraints based on the history of the system, rather than just the current state of the Markoc Decision Processes.
Mehdi Dastani is a computer scientist and chair of the Intelligent Systems group in the Department of Information and Computing Sciences at Utrecht University. His research focuses on formal and computational models for intelligent systems. His current research focuses on theories and applications of multi-agent systems, in particular specification and programming languages for autonomous agents and multi-agent systems, logics for reasoning about (multi-)agent specifications and programs, and combining logic and machine learning approaches to develop autonomous agents and multi-agent systems.
Virginia Dignum, Umea University
The Role of Formalisation in Responsible AI
Virginia Dignum is Professor of Responsible Artificial Intelligence at Umea University, Sweden, where she leads the AI Policy Lab. She is also senior advisor on AI policy to the Wallenberg Foundations. She has a PHD in Artificial Intelligence from Utrecht University in 2004, is a member of the Royal Swedish Academy of Engineering Sciences (IVA), and Fellow of the European Artificial Intelligence Association (EURAI). She is a member of the United Nations Advisory Body on AI, the Global Partnership on AI (GPAI), UNESCO's expert group on the implementation of AI recommendations, OECD's Expert group on AI, founder of ALLAI, the Dutch AI Alliance, and co-chair of the WEF's Global Future Council on AI. She was a member of EU's High Level Expert Group on Artificial Intelligence and leader of UNICEF's guidance for AI and children. Her new book "The AI Paradox" is planned for publication in late 2024.
Wojtek Jamroga, University of Luxembourg and Polish Academy of Sciences
Pretty Good Strategies to Cast Your Vote Securely
Benaloh challenge allows the voter to check whether the vote has not been altered while being cast. It is especially important when the vote is being sent in encrypted form, and the encryption cannot be fully trusted. We present a simple game-theoretic analysis of the mechanism. In particular, it has been claimed that the voter has no "natural" rational strategy to play against the encryption device, which undermines the whole idea. Here, we claim the contrary, i.e., that there exist simple rational strategies that justify the usefulness of Benaloh challenge. This talk is about an application of game theory to the analysis of multi-agent interaction in elections.
Wojtek Jamroga is a full professor at the Polish Academy of Sciences and a research scientist at the University of Luxembourg. His research focuses on modeling, specification and verification of interaction between autonomous agents. He has coauthored around 150 refereed publications, and has been a Program Committee member of most important conferences and workshops in AI and multi-agent systems. According to Google Scholar, his papers have been cited over 3100 times, and his H-index is 29. The research track of Prof. Jamroga includes the Best Paper Award at the main conference on electronic voting (E-VOTE-ID) in 2016, and a Best Paper Nomination at the main multi-agent systems conference (AAMAS) in 2018. His teaching record includes numerous courses at ESSLLI (European Summer School in Logic, Language and Information), EASSS (European Agent Systems Summer School), and ESSAI (European Summer School on AI), several courses at doctoral schools, and tutorials at top conferences in AI and multi-agent systems – all of them on formal methods for multi-agent systems.
Aniello Murano, University of Naples Frederico II
TBA
Laurent Perrussel, University Toulouse Capitole
Strategic Reasoning and Mechanism Design
In this talk, we will see how Logics for Strategic Reasoning may play a key role for automating Mechanism Design. After reminding the key components of Strategy Logic, we show how classical Mechanism properties may be encoded as plain logical objects and how verying the correctness of a mechanism boils down to Model Checking. We then go further by showing how Mechanism Design can then be rephrased as Synthesis.
Laurent Perrussel is a full professor at University Toulouse Capitole since 2015. From 1995, LP is a member of the AI research group. LP's current research is on strategic reasoning and GDL and he is an expert on multiagent systems and reasoning about actions. He tutored or co-tutored seven PhD students and he published ~35 A-A* conference papers and journal papers. He gave tutorials at KR, IJCAI and AAMAS in the last years and he will be a lecturer at the summer school ESSAI-24.
Tran Cao Son, New Mexico State University
Recent Improvements to Action Language mA*
The action language mA* was developed for reasoning about actions and change (RAC) in multi-agent domains. It extends the approach to RAC using action language in single agent environment to multi-agent environment in that it has an English-like syntax and a transition function based semantics. A key insight of mA* is the identification of the different roles of agents in an action occurrence which are dynamic. For this reason, mA* includes specifications for agents' observability of action occurrences. mA* utilizes the notions of Kripke structure and update models in defining the transitions between states. It is shown that under certain conditions, the semantics of mA* yields intuitive results. In designing mA*, the authors of the language emphasized its simplicity and usability (e.g., using mA* as a specification language for multi-agent planning). This design decision left open some of the key problems in reasoning about beliefs. The first problem is the second-order false belief and the second one is dealing with false beliefs and maintaining KD45 properties. In this talk, I will present different solutions to these two problems. For the first problem, I will present two solutions and discuss their differences. I will introduce a new definition of the update operator between a pointed Kripke structure and an update model that solves the second problem.
Son Tran is a Professor and Head of the CS department at NMSU. His research has been focused on logic based AI in automated planning, diagnosis, commonsense reasoning, multi-agent systems, and the development of practical applications using answer set programming such as distributed constraint optimization, multi-agent path findings, automated negotiation, and web-service composition. He led the development of several state-of-the-art planning systems for agents with incomplete information and is currently leading the effort to create an epistemic planner for multiple agents. In recent years, he is interested in reasoning about actions and change (RAC) in multi-agent domains. explainable planning, generating explanations for planning failure, and the development of reasoning framework for trustworthiness of CPS systems. He has published several research papers in premier conferences in AI.
