Maniadakis and Trahanias propose a new approach to designing systems modeled on the mammalian brain, based on multiple components that communicate with each other and cooperatively adapt and evolve. This is in contrast to previous attempts at brain modeling that focus on modeling specific parts of the brain, largely in isolation.

The proposed framework consists of cortical agents and link agents, which represent the neurons that make up a mammalian central nervous system (CNS), and is described in a modular way, where each part is expandable and autonomous, yet cooperative. New axons/neurons--and whole systems of axons/neurons--can be added or removed at any point, making for a very efficient and useful design; this dynamic aspect is due to the authors’ hierarchical cooperative coevolutionary (HCCE) scheme.

To test the HCCE scheme, the authors implement a robot system patterned after the mammalian biological system that learns to take appropriate actions, depending on light stimuli. With this design, brain lesion simulation can be achieved easily, by simply deactivating individual agents. The authors compare the HCCE performance for the problem of wall avoidance with the approaches used in other systems, such as enforced subpopulations (ESP) and unimodal evolution. Test results indicate that the HCCE scheme’s performance is better than the other systems’. This is attributed to large numbers of coevolving brain areas and interconnections, to complete a single complex system that exhibits phenomena found in biological systems. For future brain modeling, this novel method has the potential to accelerate the successful implementation of system components, while allowing for the flexibility to add new components and redesign existing ones.

The paper is a good overview of agent-based methods for brain modeling. Readers should have basic knowledge of artificial intelligence, although the authors do provide explanations of the underlying concepts and methods. Also, even though the authors do not shy away from equations, readers with a weak mathematical background should be able to follow the main ideas. Overall, this idea of agent-based coevolutionary systems for modeling complex systems is very interesting, and is likely to make the reader curious about other potential applications.