Information Technology Reference
In-Depth Information
Chapter 1
Introduction
From our most early experiences with reality, we start to recognize patterns in the
surrounding environment. This allows us as human beings to be aware of the
different objects that we are related to. The scope of pattern recognition is broad
since it is observed at different levels in the world. This awareness occurs for a cell
that divides and specializes itself and for an expert standing in front of a painting
trying to make a distinction between the pure object and the pure subject of that
object. This natural skill of living, which is the basis of the learning process, is
artificially imitated by means of sequences of mathematic-computational steps
known as machine learning. Without going into an epistemological discussion
about whether or not the perceived reality is biased by our senses and capabilities
to process what is perceived, the high complexity of the tasks for machine learning
represents an actual challenge. The questions of why and what we learn for, which
are generally related to adaptation for evolving, are outside the scope of this work.
Instead, we will focus on how to learn artificially, and we will propose mathe-
matical procedures that are able to distinguish, represent, and allocate learning
objects, as well as assess these procedures in novel application fields.
Learning can be defined as the process of inferring general rules from given
examples. To obtain an artificial approach to the learning process, we have to
simplify the representation of the objects of the real-world whose patterns must be
recognized and/or learned. Afterwards, a way to search for data of interest in these
objects must be proposed. The data of interest will usually have (or we attempt to
find) a certain meaning or interpretation that allocates them to a specific field of
application. Thus, real objects or phenomena are represented as numbers or data,
i.e., collections of ones and zeros for computer processing. Data are obtained from
measurements applied on the real objects by means of sensors. The estimation and
analysis of the distributions and groupings as well as the inference on possible
former generators of these data are critical for pattern recognition.
There are several definitions that have been proposed for pattern recognition
[ 1 ]. In this thesis, we will assume a common accepted explanation of pattern
Search WWH ::




Custom Search