Neuromation Rating Review
|Start ICO||06 Jan 2018|
|End ICO||07 Jan 2018|
We have reviewed project's progress as of December 2018 and identified that since August 2018 the project's GitHub hasn't been updated, as to the product, there is still only the demo version. However, the project's community has grown significantly since the previous date of analysis, which is a pro. However, taking into account all the abovementioned disadvantages and the fact that the ICO market conditions have changed not in favor of ICO projects, the project's rating has been decreased to Risky.
We assign the Neuromation project a "Risky" rating.
The project is entering a dynamically developing market with impressive growth rates, which will remain high in the coming years according to experts' forecasts.
The investment attractiveness of the token, due to the mechanism of "burning" will be at a high level in the next 3 years; this allows the project to attract longer-term investors.
Speaking about the strengths, we note that the team has all necessary skills for a successful start to the project and its subsequent development; the project’s media (marked by various awards) and the upcoming Token Sale are carried out to a high level.
At the same time, we see a number of risks that could prevent the project’s successful development.
First and foremost, there is growing competition. In this case we are talking not about direct competitors to Neuromation, but about different projects using a mechanism for rewarding miners in a similar way to Neuromation (as an alternative to classic crypto mining).
Not all parameters of the token sale (e.g. the soft cap) are disclosed.
The global marketing strategy for further promotion of the project is not described; thus, what funds will be aimed at future promotion of the platform remains unclear.
General information about the Project and ICO
Neuromation is a platform on Ethereum blockchain which enables building a library of synthetic data using the computing power of private and commercial mining service providers. Using the sets of these data, Neuromation can effectively train models of neural networks.
In other words, it is a platform designed to introduce the process of developing artificial intelligence models to various industries using neural networks.
One of the main goals of the Neuromation platform is to become a center for artificial intelligence services for international business, providing a new approach to combining supply and demand in each of these areas on a large scale.
The Neuromation company is based in Estonia and fully complies with the legislation of Estonia for regulating crowdfunding. Participation in the ICO is open to all except for citizens of China, Hong Kong and citizens in the list of OFAC. US citizens will have to undergo additional accreditation as "qualified investors" via the following service: https://verifyinvestor.com/
Slack: No information
Twitter (1505 followers)
https://t.me/Neuromation (565 followers)
https://t.me/neurotoken_io (169 followers)
Facebook (3790 followers)
GitHub: No information
Smart contract platform: Blockchain Ethereum
Contract type: ERC-20
Start date: October 25, 2017
End date: November 28, 2017
Target sales volume of tokens for Pre-Sale: 60,000,000 NTK (This number of tokens is allocated simultaneously on the pre-sale and main-sale).
Bonus: When applying to the "whitelist" on the website, a 25% bonus is required as well as bonuses depending on the date of purchase:
Price: 1 NTK = 0.001 ETH
Minimum purchase amount at Pre-Sale: 3 ETH
Maximum purchase amount at Pre-Sale: no restrictions
Start date: November 28, 2017
End date: January 1, 2017
Purpose-oriented token volume: 60,000,000 NTK
Price: 1 NTK = 0.001 ETH
Minimum purchase amount: no restrictions
Maximum purchase amount: no restrictions
Maximum token emission: 100,000,000 NTK (There will be no additional emissions)
- First week: 15%
- Second week: 10%
- Third week: 5%
- Bonuses depending on the amount of purchase:
- >1,000 ETH +1%
- >2,000 ETH +2%
- >3,000 ETH +3%
- >5,000 ETH +4%
- >10,000 ETH +5%
Accepted payment: ETH
Structure of distribution of funds raised during the token sale:
- At least 40% - platform development
- Up to 40% - liquidity reserve (prepayment for server capacity)
- 10% - PR and Neuromation marketing services
- 10% - remuneration to partners / consultants / early investors
Tokens unsold during the token sale will be burned.
Tokens assigned to the team as a reward will be frozen for a period of up to 24 months with a monthly partial release for the entire period.
Some NKT tokens will be removed from circulation within the next three years. The mechanism of "burning" tokens is described in detail in "Investment attractiveness of the token".
