Artificial Intelligence (AI) happens to be the next big thing. AI is all about data. Datasets used for machine learning are still labeled by hand, which requires a lot of effort. This creates a lot of friction: labeling quality is not guaranteed, and the initial source data is not secured. Dbrain allows people to work together in secure, seamless, integrated processes for buying, designing, and building AI apps, from start to finish.
Dbrain is an open blockchain platform that links crowdworkers and data scientists enabling them to transform raw data into real-world AI solutions. Crowdworkers do simple tasks of data labelling and validation, and are paid instantly in cryptocurrency for their work. Data scientists use the resulting datasets to train Neural Networks (NN) and build AI apps. Businesses use existing AI solutions or specify new ones to serve their particular needs. Dbrain automates AI production and data workflow by providing efficient tools to all parties, including a web application, a Telegram bot, and a mobile app.
Blockchain technology helps us meet many of AI’s current challenges. Using the blockchain, we can confidently manage high-quality data labelling, security concerns, intellectual property rights, and international micropayments. Using existing commercial computation infrastructure allows us to build an affordable, scalable toolkit for developing and deploying AI apps.
Right now, AI is off limits to all but the wealthiest and most powerful operations. Dbrain makes AI affordable to more customers. We make AI buildable by more developers. We make AI profitable for more workers. We democratize AI.
Join us to make AI happen!
Challenges
Blockchain technology helps us meet many of AI’s current challenges. Using existing commercial computation infrastructure allows us to build an affordable, scalable toolkit for developing, integrating, and deploying AI apps.
Currently, the most important practical challenges for widespread AI adoption are a shortage of high-quality datasets, unreliable data security and unreliable data quality, and the lack of a common framework shared by all parties interested in AI production.
High-quality datasets
Large, high-quality datasets contribute more than 80% to AI application success, making these data even more important for machine learning than algorithms. Datasets are still labeled by hand and require a lot of human effort. Regardless of size, poorly labelled datasets can nullify Neural Network (NN) model function and impede AI progress in general. Leveraging the labor of online crowdworkers is the most effective solution for creating large datasets. However, existing data-labelling tools and platforms fail to ensure quality and fail to meet the current demand for AI datasets.
Dbrain guarantees quality of datasets without any work duplication. To align the incentives for crowdworkers, validators, AI developers, and data owners, Dbrain implements the Subjective Proof of Crowdwork Protocol (SPOCK), which validates data quality automatically and guarantees real-time, fair, transparent billing to workers and data owners.
Security and trust
Sharing sensitive data with third parties, and even in-house developers, poses certain security risks. AI developers can replicate third-party software within a very short time when given access to someone else’s data. Labeled data, rather than software, are the defensible barrier for many businesses. Data owners lose revenue when datasets are leaked to third parties.
Dbrain protects data owners’ interests and prevents leaks at all stages of AI app development. No matter who uploads data on the platform, the Protocol for Indirect Controlled Access to Repository Data (PICARD) protects datasets and AI apps hosted on the platform. It also allows data scientists to train AI models using datasets without downloading them, and to sell AI solutions to business clients later on. The protocol guarantees security and trust in the Dbrain community with regards to data access control and reward distribution.
Abundant crowdwork supply
In the 10 largest developing countries, the total number of internet users is close to 2 billion; with nearly 50% technology penetration, the online population is growing rapidly. The number of internet users in these countries is greater than in all other countries combined. At the same time, the World Bank estimates that there are around 2 billion unbanked people in the world. Clearly, internet connectivity reaches the developing world much faster than the banking system, and many people connected to the internet are still excluded from the global financial system. Cross-border payments via banks are expensive, slow, and location dependent. Cryptocurrencies can solve this problem by reaching any person connected to the internet.
The supply of online crowdwork is abundant globally. The World Bank estimates that in 2013 the minimum total supply of crowdwork was $239B, while the market demand in 2016 was $4.8B, or 50 times less than the work supply. Only demand limits the market growth.
