The Communication of Inductive Inferences

W. H. E. Davies, Peter Edwards

Research output: Chapter in Book/Report/Conference proceedingChapter

7 Citations (Scopus)
5 Downloads (Pure)

Abstract

We propose a new approach to communication between agents that perform inductive inference. Consider a community of agents where each agent has a limited view of the overall world. When an agent in this community induces a hypothesis about the world, it necessarily reflects that agent's partial view of the world. If an agent communicates a hypothesis to another agent, and that hypothesis is in conflict with the receiving agent's view of the world, then the receiving agent has to modify or discard the hypothesis. Previous systems have used voting methods or theory refinement techniques to integrate these partial hypotheses. However, these mechanisms risk destroying parts of the hypothesis that are correct. Our proposal is that an agent should communicate the bounds of an induced hypothesis, along with the hypothesis itself. These bounds allow the hypotheses to be judged in the context from which they were formed. This paper examines using version space boundary sets to represent these bounds. Version space boundary sets may be manipulated using set operations. These operations can be used to evaluate and integrate multiple partial hypotheses. We describe a simple implementation of this approach, and draw some conclusions on its practicality. Finally, we describe a tentative set of KQML operators for communicating hypotheses and their bounds.
Original languageEnglish
Title of host publicationDistributed Artificial Intelligence Meets Machine Learning: Learning in Multi-Agent Environments
EditorsG Weiss
PublisherSpringer-Verlag
Pages223-241
Volume1221
ISBN (Electronic)978-3-540-69050-4
ISBN (Print)978-3-540-62934-4
DOIs
Publication statusPublished - 1997

Fingerprint

Communication

Keywords

  • multi-agent systems
  • learning
  • communication
  • induction

Cite this

Davies, W. H. E., & Edwards, P. (1997). The Communication of Inductive Inferences. In G. Weiss (Ed.), Distributed Artificial Intelligence Meets Machine Learning: Learning in Multi-Agent Environments (Vol. 1221, pp. 223-241). Springer-Verlag. https://doi.org/10.1007/3-540-62934-3_51

The Communication of Inductive Inferences. / Davies, W. H. E.; Edwards, Peter.

Distributed Artificial Intelligence Meets Machine Learning: Learning in Multi-Agent Environments. ed. / G Weiss. Vol. 1221 Springer-Verlag, 1997. p. 223-241.

Research output: Chapter in Book/Report/Conference proceedingChapter

Davies, WHE & Edwards, P 1997, The Communication of Inductive Inferences. in G Weiss (ed.), Distributed Artificial Intelligence Meets Machine Learning: Learning in Multi-Agent Environments. vol. 1221, Springer-Verlag, pp. 223-241. https://doi.org/10.1007/3-540-62934-3_51
Davies WHE, Edwards P. The Communication of Inductive Inferences. In Weiss G, editor, Distributed Artificial Intelligence Meets Machine Learning: Learning in Multi-Agent Environments. Vol. 1221. Springer-Verlag. 1997. p. 223-241 https://doi.org/10.1007/3-540-62934-3_51
Davies, W. H. E. ; Edwards, Peter. / The Communication of Inductive Inferences. Distributed Artificial Intelligence Meets Machine Learning: Learning in Multi-Agent Environments. editor / G Weiss. Vol. 1221 Springer-Verlag, 1997. pp. 223-241
@inbook{629e0b46ef5a465bba1b5479eb9510c2,
title = "The Communication of Inductive Inferences",
abstract = "We propose a new approach to communication between agents that perform inductive inference. Consider a community of agents where each agent has a limited view of the overall world. When an agent in this community induces a hypothesis about the world, it necessarily reflects that agent's partial view of the world. If an agent communicates a hypothesis to another agent, and that hypothesis is in conflict with the receiving agent's view of the world, then the receiving agent has to modify or discard the hypothesis. Previous systems have used voting methods or theory refinement techniques to integrate these partial hypotheses. However, these mechanisms risk destroying parts of the hypothesis that are correct. Our proposal is that an agent should communicate the bounds of an induced hypothesis, along with the hypothesis itself. These bounds allow the hypotheses to be judged in the context from which they were formed. This paper examines using version space boundary sets to represent these bounds. Version space boundary sets may be manipulated using set operations. These operations can be used to evaluate and integrate multiple partial hypotheses. We describe a simple implementation of this approach, and draw some conclusions on its practicality. Finally, we describe a tentative set of KQML operators for communicating hypotheses and their bounds.",
keywords = "multi-agent systems, learning, communication, induction",
author = "Davies, {W. H. E.} and Peter Edwards",
year = "1997",
doi = "10.1007/3-540-62934-3_51",
language = "English",
isbn = "978-3-540-62934-4",
volume = "1221",
pages = "223--241",
editor = "G Weiss",
booktitle = "Distributed Artificial Intelligence Meets Machine Learning: Learning in Multi-Agent Environments",
publisher = "Springer-Verlag",

}

TY - CHAP

T1 - The Communication of Inductive Inferences

AU - Davies, W. H. E.

AU - Edwards, Peter

PY - 1997

Y1 - 1997

N2 - We propose a new approach to communication between agents that perform inductive inference. Consider a community of agents where each agent has a limited view of the overall world. When an agent in this community induces a hypothesis about the world, it necessarily reflects that agent's partial view of the world. If an agent communicates a hypothesis to another agent, and that hypothesis is in conflict with the receiving agent's view of the world, then the receiving agent has to modify or discard the hypothesis. Previous systems have used voting methods or theory refinement techniques to integrate these partial hypotheses. However, these mechanisms risk destroying parts of the hypothesis that are correct. Our proposal is that an agent should communicate the bounds of an induced hypothesis, along with the hypothesis itself. These bounds allow the hypotheses to be judged in the context from which they were formed. This paper examines using version space boundary sets to represent these bounds. Version space boundary sets may be manipulated using set operations. These operations can be used to evaluate and integrate multiple partial hypotheses. We describe a simple implementation of this approach, and draw some conclusions on its practicality. Finally, we describe a tentative set of KQML operators for communicating hypotheses and their bounds.

AB - We propose a new approach to communication between agents that perform inductive inference. Consider a community of agents where each agent has a limited view of the overall world. When an agent in this community induces a hypothesis about the world, it necessarily reflects that agent's partial view of the world. If an agent communicates a hypothesis to another agent, and that hypothesis is in conflict with the receiving agent's view of the world, then the receiving agent has to modify or discard the hypothesis. Previous systems have used voting methods or theory refinement techniques to integrate these partial hypotheses. However, these mechanisms risk destroying parts of the hypothesis that are correct. Our proposal is that an agent should communicate the bounds of an induced hypothesis, along with the hypothesis itself. These bounds allow the hypotheses to be judged in the context from which they were formed. This paper examines using version space boundary sets to represent these bounds. Version space boundary sets may be manipulated using set operations. These operations can be used to evaluate and integrate multiple partial hypotheses. We describe a simple implementation of this approach, and draw some conclusions on its practicality. Finally, we describe a tentative set of KQML operators for communicating hypotheses and their bounds.

KW - multi-agent systems

KW - learning

KW - communication

KW - induction

U2 - 10.1007/3-540-62934-3_51

DO - 10.1007/3-540-62934-3_51

M3 - Chapter

SN - 978-3-540-62934-4

VL - 1221

SP - 223

EP - 241

BT - Distributed Artificial Intelligence Meets Machine Learning: Learning in Multi-Agent Environments

A2 - Weiss, G

PB - Springer-Verlag

ER -