New Criteria and a New Algorithm for Learning in Multi-Agent Systems

By Rob Powers, Yoav Shoham. 2004

128 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.

Citations in the corpus

Empirically Evaluating Multiagent Learning Algorithms (2014, methodology): “The Powers and Shoham (2005) implementation of meta used a distance measure based on the Hoeffding Inequality, even though the pseudo-code called for using an ℓ 2 norm.”
An Overview of Cooperative and Competitive Multiagent Learning (2005, methodology): “[128] provide suggestions for different criteria for evaluating MARL algorithms.”
If multi-agent learning is the answer, what is the question? (2007, background): “In [45] we define a criterion parameterized by a class of ‘target opponents’; with this parameter we make three requirements of any learning algorithm: (1) (Targeted”
Emergence of Norms through Social Learning (2007, background): “Most multiagent reinforcement learning literature involve two agents iteratively playing a stage game and the goal is to learn policies to reach preferred equilibrium [Powers and Shoham, 2005].”
A Survey of Learning in Multiagent Environments: Dealing with Non-Stationarity (2017, methodology): “Moreover, Manipulator has the same guarantees as MetaStrategy against a richer class of target opponents, memory-bounded opponents.12 These are defined as opponents that play a conditional strategy where actions can only depend on recent periods, this is its distribution over actions can only depend”
Learning against multiple opponents (2006, methodology): “We will use the testing environment first described in [22] by testing against a number of existing approaches from the multi-agent learning literature over a wide variety of repeated games from GAMUT [19].”
Multi-agent Reinforcement Learning: An Overview (2010, methodology): “Not only algorithms speciﬁcally designed for static games are tested on such tasks (e.g., AWESOME [31], MetaStrategy [93], GIGA-WoLF [14]), but also others that can, in principle, handle dynamic stochastic games (e.g., EXORL [116]).”
On the Rate of Convergence of Fictitious Play (2010, unknown): “3 a 1 0 −1 − a 2 1 0 − a 3 0”
A Comprehensive Survey of Multiagent Reinforcement Learning (2008, unknown): “Some steps in this direction have been taken in [55] and [57].”
AWESOME: A general multiagent learning algorithm that converges in self-play and learns a best response against stationary opponents (2003, background): “larger classes of opponents ( Powers & Shoham, 2005b; Powers & Shoham, 2005a).”