Walking in the valley of the giants

by Martin Birch Non-Executive Chairman

With this article, I intend to introduce the reader to the Neuromation concept of artificial intelligence (AI) with particular reference to Neuromation’s specific specialty, being deep learning, and its application to the global marketplace in the 21st century. I am aware that some readers will have advanced knowledge of the technical aspects of AI, however, this paper is not meant to be a technical description of the advanced minutiae of AI, more an investigation into the business case from Neuromation’s perspective and a description of how the Neuromation ecosystem operates and how the entire ecosystem relates to the Neurotoken (NTK).

First, however, it is important to define what we mean by deep learning in the context of an AI system. Deep learning is an AI function that in essence mimics the workings of a human brain in processing data to make decisions. Deep learning is thus able to learn without human supervision, drawing upon data that may be both unlabeled and unstructured to make decisions at a very high speed. The reader may like to visualize a massively multi-layered neural network with each layer coming to a specific conclusion before passing it onto the next layer where that conclusion is then used to develop a further conclusion and so on with the whole process being carried out at ultra-high speed potentially hundreds of thousands if not millions of times to create a solution.

In order to build a deep learning methodology, it is necessary to construct the operating assumptions that will apply in order to manage the data and how this will be considered and reacted upon by the AI process, this starts out it’s life as an algorithm in the purely conceptual stage which then has to progress to a fully tried, tested and proven production model that can then be used to add value to real-world applications across a wide spectrum of industries and business classifications. The Neuromation deep learning ecosystem is thus a super-efficient and highly concentrated form of MLOps which is itself a compound of 2 disciplines being Machine Learning (the ML part) and DevOps (the Ops part) with DevOps itself also being a compound of software development and IT operations. Machine learning models are usually tested and developed in separate or isolated experimental systems leading to a theoretical conclusion at the algorithm stage.

Once an algorithm is ready to be progressed along the development chain MLops is practiced between Data Scientists, DevOps, and Machine Learning Engineers to ultimately transition the theoretical algorithm to production systems where it can be included in products or services that are then sold into the physical world. It goes without saying that the quicker and more accurately a scientific team can achieve answers from this process the bigger the market advantage will be to the company that is using MLops to accelerate its product development cycle.

In its simplest form, the three main ingredients of a successful AI / deep learning / MLops structure are highly qualified scientists to develop the original algorithms, massive computer processing power, and an advanced set of MLOps tools. The tools in this case being very specialist software for each task that allow the data scientists to construct the theoretical models and run the entire process. Obviously different aspects of AI and MLops require different toolsets and the Neuroamation suite of MLops tools has been created by our scientific engineering team to focus primarily on the deep learning aspect of AI which Neuromation calls home.

To further underscore the importance of a top-of-the-range suite of tools in MLops development cycles I would like to leave AI for a moment for a short diversion into the world of vehicle repair.

Imagine, if you will, that you have a car made by one of the quality car manufacturers, a Mercedes for example. The car is just 5 years old so is generally in decent condition but it has a manual transmission and has been used as an executive taxi in a city environment. All that stopping and starting in city traffic has worn out the transmission to the extent that both the gearbox and clutch assembly need replacing to extend the vehicle’s life. Now, while it may be possible to carry out this piece of automotive surgery with just an adjustable spanner, a hammer, and a screwdriver you run the risk of not being able to join the whole transmission together properly, sure the car might run for a while but the end result is likely to be fused clutch plates and a gearbox that contains shards of useless metal where the gear cogs used to be. In addition, the specialist tools to carry out this repair are expensive and you are only going to need them once for this one-time transmission replacement

Now, further imagine, what if you had access to a fully equipped workshop with every tool you might require, hydraulic lifts, diagnostic systems, and a full set of air guns with every conceivable attachment. Not only would you be able to complete the transmission replacement much more quickly but you would have absolute confidence in the end result as you had the tools immediately available to make sure everything fits together perfectly. Once the transmission is installed you would thus be able to drive away in a perfectly working Mercedes having had no requirement to spend a fortune on your own tools.

The Neuromation developed a suite of tools for AI deep learning MLops fulfills the same process as that fully equipped automotive repair shop. Neuromation’s clients do not need to spend money creating their own MLops tools as Neuromation has already done so for them and has the full MLops suite available on the Neuromation platform, thus Neuromation’s platform clients get all the massive computer processing power they need on-demand when they need it and a complete set of proven deep learning MLops tools as part of the package which they effectively rent on an hourly, daily, or even weekly basis to complete their projects. To further promote ease of use all Neuromation tools are open source and all tools are installed in the same cluster. Neuromation has proven that it can save a team of 5 data scientists more than $100,000 a year by using the Neuromation platform as their primary MLops development ecosystem.