Global job automation trends and widespread internet penetration in emerging markets accelerate the transformation of the classic employment model into an online work model, which is divided into small, well-defined tasks, most of which do not require any special expertise. Many people in developing countries are willing and able to perform simple jobs for less than employers from the developed world currently pay to their workers. However, compared to local wages, workers could still earn significantly more by working at crowdsourced tasks online. The only missing link to connect the two parties in this win-win game is an efficient channel for employment, payment, and job validation.
Last-mile infrastructure
Even the most sophisticated AI platform is useless without access to end users. To use AI solutions in the real world, businesses need to find AI developers, the scarcest resource on the market. Developers need access to scalable and affordable AI computation infrastructure to train and deploy their AI Apps. They also need access to raw data and crowdworkers for data labeling and model output validation. Labelers need simple, accessible interfaces, and micropayment channels to be paid for their work.
An efficient infrastructure for AI development lowers barriers for all stakeholders. Labelers and businesses benefit the most. Telegram messenger solves the problem of access for labelers. Labeling tasks vary by level of difficulty. For simple tasks, we offer the Dbrain Telegram Bot, and literally anyone with a connected device can use it to label data and earn Dbraincoins (DBR) in return.
Platform
Dbrain is an open blockchain platform for turning raw data into real-world AI solutions. We make AI accessible to businesses and allow anyone to earn money for their effort.
The platform automates most of the data preparation and human-in-the-loop workflow. It provides crowdworkers with a user-friendly labeling and validation tool based on a unique web application and Telegram bot. Blockchain protocols and internal cryptocurrency ensure transparency and fair revenue distribution among all stakeholders.
AI production line
Dbrain levels the playing field for all participants on the AI market.
For crowdworkers, we provide an opportunity to earn money for training and supervising AI networks and receive a fair share of future AI revenues securely via smart contracts. For AI developers, we lower barriers significantly for creating commercially viable AI products and provide scalable and elastic access to accumulated datasets, unique data providers, business clients, and a distributed pool of workers who create new and process existing data. We enable data providers to monetize their existing datasets and live data streams. For businesses, we offer a wide range of turnkey AI solutions, integration, and customization for particular needs.
Blockchain and crypto
The Dbrain platform works on the Ethereum network and relies on its smart contracts. We’re building a scalable permissioned blockchain anchored to the Ethereum network via state channels. Our solution can securely process thousands of transactions per second which all involved parties can verify independently. We implement two blockchain protocols for decentralized access to our platform and an in-house cryptocurrency.
SPOCK protocol
To align the incentives of crowdworkers, validators, AI developers and dataset owners, Dbrain implements the Subjective Proof of Crowdwork Protocol (SPOCK), which automatically verifies data quality and guarantees real-time, fair and transparent billing to workers and task requesters.
All work tasks performed on the Dbrain platform require multiple validations by other random labelers. Validators either do the same work for the simplest tasks such as image classification, or confirm the correctness of complex tasks. When the majority of validators agree on the task result quality, then the original worker receives a payment and a higher rating. Workers get a lower rating and no payment for rejected tasks.
All work tasks performed on the Dbrain platform require multiple validations by other random labelers. Validators either do the same work for the simplest tasks such as image classification, or confirm the correctness of complex tasks. When the majority of validators agree on the task result quality, then the original worker receives a payment and a higher rating. Workers get a lower rating and no payment for rejected tasks.
We have several requirements for our rating and task validation system to be able to process task completion and validation in real time:
- Online calculations: We need to evaluate work results as they arrive using only data stored publicly in our Ethereum smart contracts and in our permissioned blockchain that is accessible to relevant task requesters and workers.
- Transparency: All rating changes and billing events should be visible to task requesters and workers online.
- Reproducibility: Calculations must be simple enough that an involved party can reproduce them independently.
- Aligned incentives: The system should motivate workers to behave diligently and conscientiously by providing a good reward and punishment balance.
PICARD protocol
The Protocol for indirect controlled access to repository data (PICARD) protects datasets and AI applications hosted on the Dbrain platform and allows data scientists to train AI models using the datasets without downloading them, and to sell AI solutions to business clients later. The protocol allows data scientists to work on a contract basis as well as to contribute to community owned datasets and public kernels. It also allows participation in Kaggle-like competitions on openly listed challenges.
