What is cognitive neuroscience?

Neuroscience is a physical science -- it seeks to understand physical mechanisms of

the nervous system.

Cognitive neuroscience is the branch of neuroscience that seeks to understand the

mechanisms of the nervous system that are directly related to cognitive (mental)
processes. These mechanisms are thought to reside in the brain. Because cognition
refers to functions of the mind, we must begin our study of cognitive neuroscience by
first examining the relation between the mind and the brain.

The question of how the mind and brain are related is called the mind-brain problem.
A matter of correlation

We take a functionalist approach to the study of cognitive neuroscience. This

reduces the mind-brain problem to the computational mind-brain problem. But we
are still left with the problem that theories of the computational mind in cognitive
science and theories of the brain in neuroscience represent two independent
systems of description. Cognitive neuroscience has not developed to the point
where it has established causal relations between cognitive and neural phenomena.

All science undergoes a natural progression from observation to correlation to

causation. Cognitive neuroscience is largely still at the stage of correlation.

Most of the work in this field aims to determine:

1) the neural structures that carry out cognitive functions, and
2) the neural mechanisms by which the structures carry out the cognitive functions.
The cognitive neuroscience triangle (Kosslyn & Koenig 1992)

This concept is useful in the study of cognitive neuroscience. Cognitive

neuroscience attempts to establish correlations between cognitive phenomena and
neural phenomena, using 3 major domains:
(a) cognition (behavior & models)
(b) brain (neurophysiology & neuroanatomy)
(c) computation (analyses & models)
Emergence of the distributed paradigm in neuropsychology

To understand the difference between the modular and network paradigms, it is

necessary to examine the history of understanding the relation between brain
function and cognition. This history has been dominated by two parallel trends:
localizationism and globalism, and has led to the emergence of a compromise
paradigm, the distributed paradigm.
Localizationism vs globalism

Historically, there has been a controversy for about 200 years in neuropsychology
over the question of whether different mental functions are carried out by different
parts of the brain (localizationism) or the brain works as a single, integrated whole
(globalism).
Phrenology

In the 17th & 18th centuries, the theory of faculties was dominant in psychology.
All psychological processes were understood as "faculties" of mind, incapable of
further subdivision.
In 1796, Franz Joseph Gall began measuring bumps on the heads of Viennese
residents. He postulated that the brain is a collection of centers corresponding to
specific "faculties".
He thought that even very elaborate & abstract functions e.g. cautiousness,
generosity, hope, were discretely localized to single areas of cerebral cortex.
Cranial bumps were thought to reflect development of cortical area underneath and
consequently the corresponding mental trait.
Incorrect assumptions:
a) cognitive functions are implemented by discrete cortical regions
b) development of a cognitive function increases the size of its region
c) enlargement of cortical regions causes expansion of the outer cranial surface

Correct assumptions:
a) mental abilities can be specified and analyzed
b) the cerebral cortex is important for mental ability
c) the brain is not a single, undifferentiated system
Globalism

Phrenology was criticized by Pierre Flourens (1824) who found that mental functions
are not localized, but that the brain acts as a whole for each function.

The Paris Academy of Sciences commissioned him to investigate the claim of Gall
that character traits are localized in specific cortical regions.

He studied the effects of brain lesions on the behavior of pigeons.

The pigeons could recover after parts of the brain were removed, regardless of the
location of the damage.

He concluded that the major brain divisions are responsible for different functions.
Cerebral cortex: perception, motricity, judgment
Cerebellum: equilibrium, motor coordination
Medulla: respiration, circulation

However, he found no localization of cognitive function within the cerebral cortex. He

concluded that the cortex has equipotentiality for cognitive function: lost function with
ablation does not depend on the location of damage, but only on the amount of tissue
lost.
The controversy continues in the 19th and 20th centuries

Some clinical evidence suggested localization of function:

In 1861, Paul Broca showed that a lesion of the posterior third of the left inferior
frontal gyrus causes a motor speech disturbance without affecting understanding of
speech. He believed that the "motor images of words" are localized in this part of the
brain.

In 1874, Carl Wernicke described a patient who had difficulty comprehending speech
after damage to the left superior temporal gyrus. Knowing of Broca’s work, he also
distinguished elementary from complex function in language.

Friedrich Goltz - 1881 – was the major opponent of localizationism of his time; he
postulated that brain works as a whole.

During the 1st half of 20th century, several influential neuroscientists continued to
advocate globalism. Karl Lashley was most important. He proposed two principles of
brain function:
a) mass action: the brain works as a single system
b) equipotentiality: all parts have equal ability to perform different tasks
He based his ideas on a long series of experiments to try to find the locus of learning
by studying maze learning in rats with various brain lesions. He declared that brain
function is widely distributed because he couldn't find such a locus. He concluded
that only the extent of damage was important, not the location.