Description of the services and scope of the project
The Neuromation platform is designed for:
- Synthesizing large volumes of data with perfectly accurate markings without any inaccuracies arising from human factors.
- Effective training of large capacity, in-depth neural networks.
- Replacing costly and slow crowdsourcing with a fast, efficient and accurate tool for creating and marking data.
- Generation artificial data and comparison with real data.
The platform will use distributed processing of data and proof-of-work blockchain tokens, which can facilitate the transformation of the process of developing models of artificial intelligence by combining components necessary to create solutions in the field of in-depth entraining.
Neuromation will work with service providers (commercial and private) that will provide resources for:
- Creating artificial data sets.
- Implementation of distributed computing services.
- Developing machine learning models.
Each of these services will be paid for by the domestic currency of the platform i.e. Neurotoken tokens. Platform users will be able to buy tokens on a Neuromation client portal using a simple and intuitive interface.
The platform provides the following types of services:
- Purchase of data
- Classification of data
- Data generation
- Buying a model
- Model training
- Training of neural networks
- Serving AI requests
The user interface consists of three categories, containing 3 main modules with certain internal processes and libraries used:
Module of artificial data sets:
- Creating a data generator
- Order for creation of a data set using a data generator
- Request for data labeling
- Repository of data sets for in-depth training
- Data generator repository
- Data sets (marketplace)
Machine Learning Module:
- Defining the model for in-depth training
- Importing the model
- Order for training for the selected data set
- Request for a custom model on the marketplace
- In-depth Learning Model Repository (Marketplace)
- Purchase of tokens
- Register as a supplier or consumer of processing power
- Registration as a consumer or service provider
- Downloading and installing Neuromation Node software
- User data
- User Models
The technology of the server part of the platform consists of two components that perform certain tasks:
- Analysis of nodes (minimum level of processing costs is equal to minimum amount in Neurotoken paid for the unit of calculation, the maximum performance of the node based on bandwidth / processing power / storage capacity.)
- Distributed data generation package.
- Distributed training package.
- Intermediate node synchronization software.
Trading (market) module:
- Ensuring effective comparison of orders for the purchase and sale of data sets, models and marking services.
- Providing system liquidity so that asset prices help to scale the system.
The platform will allow users to create data sets generators, generate large amounts of data and train in-depth training models; users will be able to trade datasets and models within the platform.
The widespread introduction of large-scale generation of artificial data sets is hindered by a lack of computing power. Neuromation can offer cryptocurrency managers the use of their equipment and to receive financial benefits for solving this problem. In order to do that, a miner needs to download the software – the Neuromation Compilation Node. When it is proposed to perform a Neuromation task, a miner (node) is selected through bidding to perform the tasks, such as generating artificial data or training an in-depth training model. In this case, the system chooses nodes according to the following parameters:
- performance / node capacity (Efficiency score).
- declared cost of the service (the price is set by the miner).
The algorithm of choice is arranged in such a way that miners charging a lower price for services will be selected first of all. The same applies to equipment power - nodes with more productive equipment will be selected firstly. Different computing resources may be required for different tasks; one node may be sufficient for hosting trained models but for such capacious tasks as learning networks, much more computing resources will be required. The speed of a task is proportional to the number of miners connected to its execution. The maximum number of nodes for performing large tasks is not defined at this stage.
When a task is completed, the miner receives payment in Neurotoken, and his equipment continues to be used for crypto mining.
Neuromation is planning to open its own laboratories to develop artificial data and train in-depth training models in real applications. Each of its laboratories will investigate a specific problem; with the development of the platform a part of generation and training will move to the laboratory. Currently, 2 laboratories have already been launched:
- A retail laboratory dedicated to creating in-depth training models that can recognize, classify the availability, filling and accuracy of the layout of objects on store shelves, as well as other indicators.
- The Laboratory for Medical Devices, in cooperation with MonBaby - a manufacturer of infant monitoring devices, is developing a smart camera that can monitor the movements of a child and transmit to its parents data on respiratory movements, body position (on the back or on the abdomen), fall detection and other types of alerts that can be customized individually.
Neuromation is planning to open an industrial automation laboratory for Enterprise Automation, where artificial data will facilitate the introduction of new solutions in production, supply chains, financial services, agricultural industries, etc.
