New Zealand

Yi-Ting Tsao, Ke-Ting Xiao, Von-Wun Soo and Chung-Cheng Chiu (June 1st 2010). Graph Laplacian Based Transfer Learning Methods in Reinforcement Learning, Autonomous Agents, Vedran Kordic, IntechOpen, DOI: 10.5772/9654. 

Kunikazu Kobayashi, Koji Nakano, Takashi Kuremoto and Masanao Obayashi (February 1st 2010). Objective-based Reinforcement Learning System for Cooperative Behavior Acquisition, Application of Machine Learning, Yagang Zhang, IntechOpen, DOI: 10.5772/8615. 

Erik Billing, Thomas Hellström and Lars-Erik Janlert (February 3rd 2012). Robot Learning from Demonstration Using Predictive Sequence Learning, Robotic Systems - Applications, Control and Programming, Ashish Dutta, IntechOpen, DOI: 10.5772/26165. 

While reinforcement learning (RL) has helped artificial agents solve challenging tasks, high sample complexity is still a major concern. Inter-agent teaching -- endowing agents with the ability to respond to instructions from others -- has been responsible for many developments towards scaling up RL. RL agents that can leverage instructions can learn tasks significantly faster than agents that cannot take advantage of such instruction. That said, the inter-agent teaching paradigm presents many new challenges due to, among other factors, differences between the agents involved in the teaching interaction. This paper is a summary of our JAAMAS article, where we propose two frameworks that provide a comprehensive view of the challenges associated with inter-agent teaching. We highlight state-of-the-art solutions, open problems, prospective applications, and argue that new research in this area should be developed in the context of the proposed frameworks.


AAMAS 2020

Agents Teaching Agents: A Survey on Inter-agent Transfer Learning

learning from demonstrations

transfer learning

reinforcement learning

AAMAS is the leading scientific conference for research in autonomous agents and multi-agent systems. The AAMAS conference series was initiated in 2002 as the merging of three respected scientific meetings: the International Conference on Multi-Agent Systems (ICMAS), the International Workshop on Agent Theories, Architectures, and Languages (ATAL), and the International Conference on Autonomous Agents (AA). The aim of the joint conference is to provide a single, high-profile, internationally-respected archival forum for scientific research in the theory and practice of autonomous agents and multi-agent systems.

Browse keynotes, discussions, panels and over 300 presentations.


AAMAS is the leading scientific conference for research in autonomous agents and multi-agent systems. The AAMAS conference series was initiated in 2002 as the merging of three respected scientific meetings: the International Conference on Multi-Agent Systems (ICMAS), the International Workshop on Agent Theories, Architectures, and Languages (ATAL), and the International Conference on Autonomous Agents (AA). 

technical paper

Opportunism is a behavior that takes advantage of knowledge asymmetry and results in promoting agents' own value and demoting others' value. It is important to eliminate such selfish behavior in multi-agent systems, as it has undesirable results for the participating agents. In order for monitoring and eliminating mechanisms to be put in the right place, it is needed to know in which context agents are likely to perform opportunistic behavior. In this paper, we develop a formal framework to reason about agents' opportunistic propensity. Opportunistic propensity refers to the potential for an agent to perform opportunistic behavior. In particular, agents in the system are assumed to have their own value systems and knowledge. With value systems, we define agents' state preferences. Based on their value systems and incomplete knowledge about the state, they choose one of their rational alternatives to perform, which might be opportunistic behavior. We then characterize the situation where agents are likely to perform opportunistic behavior and the contexts where opportunism is impossible to occur, and prove the computational complexity of predicting opportunism.


A Formal Framework for Reasoning about Opportunistic Propensity in Multi-agent Systems

We study a sequential preference aggregation procedure based on voting rules for settings where several agents express their preferences over a common set of variable assignments via soft constraints. We evaluate this approach by providing both theoretical and experimental results. 


Sequential Voting in Multi-agent Soft Constraint Aggregation

Many real-world decision problems are inherently multi-objective in nature and concern multiple actors, making multi-objective multi-agent systems a key domain to study. We argue that trade-offs between conflicting objective functions should be analysed on the basis of the utility that these trade-offs have for the users of a system. We develop a new taxonomy which classifies multi-objective multi-agent decision making settings, on the basis of the reward structures and utility functions. We analyse which solution concepts apply to the different settings in our taxonomy, which allows us to offer a structured view of the field and identify promising directions for future research.


