Название: The Shallow and the Deep: A biased introduction to neural networks and old school Machine Learning
Автор: Michael Biehl
Издательство: University of Groningen Press
Год: 2023
Страниц: 294
Язык: английский
Формат: pdf (true)
Размер: 10.6 MB

The Shallow and the Deep is a collection of lecture notes that offers an accessible introduction to neural networks and Machine Learning in general. However, it was clear from the beginning that these notes would not be able to cover this rapidly changing and growing field in its entirety. The focus lies on classical Machine Learning techniques, with a bias towards classification and regression. Other learning paradigms and many recent developments in, for instance, Deep Learning are not addressed or only briefly touched upon. Biehl argues that having a solid knowledge of the foundations of the field is essential, especially for anyone who wants to explore the world of Machine Learning with an ambition that goes beyond the application of some software package to some data set. Therefore, The Shallow and the Deep places emphasis on fundamental concepts and theoretical background. This also involves delving into the history and pre-history of neural networks, where the foundations for most of the recent developments were laid. These notes aim to demystify Machine Learning and neural networks without losing the appreciation for their impressive power and versatility.

Unsupervised learning is an umbrella term comprising various methods for the analysis of unlabeled data. Such data sets do not contain label information associated with some pre-defined target as it would be the case in classification or regression. Moreover, there is no direct feedback available from the environment or a teacher that would facilitate the evaluation of the system’s performance. A comparison of its response with a given ground truth or approximate representation thereof is not available or possible.

In supervised learning, available data comprises feature vectors1 together with target values. The data is analysed in order to tune parameters of a model, which can be used to predict the (hopefully correct) target values for novel data that was not contained in the training set. Generally speaking, supervised machine learning is a promising approach if the target task is difficult or impossible to define in terms of a set of simple rules, while example data is available that can be analysed.