Perceptrons is an experimental collection of on-chain AI models.
While many projects have stored outputs from AI models on-chain, Perceptrons attempts to store the actual AI models themselves, allowing users to query the artwork and run live image recognition tasks.
How do Perceptrons work?
Perceptrons are built of neural networks that take images as inputs and classifies them.
Neural networks are mathematical functions which take a variable as an input, combine it with certain parameters and return an output. Parameters are estimated by going through large amounts of previously categorized data.
In Perceptrons, many different neural network designs are generated. The parameters of those networks are then estimated by going through the training data, and the resulting model is uploaded to the Bitcoin blockchain through inscriptions.
Collectors will be able to interact with their Perceptron piece through a dedicated front-end developed by Generative.xyz.
Perceptrons also change over time; decaying, dying and being reborn. In the future, collectors will also be able to tinker with the underlying models of their Perceptrons, thus manipulating the artwork.
Who made the artwork?
Part of the Generative team, consisting of artists 1605, 8745, and 131pow (with the help of 3700 and many others).
Generative is a Bitcoin inscription platform and marketplace founded by a former Fingerprints member, alongside many developers, artists, designers, writers, and technologists.
The neural network code of a Perceptron occupies about 400kb.
The cost of permanent storage in the Ethereum blockchain is about $50/kb as of launch, which would make storing the models directly in the Ethereum blockchain prohibitively expensive.
Through ordinals inscriptions (made possible in the Taproot Bitcoin upgrade), storage cost is significantly lower, at $0.05/kb, thus making the project feasible.
For more information on how it is possible to store this kind of data directly on the Bitcoin mainnet, please refer to this link.
Where can I learn more about Perceptrons?
page: 1 / 14
page: 1 / 14