Towards Mode Balancing of Generative Models via Diversity Weights

Sebastian Berns, Simon Colton, Christian Guckelsberger

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

22 Downloads (Pure)

Abstract

Large data-driven image models are extensively used to support creative and artistic work. Under the currently predominant distribution-fitting paradigm, a dataset is treated as ground truth to be approximated as closely as possible. Yet, many creative applications demand a diverse range of output, and creators often strive to actively diverge from a given data distribution. We argue that an adjustment of modelling objectives, from pure mode coverage towards mode balancing, is necessary to accommodate the goal of higher output diversity. We present diversity weights, a training scheme that increases a model's output diversity by balancing the modes in the training dataset. First experiments in a controlled setting demonstrate the potential of our method. We discuss connections of our approach to diversity, equity, and inclusion in generative machine learning more generally, and computational creativity specifically. An implementation of our algorithm is available at https://github.com/sebastianberns/diversity-weights.
Original languageEnglish
Title of host publicationProceedings of the 14th International Conference on Computational Creativity (ICCC 2023)
EditorsAlison Pease, Joao Miguel Cunha, Maya Ackerman, Daniel G. Brown
PublisherAssociation for Computational Creativity
Number of pages10
ISBN (Electronic)978-989-54160-5-9
Publication statusPublished - 2023
MoE publication typeA4 Conference publication
EventInternational Conference on Computational Creativity - Waterloo, Canada
Duration: 19 Jun 202323 Jun 2023
Conference number: 14
https://computationalcreativity.net/iccc23/

Conference

ConferenceInternational Conference on Computational Creativity
Abbreviated titleICCC
Country/TerritoryCanada
CityWaterloo
Period19/06/202323/06/2023
Internet address

Fingerprint

Dive into the research topics of 'Towards Mode Balancing of Generative Models via Diversity Weights'. Together they form a unique fingerprint.

Cite this