Creaition – revolutionizing the design process with machine learning

From the drawing bank to a machine learning research project

As one of the leading consultancies in the area of data science, AI & machine learning, we regularly meet people at STATWORX who make a lasting impression on us with their revolutionary ideas. Many of these people have one thing in common with their product ideas: They want to reduce a personal pain point in their way of working. One of these exciting people with an unusual idea is Marco Limm.

Marco Limm has The start-up Creaition founded, which wants to optimize the extremely complex and very iterative process of design development with its product. But how did he get this idea?

While studying transportation design in the USA, he first came up with the idea of using the design process with the help of Machine learning to optimize. He was motivated by the fact that many of his works were produced, to put it in his own words, “for the bin” and thus spurred him on to find a more efficient way.

The idea of developing his own product, which acts as a digital muse for designers, has been very busy for Marco since the third semester. With the motivation to make work easier for other designers as well, and spurred on by curiosity as to whether his idea could actually be implemented, he developed a research paper from the idea, which he dedicated himself to both his bachelor's and master's thesis.

The development of the creaition prototype

First attempts with 3D models, image data & scarce resources

Marco sees the most important influence of designers at the beginning and end of the classic design process — in his opinion, designers should be able to focus much more on this. The middle part of the design process includes, for example, “drawing car fronts for hours,” says Marco. “The machine can take over the morphology; you don't need a human designer for that. ”

Based on the basic idea of optimization through machine learning, Marco started developing the algorithm together with Kevin German as part of their joint bachelor thesis at the university. The aim was to train artificial intelligence (AI) that independently creates design suggestions based on historical data.

First, the duo tried to work with 3D models that Marco had created himself. In this attempt, the two quickly reached their limits, as they did not have sufficient computing power and infrastructure available at university. Another problem was a lack of data.

In a second step, they tried their hand at images that they obtained from various sources (archives, online, etc.). That worked better already, but the results that the algorithm generated from this data were more art than industrial design. Marco tells us that one of his professors would have liked to publish these results as an art project. “The result was beautiful compositions that could certainly be placed in a museum, but unfortunately that wasn't the real goal. ”

3-layer approach for the machine learning algorithm

From these first two failed attempts, it was possible to draw the conclusion that they needed a less complex but also not too abstract data basis. It went back to the research phase, rummaging through books, examining various approaches, collecting new impressions. As a result, they launched a completely new approach in the third attempt. In this new approach, they divided the design into 3 layers: silhouette, surfaces, graphics. They used the three layers that make up a design to reduce the complexity of the data.

With the 3-layer approach, Marco was able to create a suitable data set and use this data to train the algorithm. The result was 40,000 designs from which they could now choose.

This result gave rise to the next challenge: Evaluating 40,000 designs did not really make work easier. The solution? A genetic algorithm that recognizes individual tastes based on designers' reactions to various design suggestions and generates new suggestions based on this. They called it “bottle Tinder” — because here too, you keep swiping, depending on whether you like the design or not.

With the “bottle Tinder” principle, designers are constantly presented with new design suggestions, which they must then evaluate. After about 15 minutes, the algorithm is able to extract the person's taste and then apply it at will.

Creation on the go at national and international trade fairs

With this prototype in its luggage, the creaition project attracted attention at various trade fairs and received consistently positive feedback there. Your colleagues from the design industry see great potential in their product — who wants to draw car fronts for hours on end?

Creaition was represented at the following trade fairs, among others:

• Salone del Mobile iin Milan (2019)
• Dutch Design Week (2019)
• Association of German Industrial Designers Stuttgart — Intelligence of the Future (2019)
• Pforzheim University, Werkschau (2019)
• Aalborg Denmark, ArtSit (2019)

Next steps: pilot project

We at STATWORX are currently supporting creaition in finding a suitable partner for a pilot project. The aim of the proof of concept is to develop new design drafts based on extracted design DNA from the client.

In concrete terms, this means that the machine learning algorithm is retrained on the basis of your 2D data and filters out a design DNA, which in turn can be used to generate any number of new design suggestions. As a rule, companies that are eligible for this type of AI optimization already have a wealth of data (especially 2D data from their products) from which they have not yet benefited much. So use the potential of your data and help your design department reduce repetitive and monotonous work steps so they can focus on essential, creative and creative work.

Would you like to find out more about Creaition? You can download the full conference paper here, which Marco Limm and his co-authors (Kevin German, Matthias Wölfel and Silke Helmerdig) wrote for ArtSit on this topic.

If you are interested in working with us on this pilot project, please feel free to contact us via our contact form or send an email to hello@creaition.io.