Abstract

This submission analyses various sharing models for institutional resources (infrastructure) to support research and teaching activities around artificial intelligence and data science. We wish to explore beyond traditional partnership models between organisations or tertiary institutions and include optimal mathematical models for pooling infrastructure. Given a region of interest, the selected models should help identify the best clustering approach, building on the strength of each entity within a cluster. Additional factors such as energy cost and reusability will be considered.
Our intuition is to draw inspiration from game theoretic algorithms, where negotiation algorithms are integrated with coalition games to produce near optimal heuristics for pooled resource allocation.
We will present the underlying mechanism design and a set of adapted algorithms to analyse the resulting games. Our experiments will focus on computing infrastructure (e.g., standalone servers, commodity clusters and high-end infrastructure).