Self-contained economy for distributed deep learning applications. It allows anyone and everyone to become a shareholder of AI services.
Infrastructure-as-a-Service closed-loop market for distributed applications. It allows trustless collaboration between stakeholders for development and operation of parallel-processing-oriented digital services. The project mission is to provide economic stability in the age of automation, by allowing anyone and everyone to become a shareholder of AI services. This is a technical and logical framework for developing, deploying and maintaining digital and hybrid (digital+quantum) services using self-enforcing agreements between stakeholders. Main pain points/needs: ● The training of commercially-relevant AI solutions require allocation of vast amounts of computing power, making it difficult for independent developers to compete with Big Tech. This is why more than half of the TensorFlow models posted on Github are untrained. As an example, Open AI’s GPT-3 would cost an estimated $12.6M, to train using AWS infrastructure. ● Startups operating in centralized markets are leaking resources through the legal and administrative tasks associated with forming the team, selling the solution and managing the earnings. ● Lack of general awareness regarding leading-edge AI capabilities exposes humanity to potentially dangerous predicaments. The most competent AI models are currently developed and used by centralized entities with minimal or even zero transparency.
How it's made
The operating model enables TensorFlow developers to access parallel computing resources for free through a system of incentives. It also allows stakeholders to deploy their solutions and distribute the generated revenue according to the terms set in Smart Contracts in a fully-trustless manner. Our traction so far: * The distributed Solidity code is fully implemented and accounts for the entire economic system. * Proof of concept on the distributed training nodes consisting of: - retrieving project logic from github when the contract broadcasts the request to initiate training - distributed training process on multiple EC-2 machines, synched over TCP-IP - dynamic endpoint allocation for serving customers when the agent is fully trained - customer credit verification upon fulfilling request. * Cross-platform client that allows the following distributed actions: - create new ATN contract (this is the equivalent of a User entity on a centralized backend) - create and publish new project as developer (by linking github repository and setting up trustless distribution of revenue) - invest in project using Eth. - access utility function of trained agent both through client and with API calls. The relevant repositories: Economy (Solidity) : https://github.com/autonet-code/contracts Client application: https://github.com/autonet-code/app Mining node: https://github.com/autonet-code/node Demo/template project: https://github.com/autonet-code/demo-project Regarding the live demo at WWW.ATNCOIN.COM - The node is not yet integrated in the client. The node screen in the client is just a mock up. - Please switch to Rinkeby before connecting your Metamask. - Refresh the page after creating your first contract in the Assets section. - Should it request a password, please reach out to Eight Rice#1340 on Discord or join our server and ask for access https://discord.gg/BAYCZWfcs9Technologies used