The platform will use a test environment of its own design – a Sensor Emulation Sandbox, for procedures to create a virtual environment in which the training of models will take place - for such areas as the automatic piloting of unmanned aerial vehicles, monitoring of industrial processes, manipulation of objects, etc.
There are no repositories available on Github; platform development is conducted in closed mode. According to information from the founders, it is being developed by 12 people.
Neurotoken – an Ethereum cryptographic ERC20 token, available for storage in various wallets compatible with this standard.
The use of smart contracts will be available with the platform version 2.0; interaction with smart contracts will occur in the following order:
- The client makes an order while a request is written on blockchain
- Nodes apply for data processing
- The platform selects nodes based on the performance of the node and its cost of services
- A smart contract is created to perform the task with the selected nodes
- After the completion of the task, the nodes are paid for Neurotoken, according to the price they set for the services.
Neuromation has trained several different models:
- Object recognition (SSD, YoLo, Fast (er), R-CNN);
- Segmentation of objects (SegNet, FCN for segmentation);
- Currently, instance segmentation (Mask R-CNN) is being developed
All the models above contain a convolutional network that performs the bulk of image categorization work and is distinguished by additional layers over these functions that process the actual semantics and produce class labels, supposed delimitation of objects, segmentation, and other parameters.
In the future, it is planned to improve the chains of artificial data generation using machine learning algorithms.
Many computer learning models are freely available - computer vision programs: VGG, GoogLeNet and ResNet, speech generation programs such as Wavenet, etc. Such software is integrated into various frameworks and libraries, for example, TensorFlow, Caffe, Torch, pyTorch, Theano, Keras; converters are used for the connection of different types of framework. Neuromation will support all major frameworks, in-depth learning and automatic differentiation libraries, as well as pre-trained models. The results of the training will be available for download.
Compression will be applied to the models by removing redundant information to reduce the response time on mobile devices ("model distillation" method).
In cases when existing models and tools are not suitable for a client to develop a model, he needs to make a request with special characteristics, which will be placed on the Neuromation exchange and processed by specialists.
According to experts [https://www.statista.com/statistics/607960/worldwide-artificial-intelligence-market-growth/], the Artificial Intelligence market is expected to grow by about 175% by the end of 2017 in comparison with a value of $2.4 billion in 2016. At the same time, the dynamics of the growth rate of the market enables forecasting that its size could exceed $59 billion by 2025.
According to statistics, the size of the global market for Artificial Intelligence applications was estimated at about $360 million [https://www.statista.com/statistics/607612/worldwide-artificial-intelligence-for-enterprise-applications/] worldwide in 2016; according to forecasts, the market will grow by another 135% [https://www.statista.com/statistics/607681/worldwide-artificial-intelligence-for-enterprise-applications-growth/] and will amount to more than $840 million in 2017.
The main technologies used in the market are:
- Natural language processing - understanding and synthesis of human language for AI
- Machine learning - AI learning in the process of solving many similar problems
- Speech recognition - converting speech to text
- Image processing - image recognition technology
- Robotics - design, creation, use and operation of robots
- Digital personal assistants - software agents capable of performing tasks or services for an individual (for example, chat-bots) as well as other technologies such as querying method and context-aware processing.
This market can be conditionally divided into several types:
- Artificial neural networks – networks built on the principles of organization and functioning of biological neural networks
- Expert systems - computer systems that simulate human ability to make decisions
- Automated robotic systems - automation of production and business processes using robots of various types (robotized automation.)
- Embedded systems - specialized management and monitoring systems built directly into the device managed.
- Digital assistance systems – a variety of chat-bots and virtual assistants (Siri, Cortana, Google Now etc.)
Depending on the application, the market is classified as:
- Gesture control
- Cyber security
- Video analysis
- In-depth learning
Tractica's research shows that worldwide revenue from software for in-depth training will grow from $655 million in 2016 to $34.9 billion by 2025.
