Neuromation Q4 Report
Contents:
- Letter from the CEO
- Introduction
- Development Update
- Client Update
- Research Update
- NTK Update
Letter from the CEO
We are what we repeatedly do. Excellence, then, is not an act, but a habit. — Aristotle
In 2018, Neuromation went from a breakthrough idea to a global technology company pioneering cutting edge technologies and building best-in-class products. The tremendous support of our token community has enabled the growth of the company and we are sincerely grateful for your strong, sustained support.
Over the last year, the company has focused on building the necessary foundation for sustained success, and we start 2019 with a strong team, clear strategic direction, mature development processes, and disruptive core technologies.
Our numerous scientific publications and popular NeuroNuggets blog have helped establish Neuromation as a thought-leader in AI. We pioneered
the ‘Synthetic Data’ space and were profiled by both Wired and Techcrunch as ‘tech disruptors’.
We key-noted several notable conferences including the LDV Vision Summit in New York with a notable presence at the Tokyo AI Expo and Future of AI conference in Tel-Aviv. Most importantly our technical contributions were recognized at CVPR and NIPS, the two most prominent AI conferences in the world. Our teams won key global competitions, further establishing their world-class AI capabilities.
Our development efforts focused on driving an order of magnitude improvement in the developer experience. Over the last quarter, our Research and Engineering Teams have exclusively used the Neuromation Platform leading to a marked improvement in productivity and decreased model development time. In the coming quarter, we are excited
to roll out the product to a growing base of external users.
In 2018, we hired world-class talent to lead our research, engineering, finance, operations, sales and marketing functions. I am continually inspired and impressed by the talent and dedication of our wonderful team and in the new year we will look to invest in our go-to-market and commercialization efforts.
Thank you again for all your wonderful support over the last year. We are excited for all that is to come in 2019 and look forward to the journey ahead together.
Best,
Yashar Behzadi CEO, Neuromation
Introduction
Neuromation is pleased to provide our Quarterly Report, for Q4 2018. The purpose of this report is to provide current and future customers, NTK holders, supporters and other community members ongoing information regarding the progress and goings on at Neuromation.
In Q4, Neuromation added several important enterprise customers of its custom AI development and Synthetic Dataset creation services, including a multinational electronics manufacturer, a major connected digital display manufacturer and firms in the healthcare equipment and waste recycling industries. All of this client work is currently being carried out 100% on the Neuromation Platform as part of our internal Alpha testing program, providing invaluable early usage and testing data.
Neuromation also made a key executive hire in the quarter, bringing onboard Andreas Wiese as Chief Marketing Officer.
Andreas joins us from MongoDB (NASDAQ: MDB) where he was responsible for growing their cloud business leading up to a successful IPO in 2017. Prior, Andreas held key leadership positions at AppDynamics, Sumo Logic and Mesosphere, Inc. His extensive experience in product, growth and global go-to-market for enterprise software companies is a major win, and we are very excited to have him drive Neuromation’s growth in the enterprise AI space.
As we will show in the Development Update below, development on the Neuromation Platform is progressing according to schedule and we are actively adding new features and improving developer experience as we prepare for wider distribution. Private Beta testing is expected to begin in the current quarter, Q1, as scheduled, followed by expanded Beta testing in Q2. As a result of better insights gained from internal use and our initial enterprise customers, we are now targeting a full release of the Platform later this year. This schedule allows additional time for development, improved stability and more thorough testing.
On the research front, several new papers by Neuromation scientists were published or presented at international conferences on topics including molecular generation models and natural language understanding for health- care applications.
Finally, we will also include an update on NTK and the broader crypto market environment and explain in more detail some of our policies regarding NTK.
Development Update
In the last Quarterly Report, we outlined the development roadmap for the Neuromation Platform. This included Alpha testing of the Platform using our internal Research team in Q4, which has been completed successfully. Beta Testing in Q1 began as planned, and we are on track to launch the Platform later in 2019. Throughout the past quarter, Neuromation’s development team has been hard at work on building out the Platform and we are happy to announce that we are on track with all commitments. Our Q2 launch will be limited to strategic partners only, which will allow us to laser focus on their specific needs. In this way, we can be sure that the Platform’s functionality will be more fully in place and the product will already be in active use and meeting the needs of a meaningful sample of the AI development community when we open the Platform to all users later in the year.
As the core architectural principles, feature set and design for the Platform have already been laid out, active development work has been in full swing for approaching 5 months now. Internal Alpha testing is ongoing both with Neuromation research projects as well as client projects, and we have reached the stage where the full lifecycle of AI development including model retrieval, alteration & debugging, testing & iteration, and training on large numbers of GPUs are all being regularly accomplished and accelerated using the Neuromation Platform.
