FELToken with Ocean Protocol, Future and Integrations

Ocean Protocol is offering monthly grants. We proposed our solution for federated learning on blockchain called FELToken. As Round 13 is quickly approaching, it’s time to evaluate our work and think about future steps.

First and foremost what is FELToken? Our goal is to create a platform that will allow running federated learning — decentralized privacy-preserving AI — projects on a blockchain using smart contracts. More simply our tool connects data scientists with data owners so that data privacy is preserved. Data scientists can then train machine learning models without seeing actual data. This, for example, allows companies to collaborate on projects without the need to share their data.

Web Application

Our vision is to create a tool allowing everybody to start a federated learning project within a minutes without any extra knowledge about smart contracts. With that being said, this month we focused on creating a web application allowing all necessary interactions:

☑ Creating project
☑ Listing projects
☑ Showing project dashboard
☑ Starting a ML model training

These are the core functionalities required for starting a new project. All these actions can be performed through the web interface. The only requirement is that a user connects his or her crypto wallet. We hope that this simplifies the process of starting a new FL project compared to other solutions out there.

Screenshot from project deployment.

Although the main functions are finished, we are constantly working on making the app even better. Many more features and improvements still need to be done. We are also planning to work with a UI/UX designer who would help users make the interaction even smoother.

Ocean Protocol Integration

With core functions being finished, we have a solid base to work on advanced features and integrations with other tools. Ocean Protocol is exciting technology providing a marketplace for data. Rather than explaining all the details about Ocean itself, we will present 3 ways how we are planning to integrate it with FELToken. The integrations are presented in the order in which they will be later implemented.

Selling Machine Learning Models

You probably heard of GPT-3, BERT, EfficientNet or any other model out there. These pre-trained models play a crucial role in many applications. Naturally comes the idea of selling your ML model which might be useful to others. Therefore, we would like to allow our users to easily publish their models to Ocean and make a profit.

Compared to public models, models from FELToken could be quite valuable as they will be trained on private data. Profit would be then divided between parties that participated in model creation. This could further drive participation from data providers and data scientists to build new models.

Using Datasets from Ocean Protocol

On Ocean, you can publish data in different formats. Using static files is relatively straightforward. You can just download the files and then use them for training as any other data. However, Ocean provides also so-called compute-to-data. This allows running compute jobs on data without actually seeing them. We would like to integrate this with our tool so that users can easily integrate such data with datasets of their own. Moreover, our tool could allow simple interaction with such datasets.

Providing Models as API

With the compute-to-data comes one more possible use case. Rather than selling whole model projects could provide model only as API. So potential users could just send data for computation and receive model output. Though this might sound intriguing for model creators, in machine learning people often need to apply different optimization techniques to make the model suitable for their task (e.g. fine-tuning). Therefore models provided through API would make sense only for very general tasks, for example, machine translation or object detection. We will probably postpone the development of this feature until other integrations are finished

Conclusion

We completed our grant deliverables for this month. With that finished, we are looking forward to the next round. Next month, we would like to start experimenting with integrating Ocean Protocol. However, the proposed integrations might still change as we dig deeper into Ocean technology. For the latest updates please follow us on Twitter: @FELToken

In case you can think of any other possible integration please leave a comment below.

Resources

Web: https://feltoken.ai/
Ocean Protocol Grants: https://oceanprotocol.com/dao
FELToken Proposal (Round 13):
https://port.oceanprotocol.com/t/feltoken-federated-learning-token/1292

--

--

--

Programming | Machine Learning | Blockchain | PhD student | Building start-up: feltoken.ai | Follow for weekly stories/tutorials from the start-up journey.

Love podcasts or audiobooks? Learn on the go with our new app.

Recommended from Medium

NLP News Cypher | 01.19.20

Google’s Machine Learning with TensorFlow on Google Cloud Coursera Specialization Review

What is F1 Score :“The Complete Understanding ”

The Meaning Behind L1 L2 Regularization

BatteryBERT for battery materials

Depth-wise [Separable] Convolution Explained in TensorFlow

MLops Automation task ,the integration of Machine learning with devops

Unlocking the secrets of great earthquakes with deep learning

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Břetislav Hájek

Břetislav Hájek

Programming | Machine Learning | Blockchain | PhD student | Building start-up: feltoken.ai | Follow for weekly stories/tutorials from the start-up journey.

More from Medium

Ethereum’s Future: What does it mean for other blockchains?

zKasino — Decentralized Betting Platform on StarkNet

Filecoin News 18: Web3 Weekend, Launchpad Accelerator, Frontier Accelerator Demo Day, and More

Iris: A Next-Gen Decentralized Storage Layer (Part 1)