It is expected that the use of in-depth training in terms of profitability will be applied in the following 10 cases:
- Static image recognition, classification, and tagging
- Machine/vehicular object detection/identification/avoidance
- Patient data processing
- Algorithmic trading strategy performance improvement
- Converting paperwork into digital data
- Medical image analysis
- Localization and mapping
- Sentiment analysis
- Social media publishing and management
- Intelligent recruitment and HR systems
Thus, the project enters a dynamically growing market and this certainly can contribute to its successful implementation.
Competitors and competitive advantages of the project
Despite the fact that we did not find any direct competitors in the market, it should be noted that there are a lot of projects that use the principle of alternative mining in models. Competition will arise for the services of miners which may affect the final price of cost of services on the platform.
We give two projects as examples (Golem and Gridcoin), although there are many more:
Golem — the first decentralized supercomputer forming a global market of computing resources. Anyone can use Golem to lease their unused CPU / GPU computing power. The price depends on the complexity of the task; it can be a small amount or several thousand dollars. Payments are implemented directly between customers, suppliers and developers via Ethereum. The commission for each transaction is 5% of the total payment amount.
The main areas of effective use of Golem:
- CGI rendering
- scientific research
- data analysis
It is notable that the crowdsale brought 820,000 ETH to the project, which was $8.2 million considering the fact that the crowdsale took only 20 minutes. Currently, the capitalization of cryptocurrency is more than $170 million.
- SETI@home (search for extraterrestrial civilizations);
- Rosetta@home (calculation of protein structure that will help to cure some genetic diseases in the future)
- World Community Grid (development of methods to treat cancer, ebola, ZIKV and muscular dystrophy)
- Performance of calculations for the hadron collider, searching for pulsars and gravitational waves, combinatorics, various projects in the field of mathematics, physics and biology.
This network itself generates and distributes coins among participants in the calculations. Thus, the organizers of research do not need to pay participants (in contrast to GOLEM). Reward for the block is from 5 to 150 GRC. 5 GRC is paid for usual CPU / GPU mining without running the BOINC application. Reward in the range of 6-150 GRC occurs for mining with BOINC on. All payments are based on BOINC utilization.
Capitalization of this platform is more than $13 million.
At the same time, in contrast to competitors for mining capacities, Neuromation has the advantage of being able to provide an adapted platform where Neuromation nodes are configured to solve certain tasks, such as:
- Synthesizing large amounts of data
- Training of in-depth neural networks
- Generation of artificial data
- Model training
- Serving AI requests
Thus, Neuromation provides an exchange platform and environment in which participants can both contribute to the creation of an artificial intelligence model and acquire its components. Meanwhile, the competitors act as intermediary services, writing containers for Amazon, Google Cloud, Azure and others.
We also note the following as advantages for Neuromation:
- The miner himself sets the price for his services. Accordingly, it will be higher than for crypto mining.
- A partnership has been concluded with the Hacken project which will ensure cybersecurity.
- Neuromation is planning to move to its own blockchain, and the installed capacity should reach 100,000 GPU after 2018.
In addition to the traditional risks inherent in the crypto market, we draw the attention of investors to the following facts:
There are more and more projects based on mining in the crypto industry. This is not about mining of cryptocurrency, but about attracting miners with their equipment to the platform to use their capacities for the project. Therefore, we will note the risk of competition as the main one, which can greatly affect the viability and development of the project. Moreover, the documentation has no information about global marketing strategy, which is why it is not clear how the developers will attract miners to their platform.
Also we note the lack of any information about the soft cap amount.
Any other significant risks that could have a negative impact on the attractiveness of the Neuromation project were not detected.
17 people are involved in the development of the project.
The project management consists of 5 people. Leading positions are occupied as follows:
Maxim Prasolov (LINKEDIN) – CEO.
Has been working for Neuromation since February 2017. Responsible for overall strategy of the project and marketing. Former member of the team implementing the initial public offering (IPO) of Ferrexpo, the largest producer of iron ore in Ukraine, on the London Stock Exchange. Has been investing in start-ups in the field of development of unmanned aerial vehicles, AI, multimedia using augmented reality (AR) since 2014.
- State University of Management (SUM) 1994-1999
Constantine Goltsev (LINKEDIN) – Chairman, investor at Neuromation.
Professional entrepreneur from the online advertising industry. Has more than twenty years of experience developing software and products. Former CEO and founder of the innovative advertising network AdoTube .
