graffik

=graphic design±research

janhenning.raff@gmail.com
@j__________r___ <— instagram
prof. for visual communication~~media-university.de
¬ Berlin, Germany

Dear reader~viewer, I am still (late 2024) doing basic research on layout in graphic design.

This is a composition generator that has embodied spatial categories as input. Participants were asked to evaluate random compositions in an online survey. From the answers, I derived some interactive generators:

I also trained a convolutional neural network (CNN) with the data:

Only 4 edge filteres were the input for a fully connected network. The second visualization is an “activation map” representing the amount of horizontals and verticals in a predefined grid.

Training data was scarce. This is one example with a lot of congruence between humans and machine. We can observe that embodied apprehension of a composition (e.g. “this looks heavy”) is relying on just a few features.

But also the not fitting results are of interest – see static-moving:

What do we learn from pattern learning of machines? How is it similar, how is it different from embodied human vision?

Read more about my research on spatial arrangement.

Using machine learning for visual analysis

I use machine learning to analyze visual communication – I talked and wrote about it: Machine Learning for Basic Visual Research in Graphic Design.

Here is an example: all posters from the competition “100 best Posters” from 2001–2023 sorted by main color:

Check the interactive 3D version: 100 beste Plakate 2001-2024 sorted by main color.

I also explore how vision is conceived in machine learning and deep learning within the MA program AI & societies of Media University.

Now feel free to browse design±research.

… or are you looking for noise?