The Simplification Conspiracy

We study in a quantitative way the efficacy of a social intelligence scheme that is an extension of Extreme Learning Machine paradigm. The key question we investigate is whether and how a collection of elementary learning parcels can replace a single algorithm that is well suited to learn a relatively complex function. Per se, the question is definitely not new, as it can be met in various fields ranging from social networks to bio-informatics. We use a well known benchmark as a touchstone to contribute its answer with both theoretical and numerical considerations.