Other studies, notably those using electrical stimulation of exposed cortex in awake
patients by Wilder Penfield & colleagues, continued to provide evidence of
localization.
Summary of distributed view

The distributed view was clearly articulated by Alexander Luria (1975): "The higher
forms of human psychological activity and all human behavioral acts take place with
participation of all parts and levels of the brain, each of which makes its own specific
contribution to the work of the functional system as a whole."

1) Elementary functions are localized, but the brain works in a distributed manner to
produce complex functions that are not localized.

2) Complex functions are carried out by distributed combinations of simple functions.

The simple functions are localized in many different places in the brain. They can be
carried out by different elementary functions at different times, allowing them to be
performed in different ways. Thus, different "strategies" can be implemented as
different combinations of simple functions.

Resolving elements of localizationism and globalism, the distributed view has

evolved into the modern network paradigm in cognitive neuroscience.
The concept of “neural network” in cognitive neuroscience

Neuroscience has been very successful at explaining the neural basis of low-level
sensory and motor functions. These functions rely on the input and output systems
of the nervous system, where discrete structural modules represent elemental
sensory and motor components. This success has led to a reliance on modular
explanations of brain function.

However, this modular paradigm fails to explain essential cognitive functions such
as perception, attention, or memory. The modular paradigm attempts to assign
specific cognitive functions to individual brain modules. One problem with this
approach is that it assumes that the different cognitive functions are separate
entities.

This assumption is adequate for the cognitive psychologist, i.e. cognitive functions
may be conceived as being distinct at the psychological level. However, it does not
necessarily follow that these functions have separate neural substrates. The
assumption that there is a cortical module for every cognitive function has caused a
great deal of confusion in cognitive neuroscience. The concept of networks provides
a vital alternative to the modular paradigm.

The network paradigm has taken centuries to develop. Even now it is not universally
accepted, but its acceptance is rapidly growing.
The concept of “neural network” in computational cognitive
neuroscience

To understand the computational aspects of the network paradigm requires

examining the history of the concept of “neural network” in the field of artificial
intelligence. The modern history of artificial intelligence can be traced back to the
1940's, when 2 complementary approaches to the field originated.

The Serial Symbol Processing (SSP) approach began in the 1940's, when the
architecture of the modern digital computer was designed by John von Neumann and
others. They were heavily influenced by the work of Alan Turing on finite computing
machines. The Turing Machine is a list of instructions for carrying out a logical
operation.

The von Neumann computer follows this theme. It:

a) performs one operation at a time
b) operates by an explicit set of instructions
c) distinguishes explicitly between stored information & the operations that
manipulate information.
The Parallel Distributed Processing (PDP) approach (also called connectionism) may
also be traced to the 1940’s.

In 1943, Warren McCulloch and Walter Pitts proposed a simple model of the neuron
– the linear threshold unit. The model neuron computes a weighted sum of its inputs
from other units, and outputs a one or zero according to whether this sum is above or
below a threshold.

McCulloch & Pitts proved that an assembly of such neurons is capable in principle of
universal computation, if the weights are chosen suitably. This means that such an
assembly could in principle perform any computation that an ordinary digital computer
can.

In 1949, Donald Hebb constructed a theoretical framework for the representation of

short-term & long-term memory in nervous system.
The functional unit in Hebb's theory is the Neuronal Assembly: a population of
mutually excitatory neurons that when excited together becomes functionally linked.

Hebb also introduced the Hebbian learning rule: when unit A and unit B are
simultaneously excited, the strength of the connection between them is increased.

The perceptron, a single-layer network of linear threshold units, was developed by

Rosenblatt in the late 1950’s.

Rosenblatt developed a learning algorithm – a method for changing the weights in the
perceptron iteratively so that a desired computation was performed. (Remember that
McCulloch & Pitts had proposed that the weights in their logic circuits had to be
appropriate for the computation.) Rosenblatt believed that multi-layer structures could
overcome the limitations of the single-layer network.
The PDP approach has gained a wide following since the early 1980's. Many
neuroscientists believe that it embodies principles that are more neurally realistic than
the SSP approach. Because PDP models are thought to work like brain regions, they
are often called artificial neural networks.

Recent interest in artificial neural networks has focused on deep learning, with
applications in many fields, such as computer vision, speech recognition, natural
language processing, social network filtering, machine translation, and drug design,
and with results that are in some cases superior to human experts.