1.5. Overview: What is in it and what is not

This thesis navigates the intricate intersection of type design and machine learning – two domains that, much like parallel lines in Euclidean geometry, rarely meet in meaningful discourse. While each field has its dedicated practitioners and theoretical frameworks, their convergence presents unique challenges in communication and methodology.

While flesh-and-blood researchers undoubtedly conducted this investigation – complete with their caffeine dependencies and occasional moments of despair – the narrative maintains the time-honoured academic tradition of referring to itself in the third person. This thesis thus assumes an almost sentient quality, presenting its findings with the detached authority expected of scholarly works while quietly acknowledging the human fingerprints on every page.

The research offers distinct value propositions for its diverse readership:

For machine learning researchers, it illuminates the nuanced requirements of type design – a domain where success metrics extend beyond conventional loss functions into the realm of aesthetic judgment and centuries-old craft traditions.

Type designers will find methodologies for incorporating machine learning into their practice without sacrificing the ineffable qualities that distinguish exceptional typefaces from mere collections of vectors.

Font engineers gain insights into bridging the technical demands of both worlds – particularly in translating traditional type design parameters into machine-learning-compatible features while preserving their original intent.

Despite its technical foundations, this thesis deliberately eschews unnecessary jargon, offering clear pathways through complex concepts. Like a well-designed typeface, it aims to be both sophisticated and accessible – ensuring that even readers approaching from adjacent fields can navigate its arguments without requiring a doctorate in both disciplines.

The thesis comprises three main chapters, each representing distinct yet interconnected phases of investigation:

Chapter Preliminaries walks the reader through the first part of the studies, with three sections:

1. Exploration and explains the methodology of explorative research of literature collection
2. Systemisation, where the methodology of classification and definition of terminology around the machine learning pipeline
3. Compilation when the thesis explains the methodology and tools of building the database of font generation projects literature.

Chapter Surveys comprises three surveys culminating around crucial topics of type design and machine learning intersection, which investigates:

1. Applications of machine learning techniques in type design through the task which the machine learning models have been trained
2. Datasets – as the mere representation of the type design domain – used in the projects, their size, quality and potential weaknesses for training models
3. Representations, or in other words, exploration of how original font files and their information are prepared for computations of machine learning algorithms.

Chapter Projects wraps the practical part of the dissertation project, consisting of two projects:

1. Regularised type designers dataset presents a reaction to the problem of irregularity of previous font datasets. The project introduces building methods used in a novel manner to generate large, high-quality, regularised font datasets.
2. Demonstration, which aims to validate the premise of dataset regularisation and presents the comparative methodology with the findings of the experiments.

Following the main chapters, the thesis presents a concise summary of findings – a scholarly “TL;DR” if the reader will – that distils key insights into a structured format for those who are hesitant to read the whole story. While this condensed conclusion cannot capture every nuance, it offers an efficient entry point to the research’s core contributions.

The thesis appends with a carefully curated dictionary – a collection of essential terms that serves both as a quick reference and, for the particularly enthusiastic reader, an opportunity to expand one’s vocabulary in the realms of type design and machine learning. While mastery of these terms may not guarantee expertise in either field, it certainly enables more confident navigation of interdisciplinary discussions.