General Questions

Neuromation is a blockchain platform (the “Neuromation Platform”) for the distributed generation of synthetic, labeled datasets for training of deep neural networks. The Neuromation Platform implements the principle of “knowledge mining”, which assumes that the computing power of mining rigs not only supports abstract blockchain cryptographic computing but also serves to solve specific practical problems that have a physical meaning.

A blockchain (or block chain) is a continuously growing list of records, called blocks, secured using cryptography. The blocks are stored on the devices of chain members rather than on a shared server. Each new block contains information about the previous one, and therefore inherently no block can be changed without modifying all the related blocks. Thus, a blockchain is inherently resistant to data modification and can serve as a reliable means of storing information.

A cryptocurrency is a digital asset designed to work as a medium of exchange; its emission and control is decentralized based on blockchain technology. Along with counterfeiting protection, blockchain technology ensures that there is no need for any internal or external administrator capable of influencing the next transaction. Cryptocurrencies only exist electronically and do not have any physical equivalent.

Synthetic data are not obtained by direct interaction with the environment, but generated by electronic devices based on certain algorithms. The very nature of synthetic data implies unprecedented accuracy (unachievable when collected manually), a high rate of creation, as well as the availability of metadata necessary to process this information automatically.

High-quality training of neural networks requires massive datasets with highly accurate labeling. Preparing them manually is an extremely expensive and slow process. Synthetic data makes the process tens, if not hundreds, of times faster and cheaper, making it affordable for a wide range of potential users.

A neural network is a learning system that evolves in accordance with not only established algorithms, but also based on experience. It consists of neurons connected via synapses. The structure of such networks came to the computer world directly from biology.

A token sale is an event in which companies making use of blockchain technology attract participation and support for their platforms by selling cryptocurrency tokens to early adopters and enthusiasts in exchange for money today. Usually these tokens, after their issuance, can be used on the project platform as internal currency for payment of services. The project, in turn, receives the financing necessary for startup costs and development of the platform.

NeuroTokens (NTK) are a cryptocurrency of the Neuromation Platform used in all transactions related to distributed computing tasks on the Neuromation Platform and serving as a “reward” for miners for providing computing capacity.

The mining of NeuroTokens (NTK) creates synthetic datasets to solve real world problems and to train thousands of neural networks. Therefore, the huge resources of mining rigs are not just spent on blockchain computing, but work to solve fundamental and applied problems. At the heart of NeuroToken (NTK) mining is the principle of “knowledge mining”.

The value of NeuroTokens (NTK)  is backed by the practical value of the calculations performed. On the Neuromation Platform, a library of synthetic data will be formed which will contain datasets created by miners. End users will be able to purchase the right to use the information already available or to order the required datasets using NeuroTokens (NTK) as a means of payment.

Its application is possible in many fields of human activity where the use of machine learning models is or becomes a reality, from retail to transport to medical research and cell behavior simulation. End users: industrial businesses, research centers, biotechnology companies, etc.

Proof of hypothesis of training neural networks based on synthetic data is complete. Our Image Recognition MVP (minimum viable product) service for the retail sector is ready.

The Image Recognition service is a “pilot” project for the retail sector on the Neuromation Platform using deep neural networks for automatic intellectual recognition of merchandise on store shelves and using this information to optimize turnover.

Neuromation Labs (the “Labs”) are designed to develop and improve network deep learning algorithms and to create synthetic datasets. The mission for each individual Lab is to collect, process and analyze real world information from a specific area in which the Neuromation Platform can be applied in order to develop methods for automated data generation, and to investigate and solve potential problems that may arise during AI training using neural networks, optimization of learning algorithms, etc.

Each Lab will specialize in a specific industry for implementing AI. The Lab for retail industry automation is already in operation; it has developed and is testing the Image Recognition MVP service.

Questions about the Neuromation Token Sale

In order to participate in the Neuromation Token Sale (the “Token Sale”), you can (subject to eligibility restrictions) purchase the desired amount of NeuroTokens (NTK) through our website, making payment during the sale process.

The NeuroToken (NTK) pre-sale starts on October 25, 2017 (registration is required for participation in the pre-sale). Pre-sale ends on January 1st, 2018. The sale starts on January 7th, 2018 and ends on the earlier of (a) February 15th, 2018 or (b) seven (7) days after a secret cap is reached, as determined in Neuromation’s sole discretion.

60,000,000 NeuroTokens (NTK) will be issued in the Token Sale.

The price per NeuroToken (NTK) will be Ether (ETH) 0.001.

The purchase process will be managed using smart contracts on Ethereum. Participants in the process will need to transfer the correct quantity of ETH to the specified address. We recommend carefully checking the address with the transfer address field before transferring the ETH.

NeuroTokens (NTK) can be stored in any Ethereum Wallet you use, such as

Citizens of any country can take part in the Token Sale, except for:

1. Citizens of countries in the OFAC list;

2. Citizens of China and Hong Kong.

*U.S. citizens and residents must qualify as having U.S. accredited investor status exclusively through and submit verification documentation in order to participate in the Token Sale.

The exchange will be ensured through a smart contract mechanism that supports the guarantee of the transaction subject to, and in accordance with, the specified conditions.

A total of 100,000,000 NeuroTokens (NTK) will be minted, of which 60,000,000 will be issued in the Token Sale. The remaining 40,000,000 NeuroTokens (NTK) will be in reserve.

30% – development of the Neuromation Platform

40% – liquidity reserve (prepayment for server capacity)

10% – for PR and marketing of Neuromation services

10% – grants through Neuromation Ventures (in support of deep learning, etc.)

10% – to the Neuromation team

NeuroTokens (NTK) are a cryptocurrency of the Neuromation Platform used in all transactions related to distributed computing tasks on the Neuromation Platform and serving as a “reward” for miners for providing computing capacity.

Our clients will have to be proficient with neural networks only at the very beginning, immediately after we launch the service. We aim to create a user-friendly platform where every user can request data, choose a deep learning model and train it on distributed computing power, without the need to ever develop expertise in neural networks.

Service nodes represent Neuromation in-house services or any third-party service providers who provide data modeling for generating synthetic datasets or neural network researchers who can provide specific machine learning models to the clients of the Neuromation Platform.

Yes, someone at Neuromation (or elsewhere) needs to create a synthetic data generator. However, not only the training, but the data generation itself, can be sent out to distributed nodes:  you send out a script that contains 3D models and indicates how to place them and the nodes render actual images. Rendering photorealistic images may take a lot of computational power, so this is not a pointless endeavor. Moreover (and this is a very important point), once you combine synthetic data with training neural networks, you no longer have to store or transmit the entire dataset at all! On a GPU farm, some GPUs can do the rendering while others are training on generated synthetic images, and there is no need to store all generated images, only the current mini-batch. This becomes especially important as many mining rigs do not have a lot of storage space and do not have a very fast Internet connection. We have talked to miners and while they are willing to upgrade storage and bandwidth to earn more money, it is even better for them if they don’t have to.