To ensure the safety of datasets and intellectual property ownership rights, we enforce strict access control during model training and deployment. During training, we charge only for compute time, but do not allow any external access during calculations. To protect from dataset leaks via this channel, we also do not allow developers who do not own the dataset to download trained models without additional validation or permission from the owner (e.g., one can simply copy raw input data as a model output). Dataset and model owners must approve indirect usage of their intellectual property in derived work and may ask for a future revenue share or direct payments for it.
Dbraincoin (DBR)
Dbraincoin (DBR) is implemented as a standard ERC20 token. The total number of Dbraincoins in existence is fixed. We use our coins as an internal currency that participants use to pay and receive for work, data and AI application usage on the platform.
Product
The Dbrain web application integrated with Ethereum (DApp) allows every Internet user to perform tasks, earn Crypto and withdraw it with a single click. Telegram bot for simple data labeling and task validation gives us access to crowdworker audience with the least imaginable friction. Our upcoming mobile app will provide user interface for complex tasks on smartphones and tablets and allow to collect custom data from crowdworkers.
We will release our UI framework publicly, so developers and businesses can create tasks and adjust them to their particular needs at their discretion. The web app is integrated with Ethereum and allows any user to withdraw or deposit funds instantly and verify transactions from our private blockchain.
Web application
The Dbrain web application integrated with Ethereum (DApp) provides an intuitive tool for data labeling and validation tasks for crowdworkers. The complex user interface allows crowdworkers to perform advanced tasks, such as image labeling for classification and regression, object annotation with bounding boxes and segmentation masks.
Label data for AI and get Crypto
Do simple image labeling and data validation tasks, get paid instantly with Dbraincoins (DBR) and withdraw your earnings anytime.
Toolkit for data scientists and businesses to develop and deploy AI Apps will be out by Q3 2018
Telegram bot
The Telegram Bot is ideal for simple image labeling and validation tasks. Anyone with the internet connected device can label data and get paid instantly with Dbraincoins. Smartphones are more accessible and widespread in developing countries than laptops, while internet penetration is high, which gives us an edge in accessing workers in those regions.
Easy way to label data
With the Telegram bot, we reach millions of unbanked people to give them an income stream in Crypto.
Mobile application
Mobile apps are a great tool to create new data — video, audio, photos, acceleration, GPS coordinates and touch input. Our app will allow any platform user to become a data provider and earn additional Dbraincoins.
Coming Soon
Apps Store & Google Play
Competitive advantages
There are a few large existing crowdwork platforms that are mostly used for AI-related tasks (e.g., Amazon Mechanical Turk, Yandex.Toloka). However, they fail to meet most of AI developers’ needs . Developers need to find raw data and upload tasks or validate data themselves. Then developers need access to AI compute infrastructure in order to train models and build AI Apps. Finally, developers need to deploy AI Apps somewhere to make them available to business clients. The Dbrain platform covers all stages of the AI production cycle and offers a comprehensive solution.
Use cases
The Dbrain platform provides a scalable and accessible infrastructure to supercharge businesses with high quality AI, integrated via a convenient API. We offer a wide range of turnkey and custom AI solutions, integration, and customization for our clients’ particular needs. Static image recognition, video surveillance and action detection, medical data processing, and content analysis of text streams, which currently lack working solutions would benefit from business-ready AI solutions. These areas account for almost half of the future AI market; they are our target.
Image recognition
Image recognition (including classification and tagging) is one of the most commonly applied AI use cases today. Image recognition is an area that is developing rapidly and that will have a major impact on the consumer, automotive, advertising, healthcare, defense, media, and entertainment industries. People communicate in images, and images are essential for product discovery nowadays. Businesses spend billions every year on repetitive graphic design tasks.
Video surveillance
Many clients are interested in AI solutions for video surveillance. Governments, retail shops, production companies and private security companies can use this technology. Computer vision-based systems focus on object recognition and action detection in video streams. Surveillance camera companies often rely on AI to detect dangerous or criminal behavior automatically. This industry employs several hundred thousand workers, who perform routine monitoring work that AI could automate to a large extent, thus saving billions of dollars.
