IEEE Computational Intelligence Magazine - February 2022 - 20

processing (such as occipital cortex) to higher hierarchical levels
(such as pre-frontal cortex (PFC)) where the information
processed by lower hierarchical levels is integrated [30], [31].
The three main DCN frameworks, namely 1) Maturational
perspective, 2) Interactive Specialization (IS), and 3) Skill learning,
aim to answer the question of how these optimized, hierarchical
networks emerge during postnatal development. For the
purpose of this work, we will focus on the IS perspective,
which is largely supported by DCN studies [32]. The IS framework
proposes that both feed-forward and feedback connections
between different cortical regions affect the functional
specialization of cortical regions [33]. More specifically, the IS
theory provides a description of the following three major processes
that occur in the developing brain:
i) Localization: The extent of cortex activation for a given task.
ii) Specialization: The extent of functionality achieved by a
given cortical area.
iii) Parcellation: The optimization of synaptic connections of
neural circuits.
The IS framework suggests that functional brain development
is a dynamic process with localization, specialization, and parcellation
processes forming a continuous loop of development as
shown in Fig. 2. As a given cortical area gains more structural
maturation, its specialization for a given task increases, which
then triggers the parcellation (optimization) of information flow
in the cortical network formed to subserve that given task.
Optimization can take place because of structural and/or
functional maturation (i.e., the emergence of capabilities
through growth processes) of different parts of the brain, along
with more long range connections coming 'on line'. As a
result of the parcellation process, not all parts of a given cortical
region need to be activated nor are all connections
required to transmit the information to the next level of
processing. In this sense, parcellation takes place both within
and between cortical regions. The increased segregation of
information pathways gives rise to increased specialization
(i.e., a modular structure), thus leading to the gradual emergence
of hierarchical networks.
An important consideration with regard to the hierarchical
brain is that the interactions between hierarchies at multiple
levels and timescales are not hard-wired, i.e., the coordination
between modules is not fixed [34]. As a consequence, existing
modules could subserve emerging cognitive states through a
reconfiguration of the evolved circuits using neural reuse [35]
process of brain organization. The other two plausible processes
put forward to explain functional brain organization are modularity
and holism [35]. The modular functional brain structure
would imply that for each task there would be largely segregated
cortical circuits with limited overlap, whereas, the holism
organization of the brain suggests that all cortical circuits may
be engaged across all tasks. The idea of neural reuse seems plausible
with respect to optimal usage of existent circuits evolved
for a given cognitive task. In this way, while neural circuits are
modular to some extent with respect to their individual functionality,
neural reuse suggests that they (individual modules)
have the capacity to connect with each other in numerous
configurations to achieve a range of cognitive-behavioral tasks.
The three aforementioned perspectives of functional structure
of the brain are illustrated in Fig. 3.
Taken together, the IS framework and neural reuse perspective
can shed light on functional brain development at a given
time point using cross-sectional DCN studies. However, a
major component of functional brain development that is still
to be accounted for is the associated temporal information, i.e.,
at what time the developmental changes are happening [34].
Clearly, all stages of functional brain development are not the
4
4
2
3
1
5
6
2
3
1
5
6
2
3
1
4
5
6
(a) Modular
(b) Holism
(c) Neural Reuse
FIGURE 3 Three different perspectives for the functional structure of the brain are illustrated with cortical modules (denoted by circles numbered
1 to 6) and interconnections between the cortical modules (shown in green for task Y and cyan for task Z). (a) The modular structure suggests
that different modules will be activated for different tasks with limited overlap of modules. This is shown by largely segregated modules activated
for task Y (i.e., the modules 1, 2, and 3 activated for task Y) and task Z (i.e., the modules 2, 4, 5, and 6 activated for task Z). As such the modules
engaged are largely distinct with limited overlap (of module 2 in this hypothetical case) for the two separate tasks Y and Z i.e., brain functional
organization is local. (b) In holistic organization, all local brain regions are presumed to be involved for all tasks. Hence, all modules are activated
for both task Y and task Z. (c) According to the neural reuse perspective, the brain works as a whole but functional differences can be appreciated
owing to their reconfigured interactions between the same modules (use and reuse).
20 IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | FEBRUARY 2022

IEEE Computational Intelligence Magazine - February 2022

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - February 2022