Information Technology Reference
In-Depth Information
In order to allow a brain-computer music interface (BCMI) to interact with, and
allow control of musical stimuli, it is
first necessary to understand the relationships
between music, the emotion(s) induced or perceived, and neurological activity. For
this purpose, it is necessary to identify both neural correlates of the emotions induced
by particular musical stimuli and neural correlates of the perception of music.
However, the feature space that may be extracted from
—
the brain may be very large, complex, and high dimensional. For example, the
electroencephalogram (EEG) provides a time series of discrete measures of elec-
trical activity recorded from the surface of the scalp and, due to high sampling rates,
may be very large and described by a multitude of features. Similarly, functional
magnetic resonance imaging (fMRI) provides a detailed three-dimensional measure
of oxygen consumption by neurons throughout the brain resulting in a much higher-
dimensional time series, which may also be described by a multitude of features.
Therefore, there is a need for advanced analysis methods to uncover relation-
ships between cognitive processes and their corresponding neural correlates, which
may not be immediately apparent.
Machine learning describes a set of methods which attempt to learn from the data
(Alpaydin
2004
). For the purposes of brain
—
and used to describe
computer interfacing and the study of
the brain, this commonly takes the form of learning in which brain activation
patterns are associated with particular cognitive processes or differentiate groups of
processes.
Machine learning provides a data-driven approach to understanding relationships
between neurological datasets and their associated cognitive processes. For
example, it may allow the uncovering of complex neural correlates of speci
-
c
emotions or music perception, which are not
immediately apparent via other
analysis techniques.
This chapter provides an introduction into the use of machine learning methods
in the context of identifying neural correlates of emotion and music perception. We
first introduce models of emotion and empirical measures of emotional responses,
which are required by machine learning methods to allow training on labelled data.
We then go on to review features which may be extracted from neurological data
and audiological signals to describe relevant or interesting properties. Finally,
machine learning methods are described and examples are provided to illustrate
their use in uncovering neural correlates of emotion and music perception.
5.2
Measuring Emotion
5.2.1 Models of Emotion
Models of emotion differ signi
cantly in the way they conceptualise the determi-
nants of the emotional experience, the way they envision the neural mechanisms
that give rise to this experience, and the predictions they formulate for the
accompanying symptoms. Particularly, the landscape of emotion theories spans
