Much of contemporary research in Artificial Immune Systems (AIS) has partitioned into either algorithmic machine learning
and optimisation,
or modelling biologically plausible dynamical systems, with little overlap between. Although the balance is latterly beginning
to be redressed (e.g. [18]), we propose that this dichotomy is somewhat to blame for the lack of significant advancement of
the field in either direction. This paper outlines how an inappropriate interpretation of Perelson’s shape-space formalism
has largely contributed to this dichotomy, as it neither scales to machine-learning requirements nor makes any operational
distinction between signals and context.
We illustrate these issues and attempt to derive both a more biologically plausible and statistically solid foundation for
an online, unsupervised artificial immune system. By extending a mathematical model of immunological tolerance, and grounding
it in contemporary machine learning, we minimise any recourse to “reasoning by metaphor” and demonstrate one view of how both
research agendas might still complement each other.