Multi-Objective Multi-Agent Decision Making: A Utility-based Analysis and Survey

Multi-objective multi-agent decision making: a utility based analysis and survey

The literature on norm emergence and normative MAS considers norms from two perspectives, namely: prescriptive norms using deontic concepts, and emergent norms that capture preference behaviour. We find that both perspectives lend themselves naturally to specific ways of representing norms in a society, as either explicit or implicit. 
Our analysis of the norm emergence literature, contributes several insights for future research in normative MAS. For example, opportunities for the study of online norm synthesis mechanisms and the investigation of the conversion of social norms, typically observed in norm emergence, to legal norms in normative MAS.  
Conversely, concepts from normative MAS can be brought into norm emergence research. For example, the study of stable emergence to avoid instability in MAS and the investigation of whether high cognitive ability agents, typically found in normative MAS, can demonstrate norm emergence. 
We summarise our analysis and outline future challenges and opportunities for cross-over between norm-emergence and normative systems research.


Norm Emergence in Multiagent Systems: A Viewpoint Paper

This presentation presents the Multi-Agent based Real-Time INfrastructure for Energy (MARTINE). MARTINE is a simulation and emulation infrastructure for the study of power and energy systems using a combination of artificial intelligence approaches. MARTINE combines real buildings with sensoring and actuation capabilities with real-time simulation, emulation of physical resources and the intelligent decision support to players actions. The infrastructure is managed and operated by means of several multi-agent systems, which connect to physical resources but also represent additional simulated players that are not present physically in the simulation and emulation environment.


MARTINE: Multi-Agent based Real-Time INfrastructure for Energy

When robots act in an environment, there will be temporal uncertainty over the execution of their actions, i.e. the duration of an action and the time it takes place will be stochastic. The presence of multiple robots in the environment contributes towards this uncertainty.  Temporal uncertainty is often disregarded in multi-robot coordination, and so in this presentation, I will discuss why it is important to model temporal uncertainty, and detail the benefits of doing so. Following this, I will outline my current work modelling temporal uncertainty in terms of congestion in order to generate effective plans for multi-robot navigation tasks. Finally, I will discuss the requirements to achieve general multi-robot coordination under temporal uncertainty.


Towards Multi-Robot Coordination under Temporal Uncertainty (Doctoral Consortium)

This paper presents part of the work developed so far within the scope of my PhD and suggests possible future research directions. My thesis tackles the problem of multi-robot coordination under uncertainty over the long-term. We present a preliminary approach that tackles multi-robot monitoring problems under uncertain task durations. We propose a methodology that takes advantage of a modeling formalism for robot teams: generalized stochastic Petri nets with rewards (GSPNR). A GSPNR allows for unified modeling of action selection and uncertainty on duration of action execution. At the same time, it allows for goal specification through the use of transition rewards and rewards per time unit. The proposed approach exploits the well-defined semantics provided by Markov reward automata in order to synthesize policies.


Long-Run Multi-Robot Planning Under Uncertain Task Durations

Coalition formation games aim at predicting the cooperative behavior of agents when forming alliances. Agents entertain preferences over coalition structures, and the goal is to find a coalition structure that is good for both individual agents and the society as an entity. We measure the quality of partitions in terms of Pareto optimality, individual rationality, and various notions of popularity. We give both efficient algorithms and hardness results for computing partitions that satisfy these properties for additively separable hedonic games and give an outlook to results on other classes of coalition formation games, including roommate games and flatmate games.

Desirable Partitions in Coalition Formation Games

The advent of data marketplaces and the increasing awareness of data privacy call for data privacy pricing mechanisms which can set prices of data properly while assuring privacy preservation. Moreover, the trade-off between the privacy of data and the accuracy of results need to be considered. Additionally, data marketplaces have different structures, including buy-sided market, sell-sided market, two-sided market, and two-sided platform. I will explore data privacy pricing mechanisms under these four structures.


Downloads

Next from AAMAS 2020

A Formal Framework for Reasoning about Opportunistic Propensity in Multi-agent Systems

Similar lecture

Learning to Incentivize Other Learning Agents

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from AAMAS 2020

A Formal Framework for Reasoning about Opportunistic Propensity in Multi-agent Systems

Similar lecture

Learning to Incentivize Other Learning Agents

Downloads