Specific progress made during Q4 included improved security, network isolation, the addition of persistent logging and real-time telemetry and usage metrics. We have maintained a strong focus on user experience, which has seen distinct improvement via improved debugging and use of interactive sessions and a remote interpreter. Web UI features are being added as needed to complement the command line interface (CLI).
Client Update
As we announced in August of 2018, in addition to the industry research partnerships Neuromation was already conducting at that time, the company began to actively respond to interest for paid customer engagements for custom AI development services.
Enterprise demand for Neuromation’s AI development, Synthetic Datasets and Platform capabilities has been encouraging and our team of AI strategists in Tel Aviv have now developed dozens of specific proposals and are currently at work on 4 major client projects in coordination with our AI Research and Engineering teams. Additional conversations regarding development work are ongoing and a robust pipeline of potential customers in the technology, manufacturing and financial services industries in the US, Europe and Asia is being carefully managed.
We are truly thrilled with the caliber and industry positions of these initial engagements.
As is the case with our ongoing industry research partnerships, these client projects ensure that Neuromation is continually solving real-world problems using the latest advances in the eld, and also that the tools we are creating to facilitate AI development are being used by leading AI practitioners, creating a positive feedback loop that we are con dent will hasten the democratization of AI and spur creation of a community with the potential to change the world for the better. To honor client confidentiality, we are unable to publicly site specific client names at this time.
Projects
Publicly listed Japanese healthcare equipment company
Neuromation is developing a state of the art object recognition system for reliably recognizing different types of surgical instruments.
European recycling company
Building an object recognition system and producing Synthetic Datasets to recognize different types of recycled plastic or plastic waste for recycling plants.
Japanese connected digital display manufacturer
Building an object recognition system for retail products utilizing both real and Synthetic Datasets.
Electronics manufacturer
Synthetic Dataset generation for facial recognition system, ensuring improved diversity and representativeness of the dataset in multiple dimensions.
Research Update
Q4 2018 was a great quarter for Neuromation Research. The Neuromation Research Team delivered talks at major conferences and had several new papers published or submitted to major venues.
Key events during the quarter included Neuromation researchers participating in NIPS 2018, the largest and most prestigious conference in machine learning and specifically deep learning; participation in AI Ukraine, the conference where Neuromation was born some years ago; and delivery of several key presentations at Basel Life, one of the largest and most prestigious forums in healthcare, among many other events.
The Research Team also finished a large and important project in collaboration with Insilico Medicine, for which we created a baseline Synthetic Dataset and conducted a large experimental comparison of generative models for drug discovery. A joint paper about this project, nicknamed MOSES, has already been published on arXiv and is awaiting publication in a major journal.
We are pleased to announce that Neuromation’s Sergey Golovanov’s team won the ConvAI2 Competition at Neu- rIPS 2018. Winning a NeurIPS challenge is one of the top honors in competitive data science, and while we have never doubted we have an excellent team in data science, this kind of validation surpasses all expectations; see our post for more details.
“I am eagerly looking forward to 2019. I am sure it will be an excellent year for Neuromation, and a great year for Neuromation Research. Good luck!” — Sergey Nikolenko, Chief Research Officer
Neuromation and Insilico Medicine at NeurIPS 2018.
New publications with Neuromation affiliation
D. Polykovskiy, A. Zhebrak, B. Sanchez-Lengeling, S. Golovanov, O. Tatanov, S. Belyaev, R. Kurbanov, A. Artamonov, V. Aladinsky, M. Veselov, A. Kadurin, S.I. Nikolenko, A. Aspuru-Guzlik, A. Zhavoronkov. Molecular Sets (MOSES): A Benchmarking Platform for Molecular Generation Models. arXiv:1811.12823 [cs.CV], 2018.
E. Tutubalina, Z. Miftakhutdinov, S.I. Nikolenko, V. Malykh. Sequence Learning with RNNs for Medical Concept Normalization in User-Generated Texts. arXiv:1811.11523 [cs.CV], 2018.
Z. Miftakhutdinov, E. Tutubalina. Deep Learning for ICD Coding: Looking for Medical Concepts in Clinical Documents in English and in French. International Conference of the Cross-Language Evaluation Forum for European Languages (CLEF 2018), LNCS vol. 11018, pp. 203–215.
