Letter from the CEO
“Individual commitment to a group effort — that is what makes a team work, a company work, a society work, a civilization work.” — Vince Lombardi
I am excited to announce the launch of the Neuromation Marketplace, offering all community members the opportunity to utilize Neurotokens (NTK) to purchase AI products and services. The launch of the marketplace marks a significant milestone for Neuromation and we appreciate all of your incredible support throughout our development journey.
Neuromation was started with a vision to democratize AI, enabling people and businesses everywhere to contribute to and benefit from AI technologies. To achieve this vision, we focused on addressing the key bottlenecks in AI development related to access to data, computation & machine learning models.
Modern machine learning models require vast amounts of labeled data and many companies and developers are limited by the cost & complexity associated with acquiring, preparing and labeling this data. Neuromation is addressing this bottleneck by enabling access to providers of Synthetic Data on its marketplace. For the first time, developers can use their NTK to purchase or create Synthetic Data for their custom needs.
To address access to computation and machine learning models, Neuromation has developed a best-in-class platform enabling the rapid creation of cutting edge AI models and simplified management of cloud resources and environments for model training. With a focus on developer experience, the platform provides a streamlined and scalable cloud-agnostic infrastructure. Seamless and secure collaboration is also enabled through sharing datasets, experiment results, environments, and live development sessions. In the technology section of this quarterly review, we will outline the key features and technologies of our platform.
We welcome all NTK holders to visit the marketplace home page where our team of AI strategists will work directly with you to meet your AI needs. NTK can now be used to order custom-generated synthetic datasets, procure computing resources via our best-in-class developer platform and develop custom AI models. The team has worked hard over the last year to develop an optimized developer platform and we are excited to open it up to the community. The team is excited to work with our community of tokenholders and support your marketplace needs.
Moving forward, we are committed to growing the marketplace and will be looking to both partner with and invest into companies that can provide utility for our token holder community.
Thanks again for the great support and we look forward to continuing our journey toward democratizing AI.
Announcing The Neuromation Marketplace
It is our pleasure to announce the launch of the Neuromation Marketplace. Machine Learning Models, Data and Computational Power are the three pillars of AI. With this announcement we are now offering a new and effective approach for connecting supply and demand in each of these domains, allowing NTK holders the ability to purchase AI Services, compute resources and Synthetic Data. All services on the Neuromation Marketplace can be purchased using NTK.
The AI Services available via the marketplace include data labeling, model development, AI strategy consulting and fundamental research.
Marketplace users will also have the ability to rapidly and seamlessly provision the cloud resources, services and storage necessary for the training of ML models via our best-in-class developer platform. Following last quarter’s successful beta test of the platform, we are now opening access as well as offering cloud computing resources for training your deep learning models to all NTK holders.
Synthetic data creation services are also available for purchase with NTK via the Neuromation Marketplace. Synthetic Data, or digitally created data that mimics real-world sensory input, has been a core focus area for Neuromation since our founding. Leveraging both procedurally generated techniques (i.e. CGI) together with new generative neural network architectures (e.g. GAN’s), Neuromation’s technology enables the development of vast, diversified and perfectly-labeled data for training modern AI algorithms.
The Neuromation Marketplace brings together the competencies of researchers, engineers, and designers in one place, offering a new and effective approach for connecting supply and demand in the fundamental domains of AI development — Machine Learning Models, Data and Computational Power.
Our AI specialists are standing by to discuss your needs and we look forward to hearing from you.
Further information is available on the Neuromation Marketplace Page.
Following Q1 2019’s successful expansion of our platform beta testing program to a wider group of strategic partners, Neuromation is now ready to open access of our developer platform to the broader community of NTK holders via the Neuromation Marketplace.
Functionality of the developer platform is expanding rapidly and the technical team continues to work on expanding capabilities for users to run machine learning payloads in their own cloud environments.
Following our successful implementation of Google Cloud (GCP) support, with dramatic simplification of infrastructure setup and orchestration for AI development, we are making progress toward our ultimate goal of being cloud-agnostic with full AWS support expected in an upcoming release.
The team is also actively working on further optimization of cloud deployments, including simplified onboarding, faster job start times and the preloading of environments with languages and libraries.
The team has also done significant work on further reduction in our customers’ operational overheads via the creation of draft ML recipes for recurring ML tasks, which are already being tested by our internal research team in the current quarter.
Learnings from our beta testing program continue to provide valuable and encouraging feedback. The desire for “just-works” simplicity from our users is a constant motivating force, and we have received great feedback on our collaboration features and pre-defined development flows. We have also been encouraged to hear that in addition to facilitating AI development capabilities for the broader talent pool of software developers, that our platform is also receiving high marks from experienced AI researchers who may already be experts at creating algorithms and theory, but nonetheless still struggle with back-end and devops management.
We now look forward to welcoming our NTK holders to take advantage of the capabilities of the Neuromation Platform via our new Neuromation Marketplace offering, and to many more successful releases to come in the future.
During the quarter, Neuromation signed two MOUs with existing customers, both of which represent a successful expansion of our initial engagements beyond successful proof of concept and toward fully-functional production AI applications.
