doc.ai
doc.ai
Token: NRN


Join the Decentralization Revolution

ICO dates
Start date: 2017-09-28
End date:

Registrated in: Undefined

Platform: Ethereum
Type: ERC20

PREMIUM ICO

HARA

Empowering billions by utilizing a global and transparent data exchange.
Ends in 264 days 23 hours
Goal: 17,000,000 USD
Price: 1 HART = 0.059 USD

Dominium

Dominium Blockchain – The one-stop-platform for everything to do with property anywhere in the world!
Ends in 233 days 23 hours
Goal: 17,500,000 USD
Price: 1 DOM = 0.25 EUR
doc.ai categories
Other
doc.ai whitepaper
Video
doc.ai token sale
KYC passing required No | Whitelist No | Restriction for countries No
Token distribution in ICO
37%
37%
Acceppting ETH, BTC
doc.ai news, social
doc.ai search trends in Google
Random whitepaper excerpts

Contents
1 Introduction
4
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Blockchain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Decentralized Artificial Intelligence . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 An Edge Learning Network
4
2.1 Post-Cloud: Towards Fog and Edge . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Multi-agent Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Artificial Intelligence Today . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.5 Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.6 Proposal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.7 Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
...

List of Figures
1 Learning phases (Loop) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Learning phases (Sequential) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3 Proposal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4 Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
5 Payout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
6 Downpour SGD: Single Client . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
7 Downpour SGD: Multiple Clients . . . . . . . . . . . . . . . . . . . . . . . . . . 9
8 Personal Portable Biology File . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
9 Onboarding module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
10 Blood Test Decoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
11 Genomics Test Decoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
12 Me...

1 Introduction
1.1 Overview
Mobile phones and tablets are now the primary computing devices for many people. In many cases,
these devices are rarely separated from their owners, and the combination of rich user interactions
and powerful sensors means they have access to an unprecedented amount of data, much of it
private in nature. Models learned on such data hold the promise of greatly improving usability by
powering more intelligent applications, but the sensitive nature of the data means there are risks and
responsibilities to storing it in a centralized location. NeuRoN is an Ethereum-based blockchain that
tackles the paradigm of decentralizing artificial intelligence.
1.2 Blockchain
The blockchain enables parties to interact based on a set of agreed upon rules. For example, such
a rule can define payment transfers. These generic rules are refered to as smart contracts. The
Ethereum network [1] is a platform that enables the creation of peer-to-peer applications based on
smart contract...

2.2 Multi-agent Systems
Multi-agents systems are used to solve problems that are difficult to solve by individual agent.
Multiple-agent communication technologies can be used for management and organization of com-
puting fog and act as a global, distributed operating system, which combines decentralized P2P
general-purpose computing tasks distribution, multiple-agents communication protocol and smart-
contract based rewards, powered by a blockchain.
The Blockchain of EVM is an example of swarm intelligence. May miners work as reactive agents,
who without any control of agent-supervisor carry out the work on maintenance of the network,
moving only in accordance with their own motivation [3].
2.3 Artificial Intelligence Today
Current work in machine learning has shown that larger models can dramatically improve overall
performance [4]. With the advent of deep learning, the field is rapidly expanding. However, large
neural network models face infrastructure limitations. These limitations can be ov...

Figure 1: Learning phases (Loop)
Figure 2: Learning phases (Sequential)
2.6 Proposal
The proposal phase, shown in Figure 3, consists of a broadcast of the curriculum or the dataset
available for training. When the server determines there are enough peers with a particular curriculum,
the server will initiate an proposal. This requires the server to select the users or peers that they wish
to initiate a learning with, and to undergo an initial teaching. In order for the stochastic gradients to
be properly calculated, the dataset which is iterated over must be of a fixed sample size. As a result,
the users must be present for the duration of the learning and the curriculum or dataset must be fixed.
The initial teaching phase consists of an initial weight being calculated.
This initial weight, along with the model, is passed to the users for the teaching phase. For the
proposal period, the server must determine how it would like to compensate its users. Users can be
paid per epoch or for completing ...

Figure 3: Proposal
The user will teach the neural network based on the locally stored dataset. The neural network will
return only its calculated delta in the weight between the knowledge it initially entered with and the
knowledge it has attained for the day (or iteration on the dataset), shown in Figure 4.
The neural network will visit every user’s device until it has learned or calculated a stochastic gradient
descent weight from each. This completion will signal a full epoch on the dataset.
Figure 4: Training
2.8 Synchronization
A synchronization occurs after the neural network has completed an epoch. This consists of the server
aggregating all of the weights or learnings and calculating a new model or a new initial weight to
provide to the network. This initial weight is again broadcast to the users if the curriculum and users
are still available for a second epoch.
2.9 Payout
During the initial proposal period, users are provided with the amount in tokens that the server
is wi...

Figure 5: Payout
order to limit bad behavior. If a malicious party attempts to perform a DDoS or Sybil attack on the
network, it will cost a fee per transaction. If the fee is not attractive to the network of miners, the
transaction will simply not be accepted in a proposed block. If the attacker proposes transactions
that contain an attractive fee, the fee acts as a disincentive to propose the transaction. If the attacker
proposes a high volume of transactions in a short period of time, the demand on the mining network
will increase. This will increase the cost of participating in the network [6].
NeuRoN will require a similar mechanism as the network will not be able to objectively assess the
quality of data or the intent of every participant in the network. In order to reduce the potential for
DDoS or Sybil attack on the network, a buy-in will be required to participate with the network in
order to enforce a monetary barrier to entry to DDoS or Sybil attack. These tokens will be returned at
the end of ...

Figure 6: Downpour SGD: Single Client
Figure 7: Downpour SGD: Multiple Clients
Algorithm 1
Downpour SGD: Processing by worker
i
Initialize:
w
i
=
w
˜
repeat
Receive
w
˜
from the master:
w
i
←−
w
˜
Compute gradient

w
i
Send

w
i
to the master
until
forever
3 Focused Domain: Personalized Biology
The medical field is experiencing a large influx of statistical modeling and machine learning. Already
we are discovering and increased number of insights. This will likely continue and will grant a many
benefits to millions of people. Furthermore, the exponential scaling of data will result in a positive
9
...
Brad Mills
Brad Mills
Crypto trader, Head of Algorithms
Diego Gutierrez Zaldivar
Diego Gutierrez Zaldivar
TECHNICAL ADVISOR
Blockchain Advisor
CEO & Co-founder RSK
Diego Gutiérrez Zaldívar
Diego Gutiérrez Zaldívar
CEO & Co-founder @ RSK Labs
Joey Krug
Joey Krug
Co-Founder & Senior Front-end Developer
Pantera Capital Co-Founder Augur
Founder of Augur
Jonathan Smith
Jonathan Smith
Co-founder & CTO Civic, Identity/KYC Advisor
Advisor
CTO & Co-Founder at Civic Technologies
Matthew Roszak
Matthew Roszak
Co-Founder Bloq
Genaro Advisor
Chairman & Co-Founder
Walter De Brouwer
Walter De Brouwer
CEO Doc.Ai