7.3 Features and limitations

Off-chains are useful in many applications, where they are a strict improvement over doing operations under on-chain. However, we should keep in mind, that every technology has its advantages and disadvantages. To sum up, the above research here is the crucial points list:

· Off-chain is centralized. In case, you lost your running off-chain nodes (for instance, due server restart, or database unavailability), all operations on off-chain will stuck, and users won't be able to process their transactions

· They're particularly useful when you need to handle a large number of transactions for a long time. The key point here lies in cost per each update. With off-chain, you don't need to pay the fee for each transaction, which extremely reduces costs.

· Off-chains are the best work for applications with a predefined set of participants. This is because the contract, which is responsible for tracking the changes on on-chain should be aware of each participant. In case, we remove/add participant on the platform - this will cause extra updates to contract and extra spends as well.

· Off-chain should be private. As the off-chain use simplified consensus, this may cause issues, if the particular node will start sending wrong transactions.

· Off-chain deals have instant finality. That means as soon as both parties sign a state update / or block has been formed (depends on specific off-chain implementation), it can be considered final. Both parties have a very high guarantee that, if necessary, they can “enforce” that state on-chain.

7.4 IPFS as a second storage engine

IPFS stays for the interplanetary file system. It's a peer-to-peer protocol where each node stores a collection of hashed files. A client, who wants to retrieve the specific file, can call this file by its hash. The IPFS will go through all nodes on the network, asking for a file by the provided hash. In case, he found that node - then user receive this file.

The approach is similar to BitTorrent: it's a decentralized technology, which allows you to store and refer to files by their hashes. However, the IPFS is also staying for the rapid web. What does it mean? Well, that means, in case, you are the file holder. Someone requested your file through his node. The file cached on his node, then another user asked for this file, it's cached as well. So, when your node is offline, you are still able to access your file. But, let's assume, that your node stay offline for about a month, and nobody else asked for this file, even the users, who requested a month ago. What will happen? This file will be deleted from that node, where it has been cached because nobody asked him for a quite long period. This is where the rapid web is: the particular resource is treated alive as long, as someone requests it. Otherwise, the resource is dead, and it doesn't matter - can you fetch it or not, as you don't plan to use it.

The fascinating part of IPFS beside it an ideology is how does the hash forms. All hashes in IPFS looks like that:

QmRZxt2b1FVZPNqd8hsiykDL3TdBDeTSPX9Kv46HmX4Gx8

This kind of hash is a multihash or self-told hash. If you previously used IPFS, you may notice, that all hashes start with Qm. The Qm is a code number of base58 encoding. Which means that our previous example original hash is:

RZxt2b1FVZPNqd8hsiykDL3TdBDeTSPX9Kv46HmX4Gx8

7.5 File lifecycle

As we have pointed out earlier, the replication happens when someone requests the particular file from another IPFS node, where it hasn't been presented earlier, or where it has already been cleaned up. However, as with BitTorrent, here we have a master node. The master node - is a node, from which file has been uploaded. It's assumed, by default, that on master node file will stay permanently. Other nodes can hold a replica of your file. But they will store the replica only when you ask them (by requesting the file by hash from this node).

The following diagram displays the workflow:

Pic.13. IPFS file upload

On the chart listed above, we see the basic file upload flow. Now let's assume, that node 1 is online, but we don't aware of it, we only know the file hash and want to obtain it via node 2:

Pic.14. IPFS file discovery

According to this flow, we have asked our IPFS node 2 about the file mentioned above. Of course, this node doesn't have it, that is why it starts asking all available nodes for this file. Let's assume, that all known nodes to node 2 are not aware of this file, but each node may have a connection with other nodes (about which node 2 is unaware). They start asking other nodes, and finally, one of the nodes return this file. During the return process, in our example, only initial requester - the IPFS node 2, will cache this file.

7.6 Security Issues

The main issues about storing private content on IPFS is that in case someone will grab your IPFS hash - he will be able to retrieve your confidential information. However, nothing prevents you from encrypting the file. In our case, we assume, that user's data may be used by authorized third party (for instance, when we want to pass our recorded brain waves activity to the doctor).

For this purpose, we can follow the asymmetric cryptography encryption. The simple flow will involve two users participation: user (client) and doctor (to whom the data will be sent). The only thing which both sides need to know are the public keys of each other and IPFS hash itself.

According to our platform flow, the exchange of IPFS hash is possible through the smart contract. However, both sides need to agree to provide the public key to each other.

As a result, we have achieved the max security level and reduced costs on passing information between users.

8. Economic model

As we have already pointed out - the use cases of the platform may vary, and not only restricted to detecting the concentration level of the driver. This makes the technology useful for the medical services, sports industry, or even for organizations/structures, who specialize in creating the exercise machines.

In our case, we are going to provide the economics in the logistics sector, by comparing the current solution with its alternatives.

