Reinforcement learning of coordination in cooperative multi-agent systems

By Spiros Kapetanakis, Daniel Kudenko. 2002

234 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Reinforcement Learning

Citations in the corpus

Multiagent Bidirectionally-Coordinated Nets: Emergence of Human-level Coordination in Learning to Play StarCraft Combat Games (2017, background): “The idea of learning independent agents can also be found (Lauer and Riedmiller 2000; Kapetanakis and Kudenko 2002; Lauer and Riedmiller 2004; Foerster et al. 2016).”
Deep Decentralized Multi-task Multi-Agent Reinforcement Learning under Partial Observability (2017, background): “Frequency Maximum Q-Value heuristics (Kapetanakis & Kudenko, 2002) bias action selection towards those consistently achieving max rewards.”
Mean Field Multi-Agent Reinforcement Learning (2018, neutral): “For Rec-FMQ, we set the frequency learning rate α f = 0 .”
Independent reinforcement learners in cooperative Markov games: a survey regarding coordination problems (2012, methodology): “Kapetanakis and Kudenko (2002, 2004) proposed to bias the probability of choosing an action with the frequency of receiving the maximum reward for that action.”
Evolutionary Dynamics of Multi-Agent Learning: A Survey (2015, methodology): “Other examples include frequency maximum Q-learning (Kapetanakis & Kudenko, 2002), an algorithm tailored to coordinate in cooperative multi-agent systems, and a class of regret minimisation algorithms (Blum & Mansour, 2007) that guarantee performance close to the best fixed action in hindsight…”
Multi-agent Reinforcement Learning: An Overview (2010, unknown): “The Frequency Maximum Q-value (FMQ) heuristic is based on the frequency with which actions yielded good rewards in the past [59].”
QD-Learning: A Collaborative Distributed Strategy for Multi-Agent Reinforcement Learning Through Consensus + Innovations (2012, background): “Various formulations, ranging from general competitive dynamic stochastic games [11], [12], [13], [14] to so called fully cooperative [15], [16], [17], [18], [19], [20], [21], [22], have been investigated (see [10] for a more complete taxonomy).”
Game Theory and Multi-agent Reinforcement Learning (2012, background): “The frequency maximum Q-learning (FMQ) algorithm (Kapetanakis and Kudenko, 2002), for example, keeps a frequency value f req(R∗,a) indicating how often the maximum reward so far (R∗) has been observed for a certain action a.”
Biasing Coevolutionary Search for Optimal Multiagent Behaviors (2006, methodology): “The authors then suggested that the search could be improved by using more optimistic exploration actions; this direction was further explored in [9] and [10].”
Cooperative Multi-Agent Learning: The State of the Art (2005, background): “These games are later used in [137, 142, 192] to investigate multi-agent reinforcement learning and evolutionary computation approaches.”
An Overview of Cooperative and Competitive Multiagent Learning (2005, background): “Fully Cooperative Scenarios Research in simple stateless environments shows that multiple cooperating concurrent learners can greatly benefit from being optimistic about their teammates: the goal is not to match well your current teammates, but to expect them to improve as well due to their learning”
Lenient learners in cooperative multiagent systems (2006, methodology): “According to [2], this is equivalent to the performance of FMQ in these problem domains, and it is also signi.cantly better than the performance of traditional Q-learning approaches, as well as to an algorithm previously proposed in [3].”
Theoretical considerations of potential-based reward shaping for multi-agent systems (2011, background): “Multiple individual learners will converge to one of these equilibrium, but whether it will be the optimum cannot be guaranteed [6] With multiple agents in the same environment, altering the exploration of one will change the experiences of all agents [12, 13].”