Probability Pooling for Dependent Agents in Collective Learning
- Submitting institution
-
University of Bristol
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 247227310
- Type
- D - Journal article
- DOI
-
10.1016/j.artint.2020.103371
- Title of journal
- Artificial Intelligence
- Article number
- 103371
- First page
- -
- Volume
- 288
- Issue
- -
- ISSN
- 0004-3702
- Open access status
- Compliant
- Month of publication
- August
- Year of publication
- 2020
- URL
-
-
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
1
- Research group(s)
-
A - Artificial Intelligence and Autonomy
- Citation count
- 0
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- Information fusion operators typically assume that information sources are independent. It is argued that this assumption is optimal in the face of ignorance. By proposing a new probability pooling operator based on copulas, this paper shows that for collective learning with high noise and low evidence, it is instead optimal to assume some comonotonic dependence between agents' probabilities. Published in a leading AI journal, this paper is one of the first to investigate agent dependence in decentralised learning. The results are significant for multi-agent systems and the approach is being extended to distributed decision-making research funded by TB-PHASE EP/R004757/1 (£5M).
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -