moc.snrebnaitsabes@s
Diversity in Generative Machine Learning to Enhance Creative Applications.
PhD thesis by Sebastian Berns.
Queen Mary University of London, July 2024.
Contents
- Introduction
- Background
- Generative Modelling in Deep Learning
- Generative Models as Creative Systems
- Use of Generative Models with Evolutionary Algorithms
- Evaluation of Generative Models
- Measuring Output Diversity
- Human Perception of Similarity
- Artistic and Creative Uses of Generative Models
- Introduction
- Automating Generative Deep Learning for Artistic Purposes
- Discussion
- Limitations of Conventional Generative Modelling
- Introduction
- Artefact Generation via Latent Space Search
- Methodology and Setup
- Experiments
- Results
- Discussion
- Increasing the Output Diversity of Generative Models
- Introduction
- Mode Balancing
- The Vendi Score
- Diversity Weights
- Proof-Of-Concept Study on Hand-Written Digits
- Discussion
- Similarity Estimation for the Evaluation of Diversity
- Introduction
- Methodology
- Study 1: Human vs. Computational Similarity Evaluation
- Study 2: Interpretation of Similarity Dimensions
- Discussion
- Related Work
- Data Biases in Machine Learning
- De-biasing Generative Models
- Conclusions
- Future Work