First of all, the problem of losing control on the road is known for a quite long time. As a result, some companies have already built their business on this problem. However, most of such solutions don't iterate with the user but instead, rely on indirect data. For instance, the speed change over time, or the face mimics. Such approaches have the significant meaning, but they can't detect the exact reason, why the driver drive in such manner. This makes such procedures useless due to lousy detection of driver's cognitive state. That is why even the driver got into an accident, it's quite difficult for insurance companies to proof, that the accident happened by the driver's fault.

Another generation of such sleep detectors is based on interaction with the driver. These devices measure specific driver's parameter, like blood pressure or skin conductivity. However, these approaches, due to a large number of indirect factors like weather, or taken certain medicaments - may show false value, and skip the critical situation on the road.

Currently, there are two leading companies, who offer such devices in Russia - StopSleep, and SleepAlert. Their stock price varies from 120$ up 200$ per item. No additional spends require.

Delivery flow

Before moving to financial side (i.e., integration costs), we should define the business processes of the particular company, who is going control their drivers.

From this point, we should better understand the clients' pain, by reproducing the issues, which they feel during the work process. The standard delivery process scheme looks like so:

Pic. 15. Standard delivery process

The following diagram has been taken from the large delivery company “Rider”. The delivery process, according to this diagram, starts from shipment tracking (where we watch for the delivery state). Following up this diagram, we see the next essential element of passing the order to the customer - the delivery process itself. In two mentioned points - the driver is served as an executor of this process. And in case, something will go wrong; we are at risk:

1) Lose the truck, driver (as we won't know, what had happened and where he is now)

2) Lose the order (as the delivery won't happen in time)

In both cases, just tracking the driver's attention level - is not enough. It's more complicated task, which also involves additional solutions to be integrated, to be aware, where the driver is and what's going on.

As a result, to automate these processes, we have to think not only about real-time tracking but also provide an ability to watch for driver remotely and give the full picture, what is going on with him during the execution. Furthermore, in case the driver will get into an accident, we also will need the history of his activity before a crash, to find out, by whose fault that happens.

The economics of altering solutions

Comparing to NeuroIO, the solutions presented above much chipper, as they don't require any third-party cloud service to serve users. As a result, integrating this approach to a company with about 1000 employees may take at least 120.000$. Under the hood, this price does not include additional spends on technology, to maintain it: these technologies are useless without an addition software/hardware equipment. These parts include at least the tracking panel, where an operator can watch for driver's movement, some service, which will keep all story of driver's actions, and of course, some alert system - when the driver will get into an accident.

This software/hardware complex may add additional spends to the company. With the example presented above, the stock price of integrating such complex to your business process may take extra 25000$ dollars + regular payment to third-party services for the license.

The current solution economics

The following technology offers a full circle of controlling the delivery process thanks to its components:

Mobile app

The mobile app helps the driver to keep the road by analyzing his mental activity on the go and notifying him, in case he becomes tired.

Admin panel

The panel, used by an operator, will help him to understand where the driver is, and how did he feel himself. Furthermore, in case the driver loses the connection (enter the blind zone), the operator will know about it, as all data is passed to admin panel including current session state.

This complex, besides software, also include EEG device as the primary device for obtaining user state. The provided solution can work with Macrotellect and Neurosky devices. Their stock price varies from 100$ up to 140$ per item.

The costs of running the cloud-based infrastructure (based on Google cloud or Amazon) varies from company size. For instance, the company with about 10000+ drivers may cost about 1000$ per month. This cost includes two separate servers for holding database cluster and two microservers, where the backend will be rolled out.

Final compare:

Table 4

9. Conclusions

In this paper, we outlined a way, in which we can track user's specific mental state, like attention level, in a short/long run. We have discussed the problem and proposed a solution by applying the machine learning over previous grabbed points from EEG device. Also, we have defined a set of models, by which we can increase the accuracy of calculated average certain artifact level and predict it in next 5-10 minutes.

Finally, I would like to admit, that subject, which we have covered in this article, is a simple use case. But it still not applicable to challenges like measuring specific mind state and predict it for next weeks or even months, as it requires more complex expert system and more parameters, to be included in it.

10. References

1. Erdi Tosuna, Kadir Aydinb, Mehmet Bilgilib, Alexandria Engineering Journal (December 2016), 3081-3089 (accessed January 4, 2018)

2. Abu-Mostafa Y. S. (1990). Learning from hints in neural networks, J. Complexity 6, 192-198. (accessed January 4, 2017)

3. Baum E. B. (1991). Neural net algorithms that learn in polynomial time from examples and queries, IEEE Trans. on neural networks 2 1, 5-19. (accessed January 4, 2017)

4. Westreich, D., Lessler, J. & Jonsson, M. (2010), `Propensity score estimation: Neural networks, support vector machines, decision trees (CART), and metaclassifiers as alternatives to logistic regression', Journal of Clinical Epidemiology 63, 826-833. (accessed January 4, 2017)

5. Ikbal M. S., Misra H. and Yegnanarayana B. Analysis of autoassociative mapping neural networks. In IEEE Proceedings of the International Joint Conference on Neural Networks, Washington, USA, 1999. 70 (accessed January 4, 2017)

6. Schimke S., Vielhauer C., Dittmann J. Using Adapted Levenshtein Distance for On-Line Signature Authentication. Proceedings of the Pattern Recognition, 17th International Conference on (ICPR'04), 2004. (accessed January 4, 2017)

7. Davidson, RJ, and A Lutz. “Buddha's Brain: Neuroplasticity and Meditation.” IEEE Signal Processing 2007 : 171 - 174. Print. (accessed January 4, 2017)

8. Valle, Ronald S., and John M. Levine. “Expectation Effects in Alpha Wave Control.” Psychophysiology 12.3 (1975) : 306-309. Print. (accessed January 4, 2017)

9. Deloitte research. https://www2.deloitte.com/content/dam/Deloitte/us/Documents/public-sector/us-blockchain-opportunities-for-health-care.pdf (accessed January, 2016)

10. Lotte, F., Congedo, M., Lecuyer, a., Lamarche, F., & Arnaldi, B. (2007, June). ґ A review of classification algorithms for EEG-based brain-computer interfaces. Journal of neural engineering, 4(2), R1-R13. (accessed January 4, 2017)

11. Abdul Latef Haroon P.S, U.Eranna, Ulaganathan J., Raymond Irudayaraj I. eye blink controlled robot using EEG technology - http://www.digitalxplore.org/up_proc/pdf/278-14894930999-13.pdf (accessed January 4, 2017)

12. Coral health. Learn to securely share files on the blockchain with IPFS!

https://medium.com/@mycoralhealth/learn-to-securely-share-files-on-the- blockchain-with-ipfs-219ee47df54c (accessed January 4, 2017)

13. Allen Laura M [et al.] Sequence-specific MR Imaging Findings That Are Useful in Dating Ischemic Stroke [Article]. - 2012. - 5 : Vol. 32. - pp. 1285-1297. (accessed January 4, 2017)

14. Blockchain healthcare strategy.

https://gem.co/wp-content/uploads/2017/09/A.-Schumacher-2017-Blockchain-Healthcare-Strategy-Guide.pdf (accessed January 4, 2017)

15. Cuadra M.B. [et al.] Comparison and validation of tissue modelization and statistical classification methods in T1-weighted MR brain images [Journal] // Medical Imaging, IEEE Transactions on. - 2005. - 24. - pp. 1548 - 1565. (accessed January 4, 2017)

16. Josh Stark. Making Sense of Ethereum's Layer 2 Scaling Solutions: State Channels, Plasma, and Truebit - https://medium.com/l4-media/making-sense-of-ethereums-layer-2-scaling-solutions-state-channels-plasma-and-truebit-22cb40dcc2f4 (accessed January 4, 2017)

17. “Ethereum: Signing and Validating” (2017)

https://medium.com/@angellopozo/ethereum-signing-and-validating-13a2d7cb0ee3 (accessed January 4, 2017)

18. Weishaupt Dominik, Kцchli Victor D. and Marincek Borut How Does MRI Work? [Book]. - Berlin Heidelberg : Springer-Verlag, 2006. (accessed January 4, 2017)

19. M. Jones, D. Hardt. The OAuth 2.0 Authorization Framework: Bearer Token Usage - https://tools.ietf.org/html/rfc6750 (accessed January 4, 2017)

20. Telemedicine today. Available at: http://www.techracers.com/blockchain-technology-telemedicine (accessed January 4, 2017)

21. Time Series Prediction with LSTM Recurrent Neural Networks in Python with Keras. Available at: https://machinelearningmastery.com/time-series-prediction-lstm-recurrent-neural-networks-python-keras/ (accessed January 4, 2017)

22. Using Machine Learning to Predict the Weather: Part 3

http://stackabuse.com/using-machine-learning-to-predict-the-weather-part-3/ (accessed January 4, 2017)

23. A simple deep learning model for stock price prediction using TensorFlow

https://medium.com/mlreview/a-simple-deep-learning-model-for-stock-price-prediction-using-tensorflow-30505541d877 (accessed January 4, 2017)

24. How to convert raw values to voltage

http://support.neurosky.com/kb/science/how-to-convert-raw-values-to-voltage (accessed January 4, 2017)

25. Inserting metadata into the blockchain.

https://medium.com/@bkawk/inserting-metadata-into-the-blockchain-40c0734b203e (accessed January 4, 2017)

26. Coral Health vs. Other Healthcare Blockchain Companies, Part 2

https://medium.com/@mycoralhealth/coral-health-vs-other-healthcare-blockchain-companies-part-2-f9d49b03afdc (accessed January 4, 2017)

11. Application

Application 1. Mobile app screenshots

Pic 16-22

Application 2. Admin panel screenshot

Pic 23

Pic. 16. Login screen

Pic. 17. Main interface screen (Android)

Pic. 18. History item screen (IOS iPad)

Pic. 19. History item screen (IOS - iPhone)

Pic. 20. Main screen (IOS - iPhone)

Pic. 21. Welcome screen (IOS - iPhone)

Pic. 22. History item screen (Android)

Pic.23. Admin panel screen

Размещено на Allbest.ru