- University of California, Berkeley 1995-2000
Fedor Savchenko (LINKEDIN) – CTO.
More than twenty years of managing complex software development projects with an emphasis on computer graphics, 3D engines, production of CGI and virtual reality (VR). Has advanced degrees in mathematics and graphic design.
- Sevastopol's'kij Nacional'nij Institut Jadernoj Energii ta Promislovosti 1996-2001
Sergey Nikolenko (LINKEDIN) – Chief Research Officer.
Involved in the project since March 2017. Responsible for research in machine learning (in-depth network training, Bayesian methods, natural language processing, etc.), analysis of algorithms (network algorithms, competitive analysis), bioinformatics. Author of more than 120 scientific works, several books, popular courses "Machine learning", "Training of in-depth networks" and others. Has extensive experience in participating in commercial projects (SolidOpinion, Deloitte Analytics Institute).
- Steklov Mathematical Institute, St. Petersburg 2005-2008
- Saint Petersburg State University 2000-2005
Denis Popov (LINKEDIN) – Co-Founder, CIO, Advisor.
Determines technical requirements, carries out development, implementation and provides complex solutions for the project. More than 15 years of experience devoted to software development.
- National Technical University of Ukraine «Kyiv Polytechnic Institute» 1999-2005
12 people in addition are involved as advisors and platform developers, many of whom have extensive experience in their field of activity:
Esther Katz – VP Communications, specialist in public relations with 20 years of experience in the main markets, specializes in bio and financial technologies.
Andrew Rabinovic – advisor, leading world scientist in the field of in-depth learning and image recognition research. Author of numerous patents and scientific publications. Has a Ph.D. in the field of computer science at the University of California, San Diego (UCSD).
David Orban – advisor, founder and managing partner of Network Society Ventures,, specializing in innovative venture-based projects for emerging technologies and decentralized networks.
Yuri Kundin - ICO Compliance Advisor, specializes in the development of a structure and methodology for risk assessment, compliance and certification of blockchain ecosystems, crypto currency and ICO. Current director of KPMG office in San Francisco.
According to the founders, it is planned to increase project staff to 25 people by February 2018.
The team members have all the necessary skills for successful implementation of the project, such as development and entrepreneurship; there are specialists who have wide experience in the sphere of business development in leadership team.
Development strategy and roadmap
The project's final goal is to create a distributed platform based on blockchain covering all aspects of a future ecosystem for artificial data and enabling users to create data set generators and generate large data sets, train in-depth training models and trade datasets and models.
The Neuromation platform will combine the following on an integrated AI marketplace:
- Market resources
- Scientific community
- Companies and individuals
The founders are planning to achieve these goals in the coming year.
To achieve the specified goals, the team have carefully arranged the development strategy for the project depending on the expected success of the token.
The founders believe that Neuromation can become a recognized business partner in developing the potential for artificial intelligence use after 2018. Neurotokens will be the main mechanism of exchange, allowing generation of artificial data, implementing the distributed training of models, the marking of data and other services in the field of artificial intelligence.
The project is also planning to become a global pool of resources in the field of artificial data, some kind of constantly replenished library which will have data sets for any possible case.
The founders expect the planned development of the project as well as worldwide distribution of the platform which depends on the expected success of the token sale (achieving 60,000 ETH worth of sales).
Neurotokens are planned to be placed on exchanges in the first quarter of 2018.
Thus, the project has clear and transparent development plans.
The Neuromation team allocates 10% of all funds to marketing. The marketing strategy is presented neither in the white paper nor on the website. Having studied various materials online, we make the following conclusion regarding current marketing actions and achievements of the company:
- The team already has cooperation agreements with partners such as OSA HP, ECR, MonBaby. In the near future it is planned to open the laboratory for Enterprise Automation (industrial automation.) From the beginning of 2018, the number of partners will increase and according to Neuromation forecasts, this source of profit will account for almost a third of the company's total revenue.
- In September, Neuromation received a special prize from Forbes and Forklog at the d10e conference in Kiev. The awarded media prize gives the project an opportunity for private pitching with a consortium of BIC blockchain investors, founded by Mike Costas. This consortium manages a capital of $2 billion and unites more than 100 ICO investors, including crypto currency hedge funds and private individuals.
