Decentralized Machine Learning (DML): A Marketplace for Algorithms and AI Models

Executive Summary

Project Description

The development of the machine learning sector is currently being stifled by a couple of problems: inaccessibility of private data, centralization of processing power, and limited algorithm development. Furthermore, machine learning datasets have to be uploaded to dedicated servers because of privacy concerns. Because of this, sizable amounts of private — and useful — data that are stored on personal devices like smartphones and laptops cannot be accessed. Decentralized Machine Learning (DML) plans to solve this problem by providing individual devices with encrypted algorithm that allows them to conduct machine learning locally. This removes the need to upload private data to third-party servers. Rather, the data will remain in the devices, and only the prediction results gotten by the algorithm will be sent out.

Another hindrance to the development of the machine learning scene is the centralized nature of the networks used. Using just one machine, there is a limit to processing power. On DML, there will be a network of machines, and idle processing power will be used to achieve on-device machine learning.

Furthermore, only the biggest companies have the resources to build in-house machine learning models and algorithms, or purchase them from consultancy firms. DML wants to remove this scarcity and make these services available to all. They plan to do this by creating a marketplace where algorithms can be crowdsourced. Customers (including research institutes, corporate organisations, government agencies, NGOs etc.) will be able search and request for customized machine learning algorithms. DML also wants to organise machine learning competitions; these will be used to reward talented members in the DML community and at the same time, attract even more talent to the platform.

The DML platform will feature four major participants: the customers, the developers, data owners, and decentralized nodes. Here is how all four parties above will benefit from the DML ecosystem: the customers can post requests for customized algorithm in the marketplace, and the developers will respond by listing all the algorithms they currently have in the marketplace. If none of them fits the customers’ needs, the developers can then offer to develop models according to the customers’ requests. Conversely, data owners will provide datasets for algorithms; these datasets are run through the DML platform and to further ensure privacy, they will not be transferred to third parties nor stored in cloud servers. Data owners can also acts as algorithm trainers, fine tuning models and algorithms. Finally, the decentralised nodes will be used to identify and distribute encrypted algorithms to devices on the DML network.

DML’s token, also named DML, is ERC-20 compliant and it will serve as the governing currency on the platform. It will serve as incentive for users who provide private data and idle processing power; customers will also pay developers for their services in DML. Furthermore, the token will be used to incentivise decentralized nodes, and developers will use it to pay algorithm trainers for their work.

Hard Cap and Valuation

DML have not made a definitive decision on their hard cap. It is estimated to be between 28,000 ETH and 32,000 ETH.

Price Per Token

Yet to be released.

Important Dates

An “early whitelist” program for investors is going on currently. The main whitelist is to be launched sometime in February 2018 and there will be individual cpas of 0.5–2 ETH. The DML Protocol Gen 1 is scheduled to go live in September 2018.

Marketing Power

As at the time this post was published, DML had almost 6,500 members on Telegram, 600+ followers on Twitter, 55 followers on Medium, and 48 readers on Reddit.

Howey’s Test

According to Howey’s Test conducted by Research Center analysts, it is unlikely that DML will be considered a security.

Team Members’ Areas of Expertise

Business experts

Victor Cheung, Blockchain Developer.

Co-Founder of Oika, an online marketplace for wedding-related services.
Co-Founder of Capheart, an award winning first-aid app.

Michael Kwok, Project Lead Director.

Co-Founder of Oika.
Co-Founder of Capheart.

Jacky Chan, Blockchain and Software Developer.

Co-Founder of Kyokan Labs, a blockchain consultancy firm.

Target market experts

Pascal Lejolif, Machine Learning Engineer.

Former CTO of Alkia IT Services, a cloud computing, AI, and cybersecurity firm.
Machine Learning Certification from Stanford University.

Matthew Slipper, Machine Learning Engineer.

Co-Founder and former CTO of Spectrum Labs.

Scott Christensen, Machine Learning Advisor.

Co-Founder and CEO of Hanpa Group.

Kyle Wong PhD., Machine Learning Advisor.

Founder and COO of Artificial Intelligence.

Marketing experts

No team members with marketing expertise.

Legal experts

No team member with legal expertise.

Software engineering experts

Victor Cheung, Blockchain Developer.

Full stack web, app and smart contract developer. He is also proficient in Solidity, C#, JavaScript, Node.js, PHP, MySQL, MFC.

Jacky Chan, Blockchain and Software Developer.

Software developer at Uber for a year.
Software Engineer at Symphony Communications (acquired by Goldman Sachs) for a year.

Alex Tsueng, Software Developer.

Hardware validation engineer at Cisco Systems for over 4 years.
Front-end engineer at Avi Networks for 2 years.

David Yoo, Software Developer.

Software Engineer at Uber (ongoing).
Full stack engineer at Ribbon.co, an e-commerce startup, for 3 years.

Blockchain development experts

DML are in partnership with Kyokan Labs, a blockchain consultancy firm. Additionally, some members of their team have significant blockchain development experience.

Jacky Chan was a contributor to Metamask and DFINITY.
Matthew Slipper contributed to Machinomy.
Alex Tsueng also contributed to Metamask.

Token sale structure experts

No team member with token sale structure expertise.

Token economics

No team member with token economics expertise.

Disclaimers:

Nothing written in this article is a legal or an investment advice.
Information is provided on a best-effort basis and is subject to change without prior notice. Be sure to verify everything you read with a project team.

The analysis was produced by Research Center team members: Alexander Hinz, Mark Jedd, Eugene Tartakovsky.

We are constantly working on improving our work and welcome all constructive feedback. Let us know what you think.