Decentralized computing power
for Automated Machine Learning

An open-sourced network for
self-improving AI models

;
AutoMLs already outperform the best human researches in creating deep neural networks. But there’s a catch.
To autonomously produce effective AI solutions, AutoMLs need vast amounts of computing power. This makes the technology viable mostly to enterprises with significant access to the required resources, making smaller AI-based ventures dependable on a single service.

We are building efficient access to the resources the modern approach to AI so desperately requires.

ScyNet is a blockchain protocol and open economic system. It rewards computing nodes for sharing resources in a decentralized network. AI ventures that configure their own AutoML nodes on the system are able to engage with the GPU power to test and improve their models.

The autonomous cycle of improving AI models

ScyNet allows private ventures to setup AutoML nodes and train their models on available trainer nodes on the network.

 The network is governed by a Zero-Knowledge Proof Protocol, which ensures a trustless method of validation where AutoML nodes remain the effective owner of each model, while at the same time Trainers are able to prove the work they’ve put in improving the model without actually revealing the model to the validating third-party.

AUTO ML NODE-01

An AutoML node creates a stream of Training Jobs.

Each job consists of a Model, Obfuscated data, and a performance metric.

Jobs are picked by computing nodes, also known as Trainers.

small bits-05
small bits-05
small bits-05
small bits-05

Models are trained on the provided dataset until an optimal configuration is achieved.

Computing Power Node-02
small bits-05
small bits-05
small bits-05
small bits-05

As the training job is complete, the trained model is returned to the AutoML node.

Validators test the accuracy of the model via the Zero-Knowledge Protocol.

AI models are ranked and trainers are rewarded. 

The AutoML learns from the best models and creates better configurations for the next cycle of jobs.

The autonomous cycle of improving AI models

ScyNet allows private ventures to setup AutoML nodes and train their models on available trainer nodes on the network.

 The network is governed by a Zero-Knowledge Proof Protocol, which ensures a trustless method of validation where AutoML nodes remain the effective owner of each model, while at the same time Trainers are able to prove the work they’ve put in improving the model without actually revealing the model to the validating third-party.

1.

  

  

An AutoML node creates a stream of Training Jobs.

AUTO ML NODE-01

8.

 

 

The AutoML learns from the best models and creates better configurations for the next cycle of jobs.

2.

Each job consists of a Model, Obfuscated data, and a performance metric.

3.

Jobs are picked by computing nodes, also known as Trainers.

small bits-05
small bits-05
small bits-05
small bits-05
small bits-05

7.

AI models are ranked and trainers are rewarded. 

6.

Validators test the accuracy of the model via the Zero-Knowledge Protocol.

4.

Models are trained on the provided dataset until an optimal configuration is achieved.

small bits-05
small bits-05
small bits-05
Computing Power Node-02

5.

As the training job is complete, the trained model is returned to the AutoML node.

Making use of

Winner of

Project by

Partner of

The Internet of AI?
A truly decentralized source of collective AI.

Private AutoMLs usually obfuscate their training jobs so that trainers cannot reutilize the AI knowledge themselves and hence the AutoML holds exclusivity of the final result. An alternative approach that suits parties such as open-source organizations or foundations is public AutoMLs.

These participants help jumpstart the ecosystem by revealing the meaning of their data and models, in turn allowing the trainers to reutilize the models for their own purposes.

Bloomberg-2-03
Bloomberg-2-03
Bloomberg-2-03
Bloomberg-2-03
Bloomberg-2-03
Bloomberg-2-03

Let’s Chat

Applications can emerge on many fields of supervised, reinforcement and generative AI. We are open to discuss future implementations.