- In October, Neuromation took third place at the d10e conference on the development of the fintech industry, blockchain, crypto currency and the economy of joint consumption. $33,000 cash prize was awarded to Neuromation founder Konstantin Goltsev and project advisor David Orban.
- Neuromation is mentioned on a large number of sources (more than 40) such as Icobench, Digitaljournal, Bitcoin.info, Bitcoin Insider , The-Blockchain News, Crowdfund Insider, BlockTribune and many others.
The upcoming token sale is discussed on social media; developers interact with the community. Promotion of the project will also occur via a bounty program which the team is planning to introduce in the near future.
In conclusion we can say that the project already has a certain profile, but there is no information on how buyers of services and the miners that provide their computing power will be attracted, as these will be the factors that will cause the project to develop, achieve the desired goals and generate the company's main revenue.
Economy of the project
Achieving the the status of a key international service for artificial intelligence is one of the main goals and objectives of Neuromation.
Plans to capture the market are optimistic and realistic. At the end of the first year, the total volume of transactions on the platform is expected to be $71 million. Each subsequent year (until 2022), use of the platform will increase by 3-5 times.
The platform is planning to generate revenue from commissions that will amount to 5-15% of its services. It is expected that with the development of the platform, in three years’ time, commission fees for Neuromation will exceed $100 million.
Another source of income will be the Neuromation laboratories’ partnerships with other companies, which will bring about 30% of total income.
Currently, the Neuromation laboratory has signed a contract with OSA HP to sell image recognition technology to customers of the OSA Hybrid Platform and for the ECR association to enter the generator and generate synthetic data for 170,000 items in retail trade in Eastern Europe. It is envisaged to create models for the in-depth training and recognition of objects on store shelves. As a result, profit will be €4.25 million over 1.5 years. Cooperation with MonBaby should bring expected revenue from the implementation of the contract of more than €2 million over several years.
Despite the fact that the economic model is built mainly on forecast values it is worth noting its simplicity and realism.
Token investment attractiveness
NTK Tokens are infrastructural and do not give their holders ownership or voting rights. The project does not confer any rights to receive dividends. However, the smart token contract will contain a built-in function to reduce the amount of NTK in free circulation over the next 3 years.
Thus, the growth in token price will be affected by two factors:
- Growth in the number of users on the platform.
- The mechanism for "burning" tokens.
According to the documentation, within the next three years (from 2018 to 2020), Neuromation is planning to burn about 50% of all Neurotokens distributing this value by years:
- 2018 – 30%
- 2019 – 20%
- 2020 – 10%
The algorithm for burning and distributing tokens on the platform is as follows:
1. The customer buys tokens on the exchange or from Neuromation’s reserve.
2. The customer orders work on the platform and Neuromation receives tokens.
3. Neuromation burns tokens:
- Up to 6% of all tokens that will pass through the platform in the first year
- Up to 1% of all tokens that will pass through the platform in the second year
- Less than 1% of all tokens that will pass through the platform in the third year.
These values are "burn tax"; they will be charged to the customer in addition to the cost for services on the platform for each transaction. These values for "burn tax" are indicative and will depend on the turnover on the platform: the higher the turnover, the lower the % that will be taken from the customer. This amount is primarily tied to the combustion plan, so "burn tax" will not be charged to customers after the implementation of the annual combustion plan
4. The remaining Neuromation tokens are distributed among those who are involved in the performance of work.
To maintain liquidity at a sufficient level, a reserve fund will be created where it is planned to store 10% of the tokens. Tokens received in the form of commission will be sold on exchanges as well as replenish the reserve fund with a high demand for NTK.
The white paper presents forecast models of token demand. Thus, according to the presented information, demand for NTK will exceed supply by 4 times in 2018, and by 90 times in 2020. The founders explain such high predicted growth by the growing popularity of the platform as well as by the fact that the costs of services are much higher than those laid down in the calculated economic model.
It can also be noted that a bonus system is provided for investors participating in the ICO.
Given all of the facts above, we can conclude that after the launch of the platform NTK tokens have a high growth potential especially at the 3-year horizon.