Deep architecture is my master thesis project where I explored machine learning and neural networks and how they can be used in an architectural framework.
The thesis was done in two parts one on theory and the other consits of design studies, in the theory part I explored how machine learning works and what parts it consists of. In the design studies a pipeline process is created that showcase different machine learning methods and techniques that are applied to common architectural practice methods.
This project is done entirely in a digital space with no context or site specifications, this is by choice to focus on the techniques and methods behind the process and not to create a “final project” that the techniques would be applied onto.
In the design studies four steps were done the first is dividing a building footprint into smaller more easily handled pieces, the second step in the process is to spatialy organize these pieces. It does this by using a neural network trained on understanding and generating new floor plans. These floorplans are then used in the third step which consist of creating a facade, this is done in a similar way as the floor plans but a similar network is then trained on facades and the logic behind those. The final step is to link both the floor plan and the facade generated from the networks and from these extrude the generated plans into a 3D-model. After a 3D-model has been generated a nural network trained to copy common rendering techniques is applied to create a rendering of the model.
The whole codebase for the project is uploaded on GitHub as to provide an easy way for other students or practitioners to continue working on it.
- Year 2020
- Location Digital
- Status Master thesis
- Institution Chalmers
- Type Digital / Machine learning / Student project / Programming