Understanding the Advantages of Modularity in Neural Systems

Understanding, or even defining, modularity in the human brain is not as straightforward as one might expect. It is natural to assume that modularity offers computational advantages, and that evolution by natural selection would translate those advantages into the kind of modular neural structures familiar to cognitive scientists. However, explicit simulations of the evolution of neural systems have shown that, in many cases, it is actually non-modular architectures that are most efficient. In this paper, I present a further series of simulations that reveal a crucial dependence on the details of the tasks that are being modelled, and the importance of taking into account physical brain constraints, such as the degree of neural connectivity. Eventually, we end up finding modularity emerging reliably from evolution across a range of neural processing tasks.