The 4 Participants in Decentralized Machine Learning Protocol
In the previous article, we have discussed about what are the constraints of current machine learning development, and how DML Protocol will tackle it to bring the industry to a new landscape.
Today we will focus on one of the core components, which are the participants in DML Protocol. There are four types of participants, namely Customers, Developers, Data Owners and Decentralized Nodes. The four participants will interact in the DML Infrastructure:
They are companies, research institutions, governments, NGOs or any individuals that have practical needs for machine learning predictions, such as trend analyses or market research requirements.
To look for a suitable machine learning model/algorithm for prediction, they can simply search and filter in DML Algorithm Marketplace. They can even request for a customized algorithm from Developers in the Marketplace.
Unlike traditional machine learning models/algorithms development, the supply of models/algorithms in DML Protocol is crowdsourced in our Developers Community. Every Developer can list his own models/algorithms in DML Algorithm Marketplace as his/her discretion or in response to Customers’ requests.
Data owners, which are also the holders of individual devices like smartphones, tablets and PCs, can authorize specific types of datasets for models/algorithms to be run in DML App. For example, they can choose to authorize photos and videos but not geolocation data for local machine learning, or freely opt in for utilization of all types of data.
Data will be kept within the devices without transferring to any third parties or being stored in any cloud servers. Thus privacy of data will be well protected. In addition, they can participate the manual fine-tuning of models/algorithms as Algorithm Trainers, in return with extra incentives given from Developers.
They will be further divided into four types according to their functions:
- Distributing Nodes identify and distribute the encrypted algorithms to individual devices.
- Federated Nodes collect, aggregate and average all connected local prediction results by Federated Learning.
- Report Nodes further average the encrypted results processed by the Federated Nodes and generate an encrypted final report to the Customers.
- Algo Refining Nodes collect, aggregate and average the fine-tuning updates by Algorithm Trainers to generate improved models/algorithms to Developers.
Therefore with DML Protocol, we shall create a fully decentralized and autonomous machine learning environment because:
- More companies, organizations, small-and-medium enterprises and even individual proprietors, can now request machine learning services directly, without bearing hefty overheads of a team of in-house developers or traditional agencies.
- On other hand, developers can exhibit their creativity on free-will or by pure market forces, but not depends on corporate bureaucracy.
- Data owners can also receive fair rewards without compromising their data ownership and control. We expect a massive unleash of untapped data usage for machine learning as a result.
- The whole machine learning process will be handled by idle processing power of individual devices and decentralized nodes, but not on a super computer which is only affordable to tech giants or big corporations.
In the next articles, we will discuss deeper about the function of decentralized nodes in DML Protocol, and how the blockchain smart contracts will link up all participants in a trustless and middleman-free ecosystem. You will also meet our team and advisors, understand why we are building DML Protocol.
Join the Community
We will announce all the latest and exclusive news in our community. So chat with the team and our growing community by joining our Telegram channel.