Benchmarking Effects of Communication Bandwidths and Number of Options on Collective Decision-making Strategies in Discrete Collective Estimation

Abstract

Multi-option collective decision making is an emergent topic of study within the field of swarm intelligence. Many strategies have been proposed to enable decentralized and localized decision-making behaviors in intelligent swarms. However, when applied to multi-option scenarios , many proposed strategies have very different requirements on the communication bandwidth and paradigm, which make a clear and fair comparison difficult. In this paper, we seek to investigate and compare the performances of opinion-based decision-making strategies and newly proposed multi-option strategies in a discrete collective estimation scenario when the communication bandwidth and paradigm are controlled. The considered strategies’ performances are assessed via error, consensus time and failure rate. We have experimented on decision-making scenarios with random and concentrated distribution of features, as well as different number of available options. Among the considered strategies, we have observed that distributed Bayesian belief sharing (DBBS) has superior performances in all three metrics, especially at higher communication bandwidths. On the other hand, ranked voting with Borda count (RV-BC) has comparable performances to the baseline opinion-based strategies at lower bandwidths, while slightly outperforms at higher bandwidths. In contrast, opinion-based approaches are competitive at lower communication bandwidths, and are less adversely affected when facing a high number of available options.