“Perplexity in t-SNE ML” is a study on T-distributed stochastic Neighbor embedding. It is uses on Non-linear multi-dimensional data representation in 2D to allow collaboration between designer and algorithm. A designed facade typology data set is fed to the machine to develop their own understanding of the image-based cultural representation. The play with size, resolution, perplexity, and iteration values throws different outputs to be voxelized as facade assemblies.

Made with Python and Processing using machine learning workflows.

Type:    Workshop
Institution: SoomeenHahm Design Platform
Instructor: Sanghyun Suh
Research:    Machine Learning, t-SNE, Python,
   Processing, 3D Voxelization.

Automated Facade Iterations:
1. --size 39 --res 256--per 50 --iter 5000
2. --size 39 --res 256--per 150 --iter 25000
3. --size 32 --res 256--per 150 --iter 25000
4. --size 19 --res 256--per 150 --iter 5000
5. --size 21 --res 256--per 150 --iter 25000
6. --size 39 --res 256--per 150 --iter 5000
7. --size 39 --res 256--per 50 --iter 5000
8. --size 42 --res 128--per 100 --iter 5000
9. --size 30 --res 128--per 100 --iter 5000

Carlos Navarro