Neural networks and applications in scientific research

COURSE SUPPORTED BY THE NON-PROFIT FOUNDATION FOR THE DEVELOPMENT OF SCIENCE AND EDUCATION "INTELLECT"
Neural networks and applications in scientific research
COURSE SUPPORTED BY THE NON-PROFIT FOUNDATION FOR THE DEVELOPMENT OF SCIENCE AND EDUCATION "INTELLECT"
Neuro Collection
Introducing the world’s first collection of brooches created by a collaboration between protein artists (humans) and neural networks! As part of this collaboration, artists use generative neural networks, which, like the brain, create new objects and images based on observations of the world around them.

Introducing the world's first collection of brooches created by a collaboration between protein artists (humans) and neural networks! As part of this collaboration, artists use generative neural networks, which, like the brain, create new objects and images based on observations of the world around them.


How it started

Some time ago we had the idea of an innovative project, a kind of art-experiment in the direction of Art & Science. This project, which we implemented together with the MONOLAMA jewelry brand, was called "Code and Prejudice", and the collaboration itself was defined by the word neuroavant-garde.


Concept

All people think differently. Different cognitive abilities, logiс, experiences, priorities. All influence the formation of human identity, which is at the core of both sociology and philosophy.

We explore how technology can help us understand "eternal" (being, consciousness, eternity) and relevant things today (objectification, gender).

Humans are subjective and biased. With the help of technology we dissect human bias*, stereotypes and prejudices. We try to self-reflect through the prism of human cognition by a system of zeros and ones.

* Bias is a tendency or prejudice in favor of or against one person or group of people.

How it started
Some time ago we had the idea of an innovative project, a kind of art-experiment in the direction of Art & Science. This project, which we implemented together with the MONOLAMA jewelry brand, was called "Code and Prejudice", and the collaboration itself was defined by the word neuroavant-garde.

What did we do?
First, we gave the VQGAN+CLIP neural network a textual description (e.g., "self-awareness," "mental health," "sexism," "friendship"), from which the image was generated. The neural network summarized the data uploaded into the web about the proposed concept and created a unique image. Thus, we obtained an image of a human stereotype.
What did we do?

First, we gave the VQGAN+CLIP neural network a textual description (e.g., "self-awareness," "mental health," "sexism," "friendship"), from which the image was generated. The neural network summarized the data uploaded into the web about the proposed concept and created a unique image. Thus, we obtained an image of a human stereotype.

Then the resulting images were used to select sketches for future jewelry. The next step was to stylize the image so that on its basis it was technologically possible to draw it in a graphic editor and then make the jewelry. After that, Jane, the founder of the MONOLAMA brand, selected materials, made final changes and, finally, created the jewelry.

In this video you can see the main steps of jewelry creation - from image generation to the final result.
Image generated by the query "mental health"
Drawn sketch
Finished brooch
Image generated by the query "sexism"
Drawn sketch
Finished brooch
Then the resulting images were used to select sketches for future jewelry. The next step was to stylize the image so that on its basis it was technologically possible to draw it in a graphic editor and then make the jewelry. After that, Jane, the founder of the MONOLAMA brand, selected materials, made final changes and, finally, created the jewelry.
In this video you can see the main steps of jewelry creation — from image generation to the final result.
Concept
All people think differently. Different cognitive abilities, logiс, experiences, priorities. All influence the formation of human identity, which is at the core of both sociology and philosophy.

We explore how technology can help us understand "eternal" (being, consciousness, eternity) and relevant things today (objectification, gender).
Humans are subjective and biased. With the help of technology we dissect human bias*, stereotypes and prejudices. We try to self-reflect through the prism of human cognition by a system of zeros and ones.

* Bias is a tendency or prejudice in favor of or against one person or group of people.

Image generated by the query "self-awareness"
Drawn sketch
Finished brooch
How does VQGAN+CLIP work?
Around April 2021, Katherine Crowson and Ryan Murdoch began experimenting by combining the CLIP model (from OpenAI) with different GAN architectures. Katherine Crowson, an artist and mathematician, wrote a Google Colab notebook in which she combined VQGAN + CLIP.

Actually VQGAN+CLIP is two neural networks combined into one. Input is a text description — output is a picture. VQGAN is a generative transformer-based adversarial network (Esser et al., 2021). It is used to generate an image — random noise is fed to the input of the network, which is transformed into an image by the neural network. VQGAN itself cannot understand human speech (text). To somehow guide the generation process, we can use CLIP (Radford et. al, 2021). CLIP looks at what pictures VQGAN generates and compares those pictures to a given textual description.

In one sentence: CLIP directs VQGAN to the picture that best matches the given text.


Unusual brooches created using neural networks are available on the MONOLAMA website.