Dengji Zhao, Shanghai Tech University
Incentives for Early Arrival in Cooperative Games
We study cooperative games where players join sequentially, and the value generated by those who have joined at any point must be irrevocably divided among these players. We introduce two desiderata for the value division mechanism: that the players should have incentives to join as early as possible, and that the division should be considered fair. For the latter, we require that each player's expected share in the mechanism should equal her Shapley value if the players' arrival order is uniformly at random. When the value generation function is submodular, allocating the marginal value to the player satisfies these properties. This is no longer true for more general functions. Our main technical contribution is a complete characterization of 0-1 value games for which desired mechanisms exist. We show that a natural mechanism, Rewarding First Critical Player (RFC), is complete, in that a 0-1 value function admits a mechanism with the properties above if and only if RFC satisfies them; we analytically characterize all such value functions. Moreover, we give an algorithm that decomposes, in an online fashion, any value function into 0-1 value functions, on each of which RFC can be run. In this way, we design an extension of RFC for general monotone games, and the properties are proved to be maintained.
Dengji Zhao is an Associate Professor at ShanghaiTech (Shanghai, China) and lead The ShanghaiTech Multi-Agent systems Research Team (SMART). He was elected as a member of the IFAAMAS Board of Directors in 2022, the first representative from a Chinese institution since IFAAMAS was founded in 2002. He is an Associate Editor of JAAMAS. He is a senior member of IEEE and the China Computer Federation (CCF). He was program/track co-chair of DAI 2021, Workshop Track of PRICAI 2021, Competitions Track of IJCAI 2022, Demo and Competitons Track of AAMAS 2023, and Survey Track of IJCAI 2024. He is a local co-chair of WINE 2023. Most of Dengji's research is on algorithmic game theory and multi-agent systems, especially mechanism design and its applications on social networks. He pioneered and promoted a new trend of mechanism design on social networks, namely how to incentivize the existing participants of a game to invite new participants via their social connections. For this research challenge, Zhao's group have made several seminal contributions in auctions, coalitional games and matching with over 20 high-quality papers. He has also contributed a blue sky paper at AAMAS 2021 to set up the related research agenda and offered four tutorials at AAAI 2022, AAMAS 2019 and IJCAI 2017/2018. He was invited to give an Early Career Spotlight talk at IJCAI-ECAI 2022.
There will be no registration fee. Morning and afternoon coffee with light refreshments will be provided.
| Monday, 13 May | |
|---|---|
| 9:30 | Opening |
| Session 1: Reasoning About Actions and Beliefs | |
| 9:35 | Keynote: Tran Cao Son (New Mexico State University). Recent Improvements to Action Language mA* |
| 10:35 | Coffee Break |
| 11:00 | Thorsten Engesser (Paul Sabatier University Toulouse). Towards Epistemic-Doxastic Planning with Observation and Revision |
| 11:30 | Abhaya Nayak (Macquarie University). Relevance, Recovery and Recuperation: A Prelude to Ring Withdrawal |
| 12:00 | Yiduo Ke (Northwestern University). Fair Allocation of Conflicting Courses Under Additive Utilities |
| 12:30 | Lunch Break |
| Session 2: Theory of Agents and Multi-agent Systems | |
| 13:30 | Keynote: Mehdi Dastani (Utrecht University). Logics for Safe and Efficient Reinforcement Learning |
| 14:30 | Qihui Feng (Aachen University). Multi-agent Only-believing |
| 15:00 | Jaber Valizadeh (Western Sydney University). The Power of Autonomy |
| 15:30 | Coffee Break |
| Session 3: Formalisation | |
| 16:00 | Keynote: Virginia Dignum (Umea University). The Role of Formalisation in Responsible AI |
| 17:00 | (To Be Confirmed) Keynote: Aniello Murano (University of Naples Frederico II). TBA |
| Tuesday, 14 May | |
| Session 4: Games I | |
| 9:30 | Keynote: Dengji Zhao. Incentives for Early Arrival in Cooperative Games |
| 10:30 | Coffee Break |
| 11:00 | Yifan He (University of New South Wales). Solving Two-player Games with QBF Solvers in General Game Playing |
| 11:30 | Qi Wang (Western Sydney University). Modelling Congestion and price competition in EV Charging Markets |
| 12:00 | Guido Governatori (Charles Sturt University). Debunking Deontic Logic (Paradoxes). |
| 12:30 | Lunch Break |
| Session 5: Games II | |
| 13:30 | Keynote: Laurent Perrussel (University Toulouse Capitole). Strategic Reasoning and Mechanism Design |
| 14:30 | Damian Kurpiewski (Polish Academy of Sciences). STV: Towards Practical Verification of Strategic Ability |
| 15:00 | Moritz Graf (University of Freiburg). Symbolic Computation of Sequential Equilibria |
| 15:30 | Coffee Break |
| Session 6: Social Choice | |
| 16:00 | Keynote: Wojtek Jamroga (University of Luxembourg, Polish Academy of Sciences). Pretty Good Strategies to Cast Your Vote Securely |
| 17:00 | Muhammad Hassan Ali Bajawa (Macquarie University). Designing Intelligent Conversational Agents in Serious Games for Ethics Training |
| 17:30 | Sanjukta Roy (University of Leeds). Gerrymandering Graphs: Hard or Easy? |
| 18:00 | Closing |