Multiagent Incremental Learning in Networks

Learning in a Fixed or Evolving Network of Agents

“…The research challenge is then to design mechanisms that enjoy efficiency properties without being overwhelmingly burdening as far as communication is concerned. This paper completes the works of [1], which was, to our knowledge, the first attempt to tackle both features simultaneously, by describing and experimenting new protocols improving the results in various situations. Then, we extends this previous work to deal with variable topologies.…”

“…However, this solution offers no guarantee when we have to deal with some communicationnal constraints restricting the communicationnal links between the agents. A global revision mechanism was proposed in [1], but in some cases (e.g. lines networks), the obtained accuracy was not as good as Smile.…”

“…The global mechanism must then rely on some kind of propagation. In [1], we proposed a mechanism in which the agent triggering the mechanism becomes the root of a spanning tree covering the whole society. It proposes its hypothesis to its neighbors (becoming its children), who will in turn connect to unmarked neighbors and propose them the hypothesis.…”

“…In [1], we use a factorized way to deal with counter-examples. When an agent receives a counter-example from a i , it will first finish the local revision with a i to get all relevant counter-examples, finally getting an indirect accept from it, and then asks its parent to begin a new local revision with it, until all these counter-examples reach the root.…”

“…This work will be based on a hypothesis refinement framework described in [4], and some first experimental results presented in [1]: the general idea of this mechanism is to allow agents to propagate relevant examples to a learning agent. While the proposed mechanism has been shown to guarantee some theoretical consistency properties, experimental study of its efficiency and effectiveness in a multi-agent symbolic learning perspective gave some hints of possible ways to improve it.…”