According to our specialists in the field, market interest in AI computer vision applications remains high. And while our customer engagements are still primarily related to object and facial recognition, recent customer inquiries reflect an increased market demand for advanced human analytics, including emotion recognition, eye tracking, facial attribute analysis, facial recognition and gesture recognition. The company is also seeing traction for its advanced labeling capabilities, specifically for features which are extremely difficult or impossible for human labelers to estimate accurately, such as depth, velocity and lighting.
During the quarter our customer team in Tel Aviv attended the growing EcoMotion conference for the smart transportation sector, where we discussed the use of synthetic data in autonomous driving applications. They also participated in Israeli pharmacy chain Super-Pharm’s innovation day where they discussed our work in automated retail solutions using AI technology.
Customer Use Case: Greenback Recycling Technologies
Emerging standards will compensate brands with pre-paid rebates if they can demonstrate their products are being properly recycled. Greenback aims to build systems to identify and track consumer goods at recycling centers by using computer vision AI to identify specific retail goods on the recycling conveyor belt. This process is made particularly difficult by the damaged condition, shape variation and missing elements of the discarded objects in this target environment.
Neuromation developed an extensive synthetic dataset with randomized deformations of the models and general domain randomization; object detection models were trained fully on the synthetic dataset; produced demo applications on standard mobile devices that showcase the models.
This quarter, the primary activity of Neuromation’s research department was to work on several internal projects related to our main focus, synthetic data. We expect results from these projects to begin to be released in Q3 and see many interesting things coming up for Neuromation on the research front.
During Q2, the research department was pleased to have one of its research papers accepted at the top conference on natural language processing, ACL 2019, which will take place in Florence, Italy from July 28th to August 2nd, 2019. This paper was based on our collaboration with the runners up in last year’s NeurIPS ConvAI2 competition, which was won by Neuromation researchers Sergey Golovanov and Rauf Kurbanov. This successful collaboration resulted in an even better performing model than the one we submitted for that competition, with its advantages supported in the paper by a comprehensive experimental comparison. We will be publishing a detailed description of the paper when it is finally released and we visit the ACL conference in late July.
We also released a new paper on medical imaging titled “Breast Tumor Cellularity Assessment using Deep Neural Networks” in collaboration with Vladimir Iglovikov. It is up on arXiv now, and we are looking for a suitable venue to publish.
On May 22–23, Chief Research Officer Sergey Nikolenko visited the LDV Vision Summit where he discussed the advantages of synthetic data in computer vision applications. Machine Learning Engineering Lead Kyryl Truskovsky also attended the Production AI Conference, where he led the workshop “Knowledge Distillation for Model Compression”. He also participated in a Kyiv School of Economics public discussion panel on AI, business, and education. Finally, our team visited CVPR where we made a lot of contacts in the field of computer vision applications for synthetic data.
Neuromation also began a more serious foray into AI education projects during the quarter, when researchers Sergey Nikolenko and Alex Davydow taught a three-week intensive course devoted to deep learning for computer vision at the Harbour Space University in Barcelona. We hope to collaborate with Harbour Space more in the future.
New publications with Neuromation affiliation (Neuromation employees in bold)
- S. Golovanov, R. Kurbanov, S.I. Nikolenko, K. Tryskovskyi, A. Tselousov, T. Wolf. Large-Scale Transfer Learning for Natural Language Generation. Accepted to 57th Annual Meeting of the Association for Computational Linguistics (ACL 2019), 2019.
- S. Golovanov, A. Tselousov, R. Kurbanov, S.I. Nikolenko. Lost in Conversation: A Conversational Agent Based on the Transformer and Transfer Learning. Accepted to NeurIPS 2018 Competition Volume.
- A. Rakhlin, A.A. Shvets, A.A. Kalinin, A.Tiulpin, V.I. Iglovikov, S.I. Nikolenko. Breast Tumor Cellularity Assessment using Deep Neural Networks. arXiv:1905.01743, 2019
Q2 saw a significant acceleration of positive price movements in the major crypto markets, led by BTC which finished the quarter up 163% and even over the symbolic $10,000 resistance level, and ETH up 105%. The total market cap of all major cryptocurrencies grew during the quarter by 139%, emphatically suggesting a return to bull market status.
Following a 167% price increase in Q1, NTK was down 56% in the quarter in dollar terms. We cannot point to specific newsflow during the quarter that instigated this drop and believe the price movement is largely due to several large holders liquidating a portion of their holdings, with both smaller and new investors taking advantage of this dynamic to enter at lower price points.
If we look at the cohort of 2018 ICOs to which Neuromation belongs and which we have been tracking over previous quarters, the overall performance of the group was up, with over 50% of tokens seeing price increases during the quarter. This may be due to product development cycles of these companies reaching fruition and their utility tokens beginning to anticipate active usage, or it may simply be due to a bounceback from general underperformance of the group in the last quarter.
While NTK was the number one performing token in the group last quarter, up by 150%, this quarter it was in the bottom quintile of performance.
As we have said in the past, the sole usage of NTK is for the stated purpose of purchasing AI development services. Earlier in this report, we announced for the very first time the availability of the Neuromation Marketplace, which will allow for NTK to be exchanged for various AI development services. We encourage all holders of NTK who are interested in such services to visit the Neuromation Marketplace website and to get in touch with our AI specialists to discuss your needs.
In addition to these updates, Neuromation is providing specifics on two key issues related to the NTK. These are: