Video: Machine Learning and Architectural Design – Victoria Akpezi

Victoria Akpezi is a BIM Manager at ATO Architects and a facilitator at ATO Studios. She is a machine learning enthusiast and a product designer who also specializes in computational design. She contributed to the Enyo competition in 2020 and was among the top 10. She is constantly fascinated by what people can do with technology, which she considers to be the core of all she does. She explores Machine Learning, Artificial inteligence and Deep learning, taking us through the machine learning process. Click on the video below to watch her presentation.

A lot of people in our industry, especially in Nigeria have become complacent with technology.

What does it mean to be a creative?

You are in this profession because you identify as a creative person. Creativity is moving from a position where there is a known problem and an unknown solution because it is your job to created a solution to that problem. This is why you should be a expert at moving from the unknown to the known.

However, we have the opposite attitude when it comes to technology because it seems too daunting and complex to gain our interest. As creatives, we ought to be examples when it comes to the application of technology especially in an industry like ours. 

Machine Learning, Artificial Intelligence and Deep Learning – the link

Artificial Intelligence ( AI) is an entire field which involves trying to get machines to think like humans. The goal is to get to a point where the machine can make decisions the same way humans would.

Machine Learning is a subset of Artificial Intelligence that focuses on extracting patterns from large amounts of data. I.E to get patterns using algorithms in a way that the human eye cannot see as a result of the complexity of data.

Deep Learning is a subset of machine learning that focuses on a specific type of algorithm which mimics the human brain. Just like the human brain passes information using neurons, Deep Learning passes information through neural networks.

Some machine learning apps we currently know and use are:

• Siri
• Alexa
• Google Translator
• Facebook
• YouTube
• Spotify and other auto music players 

Types of Machine Learning Models

There are 4 main types of machine learning models:

• Supervised
• Semi-supervised
• Unsupervised
• Reinforcement 

The supervised and semi-supervised models have currently gained more ground in Architectural design practice.

Supervised Learning Model

This works with a bunch of unorganized data passes through an already determined sorting process. The architect gives the machine a specified training set and instructs it to sort others (from the data) according to the training set that was provided.

Semi-Supervised Learning model

Semi-supervised learning is a bridge between supervised and unsupervised learning. With Unsupervised Learning, the system is fed with a bunch of data that is without a known pattern. The machine finds the patterns and groups the data accordingly.

Semi-supervised learning is halfway between supervised and unsupervised learning. The machines learns under the architects supervision until it masters the patterns and can work alone.

Neural networks

As mentioned earlier, deep learning works with neural networks that help the machine circulate and process information. The most commonly used neural network is called a Generative Adversarial Network (GAN). The GAN has a generator and a discriminator. The generator passes raw data to the discriminator which determines if the data matches the defined pattern or not.

How does a machine learn to create a building?

She explains this using a project by Mr Stanislas Chaillou, a student from Harvard Graduate School of Design.

1. Train the machine by giving it a several-hundred good floorplans (they meet the criteria for good design) so that it can understand the patterns that exist.

2. Teach it about orientation – for instance, how the sun position affects the shape a building should have.

3. Train it to understand furniture layout and orientation – The furniture contained in the floor plan has been grouped according to colours. The machine learns to understand the relationship between circulation and furniture so it can start to extrapolate this pattern.

4. Teach is to understand connectivity and program – how does the door and window relate to the position of the bed? Help the machine to understand the connection of various elements and consider them the way a human would.

5. Use colours to represent these graphical elements we learn – As architects, the way we see a floor plan (to the human eye) will be doors, windows, furniture etc. but the computer will represent these elements in machine language it can understand. In this case – it uses colours. There will be a graphical interface that translates the colours to components that make sense to a human being. The machine understands the rules of a colour block but the human understands the floorplan.

6. Generate options – You have taught the computer how to think about design and understand how spaces work. Now, you give it a footprint and watch it generate options that you want.

Tools to get started

1. Pix2pix – turns any sketch you give into a cityscape on Venice. It also renders to other objects like bags, shoes etc.

2. Fusion360 – Uses generative design to help you optimize materials by identifying the precise area where forces act on objects so that you can maximize materials.

3. Rhino+grasshopper – An example from the BIM Africa Innovation Awards. You set up parameters and let the system generate options.

4. Revit generative design – Works with the constraints you provide to generate options.

Conclusion

Machines have become part of the workforce. What will your response be as a creative architect?

CREATIVE ARCHITECTS 2021 SPONSORS

SwiftWall Solutions 

PrintMyDrawings 

The Unreal Place

Glass.com.ng

MDAD ATELIER 

Primestack Motivo

Rygdesigns