Neuromation’s suite of tools consists of the following main products covering the entire ML lifecycle:

1. Data labeling

2. Data management

3. Development environment

4. Remote debugging

5. Code management

6. Experiment tracking

7. Hyperparameter tuning

8. Distributed training

9. Metadata management

10. Model management

11. Deployment

12. Testing

13. Monitoring

14. Interpretation

15. Pipelines orchestration

16. Resource orchestration

17. Access control orchestration

In order to make the Neuromation MLops solution available on-demand to clients anywhere in the world along with access to the required computing power Neuromation has created its own branded Neuromation platform through which all services and tools are made available. Industry reports attest that 80%+ of data science projects never make it into production while consuming far more resources and time than originally anticipated. The Neuromation platform-based solution massively reduces the risk of projects falling into this 80% by providing sound MLOps process and implementation by combining AI/ML practices with DevOps to create continuous development and delivery (CI/CD) of data and ML functionality that controls the costs and de-risks delivery.

The reader will not be surprised to learn, given the substantial technical barriers to entry into the MLOps arena, that the main players in the field are household names like Amazon, Google, IBM, and Microsoft Azure. In this environment, Neuromation is truly walking in the valley of the giants. Neuromation distinguishes itself in the AI field by not being a general player like the giants mentioned above but by being a specialist AI/MLOps solution focused on deep learning. Neuromation also includes its entire suite of tools into its client offering whereas common practice is to charge extra for such support and finally Neuromation has placed its price point per hour at a much lower cost than that charged by the family of giants mentioned earlier. Consequently, if an MLOps team is looking for a solution to both accelerate and manage its deep learning requirements, Neuromation is the natural choice based upon the richness of its suite of tools and its extremely competitive price offering.

By now the reader should have a good grasp of what Neuromation stands for, how the Neuromation ecosystem operates and how Neuromation brings endless benefits to its clients who are involved in utilizing the deep learning aspect of AI to give their own businesses an advantage over their competitors.

The Neuromation platform will be launched during this month (May 2021).

There is however one final point that needs addressing, how do Neuromation’s platform clients pay Neuromation to use the rich functionality of the Neuromation AI/MLOps platform?

NeuroToken (NTK)

Without exception, The NTK token is the only payment method accepted by Neuromation for use of the platform. Clients have three main options at their disposal in this regard

• Clients log onto the platform and make a purchase of NTK at the current market price for fiat currency using a credit or debit card with the order being routed straight through to the crypto market where the fiat currency is sold and the NTK purchased crediting the client’s account on the platform with the NTK.
• For large orders clients can pay Neuromation an amount of fiat currency directly which when received by Neuromation will be used to buy NTK on the market which will then be credited to the client’s account and used to pay for platform services.
• Clients can buy NTK tokens themselves directly on the market and remit them to Neuromation directly as a crypto wallet transfer.

By making the NTK token the only payment method for using the Neuromation platform a system of native demand has been put in place. The more business that is transacted over the platform the more demand there will be for the NTK token from clients of Neuromation. It is up to the reader to come to their own conclusions as to what this increase in demand for the NTK token will do to the market price of NTK.

As a tantalizing clue to what to expect in the near future, it is worth noting that a recent research report published by Emergen Research on the Deep Learning System Market has forecast a compound annual growth rate of 39.1% between now and 2028 in this sector of the AI industry. By 2028 Emergen has forecast the global Deep Learning System Market will be worth $93bn in revenues.

It is the intention of Neuromation to become the premier AI / MLOps / Deep Learning data center in Europe and to corner a big chunk of this $93bn market with all fees due to Neuromation being denominated in the NTK token.

For readers wishing to know more about NTK, there are 100mn NTK tokens in issue with just 78mn in circulation. The NTK market price can be followed on both the WhiteBIT and HitBTC exchanges.

This article is the first in a planned series of articles describing Neuromation’s business activities, future articles will feature case studies, new developments in the Neuromation ecosystem, and technical input from Neromation’s MLOps and Deep Learning computer engineering team.

Kyiv, May 2021

About the author: Martin serves as the Non-Executive Chairman of The Neuromation Group and is the Managing Partner of Eastern Europe focused investment bank, Empire State Capital Partners.