Pattern form recognition

The aim of pattern recognition is to design automatic, computerized tools capable of recognizing patterns. Pattern recognition (RDF) is a field that has evolved considerably over the last few decades, to the point where its contours have become increasingly blurred. It's not easy to find a definition that all RDF researchers can agree on.

Bishop reports that it was Tycho Braha's systematic collection of astronomical observations in the 16th century that enabled Johann Kepler to discover empirical laws of planetary motion. We can multiply such examples and go far back into our past.

If we look at the modern history of RDF, let's say since the 1980s, and its recent developments linked of course to the development of computing machines, we can see that it was initially concerned with the automation of perceptual tasks. One of the leading applications in this field was handwriting recognition, which remains an active field of application and research today. Beyond this, vision, through the recognition of objects in images, but also hearing and automatic speech recognition are emblematic and historic applications of pattern recognition.

This is one of the reasons why pattern recognition has long been regarded as a component of the broader, and also highly multidisciplinary, field of artificial intelligence (AI), an ultimate goal of which could be to produce robots endowed with intelligence, capable not only of learning and reasoning, but also of interacting, including physically, with the outside world, i.e. moving, recognizing their interlocutors and objects in a room, speaking, hearing, understanding, reasoning, etc. In this scheme, RDF's initial role was to produce all the algorithms needed for abstract perception of the environment (obstacles, individuals, etc.) from sensors on the outside world.

Gradually, RDF has freed itself from the tutelage of AI. By taking an interest in any shape or pattern, it has made it possible to tackle other classification tasks, and has been extended to other supervised learning frameworks such as regression. Today, RDF is one of the founding disciplines of the machine learning field, along with others such as algorithmics, complexity, cryptography, logic, optimization, statistical physics, probability, cognitive science, statistics, evolutionary theory, etc.

The modern history of pattern recognition is therefore an eventful one. It is, after all, a young discipline which has already undergone many influences and developments, and whose spectrum of applications and techniques has grown within what is now known as digital machine learning. This presentation is intended as an introduction to the field and cannot be considered exhaustive. Its main purpose is to...