Past research · Datasets and a knowledge graph
Datafying Creativity
Which parts of creative work can be computed, and what must data and knowledge contain for creativity to be identified, evaluated and generated?
This series is the ground beneath the three models. Each project gathers a body of design knowledge into a form a machine can learn from, and in doing so tests where creative work can be computed and where it resists.
The datasets come from places design research rarely looks: the colour cultures of Chinese youth subcultures, the fast-moving world of blind-box designer toys, and traditional crafts such as Jinshan farmer painting. Prometheus goes a step further and constructs design knowledge itself as a knowledge graph that models can reason over.
Together they make three things possible: identifying creativity, evaluating it, and generating creative work.
Structure
- Talks
- The Computability of Design, Brown University (2024); keynote, Hong Kong Polytechnic University (2023)
- Research line
- The Computability of Creativity, Design AI Lab
The series
(a)
Youth-subculture colour dataset
Colour palettes drawn from Chinese youth subcultures, used for culture-inspired multimodal palette generation and colorization (IEEE MIPR 2021).
(b)
Blind-box dataset
A dataset of blind-box designer toys, a consumer form where taste, rarity and collecting meet, used to generate new figures, outfits, props and scenes in 3D. To add: scale of the dataset
(c)
Chinese traditional craft dataset
Traditional crafts including Jinshan farmer painting, used to study how AI can support the inheritance of craft (Decoration, 2022). In an installation built on it, visitors sketch a person, a house or a tree, and the system paints the scene in the Jinshan style.
(d)
Prometheus
A knowledge graph that constructs design knowledge in a form machines can reason over. Ask it a question about design and it unfolds the concepts and sources linked to the answer. To add: what the graph contains and its scale