ARTIFICIAL INTELLIGENCE, NEURAL NETWORKS

Artificial intelligence is essentially a statistical approach to problem-solving. At the core of many AI models are Artificial Neural Networks (ANN) – a mathematical model consisting of layers of interconnected nodes, similar to the neurons in the human brain. These networks can learn patterns and relationships from trained datasets, enabling them to perform tasks without explicit programming; it’s about programming the unprogrammable. Unlike most generative tools that have to start solving the problem from the beginning every time you run them, once you’ve trained an AI model, you can always use it for instant problem-solving. It belongs to intuitive digital tools that have a higher autonomy in decision-making.
Neural Network models are great for:

– predictions: predicting outcomes, such as construction costs or building performances;
– suggestions: predicting which input parameters to use to achieve specific outcomes;
– image creation: exploring ideas.

CREATION

The integration of Artificial Intelligence in architecture is paving the way for innovative approaches, with a current focus on image generation. Using these tools, architects can transform sketches or basic models into lifelike realistic renders and explore different ideas suggested by AI. Additionally, we are working on new models that can be trained on various geometric operations for the generation of precise 3D models.

PREDICTION

We can use simple neural networks to predict the performance of a building (gross area, facade area, volume, material usage, cost…) with a high level of precision. To achieve this, we can either work with your dataset or create a new one from the parametric model. Once you have a clean dataset and a trained model, you can set input values and predict the outputs in less than 0.5s, saving a lot of time in the decision-making processes.

SUGGESTION

We can also utilize neural networks for inverse predictions. This implies that we can specify the desired outputs (gross area, height, volume…) and the algorithm will attempt to determine what inputs (X and Y dimensions, rotation…) should be used to achieve the required goals. It is a more complex problem to solve compared to the direct prediction of building performance, as different combinations of input parameters can lead to similar results.

DATASETS

We can generate datasets for you. When we create a parametric model, we can instruct the computer to randomize input parameters, perform all the necessary computation, and store both input and output values in a CSV file. We can also export images of your design. The computer will repeat this process more than 10,000 times. Although this may take several days, you will have a dataset to work with and learn how to create your own AI models.

Loading...