Labeled data are essential for implementing AI surveillance. Many AI developers will create AI algorithms for effective surveillance systems using our platform. Surveillance companies will upload videos from surveillance cameras onto the Dbrain platform. Crowdworkers will detect and tag unusual videos and send them to the neural network. After that, the trained model will become better at detecting emergencies, fights, robberies, and similar activities.
Medical data processing
AI is rapidly changing the healthcare analytics market . AI developers can create systems that predict and detect serious diseases at early stages much better than doctors do today. AI developers need high quality and large datasets of patient data like radiology scans and clinical records about illnesses and treatments.
One of the most important AI use cases is cancer detection. Deep learning models can assist pathologists in this task. A pathologist’s report serves as a basis for diagnosing many diseases. Despite the fact that pathologists study for many years in order to improve their cancer prediction skills, even today AI can detect cancer with higher accuracy than doctors can. Like in all AI models, more and better data are needed in order to improve its performance.
Pathologists do not have a platform for data exchange on cancer detection and on many other illnesses. Dbrain will create a protocol for exchanging data between all parties concerned in different countries, with different legal restrictions.
Natural language processing
NLP is an AI application that recognizes not only formal content of texts, but also their sentiment and meaning. AI can also detect messages that signal dangerous situations, for example, a terrorist threat or suicide intention. Telegram is one of the leading messengers worldwide. It has more than 100 million active users and delivers over 15 billion messages daily. Telegram has recently been blocked in Indonesia by the government, which said that the messenger is "full of radical and terrorist propaganda". The developers of Telegram do not provide access to users’ messages to any governments or officials. Therefore, AI in combination with a human feedback loop is the only possible solution for content moderation.
You name it
We offer a wide range of turnkey and custom AI solutions, integration, and customization for our clients’ particular needs. Become one!
Revenue model
We charge a 10% commission from every transaction on our platform to compensate our costs of running the infrastructure and maintaining a healthy platform. Our commission is much lower than those charged by the existing crowdwork platforms. We believe that zero commissions are unsustainable for a large crowdwork platform. Those who promise to never charge any money for the value they add either aren’t going to build a sustainable business, or aren’t telling the whole story, or don’t add any real value in the long run.
Our AI platform will save our clients much more than the commission we charge, because they do not have to set up any infrastructure for data labeling, AI development, training, and deployment.
Roadmap
- Q3 2017. Proof of concept and research phase to define the need for a collective AI development tool
- Q4. Product development and MVP testing on the first business clients.
- Q1 2018. Public Alpha version of web application and Telegram bot to label and validate data for crowdworkers (SPOCK)
- Q2. Public Beta for training neural networks on labeled data, mobile app for data labeling
- Q3. Launch of the fully-running blockchain platform for building AI Apps with API integration for businesses
- Q4. Scaling platform to meet new markets with a focus on ever-growing AI community and client base
Partnerships
Participant of Nvidia Inception Program, includes a custom set of ongoing benefits such as an opportunity for cooperation with major AI companies and support of Nvidia during critical stages of product development, prototyping and deployment.
Dbrain will help Chronobank, a decentralized initiative to disrupt the short-term recruitment industry, create a fair and intelligent system to evaluate employees in the labor market.
Participant of Microsoft BizSpark, a global program that helps startups succeed by giving them free access to Microsoft Azure cloud services, software, and support.
Team
Dmitry Matskevich
Chief Executive Officer
Serial entrepreneur, data geek. Founded
2 leading Big Data companies. Sold Flocktory, a B2B AI startup, for $20M in 2017
Aleksey Hahunov
Chief Technical Officer
Founder of connectome.ai and R-SEPT.
3+ years in managing R&D teams in IT/AI
Ivan Gorshunov
Chief Marketing Officer
Mobile Apps product expert at Google, serial entrepreneur, co-founder of «.etc» and Rutech, hosts the biggest product community in CIS
Dima Dewinn
Chief Design Officer
Designer, founder and CEO of Thngs, 12+ years in visual storytelling and product development.
That's the explanation about DBrain which Digital Science can convey., hopefully, can be useful for friends of cryptocurrency lovers. Thanks and look forward to information about other crypto worlds.
BOUNTY PROGRAM IS ONGOING NOW ON BOUNTYHIVE
Tidak ada komentar:
Posting Komentar