Research Blog Posts in Q4
Neuromation Team at the Basel Life
NeuroNuggets: CVPR 2018 in Review, Part I
NeuroNuggets: CVPR 2018 in Review, Part II
NeuroNuggets: CVPR 2018 in Review, Part III
Neuromation October Events in Kyiv
State of the Art in Domain Adaptation (CVPR in Review IV)
What’s In a Face (CVPR in Review V)
NeuroNuggets: Cut-and-Paste in Deep Learning
MOSES: A 40-Week Journey to the Promised Land of Molecular Generation
Neuromation Researchers Won the NeurIPS ConvAI2 Competition!
Creating Molecules from Scratch II: AAE, VAE, and the Wave Transform
During Q4 Neuromation researchers participated in the following events:
Basel Life
Sergey Nikolenko delivered a presentation titled “Deep Learning and Synthetic Data for Healthcare” where he discussed many of the major obstacles to AI adoption in healthcare, including the lack of sufficiently large datasets, lack of labeled training data, the dif culty of explaining results, and the risk of systematic bias. He then demonstrated some key advantages of Neuromation’s synthetic data approach in this eld.
In the area of drug discovery, Dr. Nikolenko presented a project in which our researchers are trying to introduce novel augmentation or data transfer techniques using GANs and other generative models. Neuromation is currently collaborating with Insilico Medicine to generate fingerprints of molecules likely to have desired properties using conditional adversarial autoencoders.
Computer VISIONers Conference for computer vision engineers.
Lead Researcher Kyryl Truskovskyi delivered a presentation on “Data Augmentation with GANs”, discussing the latest advances in synthetic data generation and their application in various industries.
NeurIPS 2018 Conference, Montreal, Canada.
A team with Neuromation researchers won the Conversational Intelligence Challenge 2 (ConvAI2), an official competition of this year’s NeurIPS 2018 conference. Sergey Golovanov, AI Researcher at Neuromation, together with his friend Alexander Tselousov formed the “Lost in Conversation” team and they won the ConvAI2 finals! Description of how the challenge was completed.
AI Ukraine Conference
Sergey Nikolenko, Chief Research Officer of Neuromation, spoke on the topic of “Synthetic Data and Generative Models in Deep Learning” presenting a survey of current state of the art in synthetic data; Kyryl Truskovskyi, Lead Researcher of Neuromation, ran the “Data Augmentation with GANs” workshop showcasing continuous integration techniques using Neuromation platform for developing generative adversarial networks for the class imbalance problem.
NTK Update
Q4 saw cryptocurrencies fall dramatically across the board, while relative market share between the two largest currencies, BTC and ETH stayed roughly stable. The total market cap of all major cryptocurrencies fell during the quarter by 41% from $222B to $130B, shedding a total of $92B in capitalization. Since the end of the quarter, in the first month of the year, these basic trends have continued with the total cryptocurrency market cap down an additional 12%, while BTC has improved it’s share by 1.4% and ETH has deteriorated relative to total by -1.3%.
Against this backdrop, NTK was down 26% during the quarter in ETH terms and lost 58% of its value in dollar terms, versus ETH down 43% and BTC down 43.5% in dollar terms. Since the end of the quarter, NTK has recovered 24.5% in dollar terms but is still down 47.39% as compared to the beginning of Q3.
If we compare the price of NTK to the cohort of the largest ICO’s of 2018, we see that NTK is once again right in the middle of the pack in Q4. Only
3 of the 27 tokens we track in this cohort were positive for the quarter (two of these were crypto lending platforms, and one was in crypto mining that has since given up those gains). Of the remaining companies, token performance in the quarter ranged from -30% to -76%. Since the end of the quarter, in the first month of the year, NTK’s relative performance with this cohort has improved, placing it in the top third of the group.
As described in the Q4 Development Update above, our current development timeline foresees Private Beta testing of the Neuromation Platform to begin in Q1 2019. General availability of the Platform is currently forecast to occur later in 2019. At that point, we will begin to see the rollout of the Neuromation Platform to the broader public and should begin to see increased use of NTK on the Platform. Once live, all transactions on the Neuromation Platform will be transacted in NTK. Regarding revenue earned from client engagements, all Neuromation revenue from AI development work will be transacted in NTK on the Platform throughout 2019. Finally, Neuromation pledges to always consider the impact on token holders of all new business ventures or product offerings.
Neuromation believes strongly in transparency and communication with our token holder community and will continue to report on our progress in terms of development and research activities on a quarterly basis.
In addition to these updates, Neuromation is providing specifics on two key issues related to the NTK. These are:
