Some message for mobile (if needed).

Mal Som is an artist made in Cambodia, planted in Los Angeles, and currently based in Berlin. Growing up in the United States as a first generation Cambodian with Buddhist upbringings, his personal philosophies have married ancient Eastern views with modern Western pop-culture beliefs. This has touched all aspects of his work. Now based in Berlin, he has begun to explore the uncharted realm of AI-generated art. While other AI artists explore the expansive and infinite possibilities this medium represents, Mal has concentrated his focus on something he described as “the collapse of diversity in a system.”

Project Strange Fruit
Within AI-generated art, there is a phenomenon known as “mode collapse.” It is not fully understood why this happens, only that it is a constant occurrence. It is the only constant in latent space. In this project Strange Fruit, Mal explored mode collapse in depth. “I train the model until a full collapse happens. I iterate on the dataset with augmenting, duplicating and looping in generated images from previous ticks to steer the collapse. And then I took a few ticks backward and that’s when I discovered partial collapse.”

Except from Mal
A partial collapse is where a neural network (GAN) is only capable of generating a small subset of outcomes. Degrees of mode collapse happens when the dataset is limited in diversity (too similar) or there are too few images. Another way of wording it: the less diverse your dataset, the quicker mode collapses occur.

What’s interesting about partial mode collapse is, I think it's the machine's way of saying “Hey bruh, after all this training, this is my conclusion, this is what I’ve learned.” Kind of like saying the answer to life is 42, after decades of processing. lol. I want the model to be predictable, the outcome to be predictable and to have more control of the latent space. So if the only constant is mode collapse, if I know where it collapses, I have a baseline. I want to be precise. I want to know what it’s going to be outputted, we all do. I want the images to be cute ^_^

There’s a philosophical side I find interesting as well, I choose to imagine the “moment” before mode collapse as the moment before the death of diversity. The last breath of the model, like it’s last words. There’s this russian novelist who coined the word defamiliarization -- the artistic technique of presenting to audiences common things in an unfamiliar or strange way so they could gain new perspectives and see the world differently.

To me art is about prolonging a movement in time, this moment in the latent space. It slows us down from our automations, it gives us back time to fully appreciate the process of a dying GAN.