New release is out! Let’s look at FELT Labs 1.0
Making federated learning even easier than before
We are thrilled to announce the latest release of our application, which includes several exciting features and enhancements. Over the past few months, we have been refining the app and adding new functionality to make it more powerful and user-friendly. This includes a Python interface, custom algorithms, user accounts, the ability to use FELT across devices, team collaboration, and more. Let’s take a closer look.
If you prefer more hands on exploration, go to app.feltlabs.ai and try it for yourself.
We added backend code and database to our application. While this is not a pure web3 solution, it simplifies many user interactions and allows many future extensions. We considered multiple architectures before we picked the one which provides a smooth user experience while leaving control to users.
User Accounts
One of the most significant changes you will notice is the addition of user accounts, right now allowing you to log in using your Ethereum wallet or Google account. This new feature enables you to monitor your compute jobs across devices and store your automation account securely. User accounts unlock many exciting features, such as sharing compute jobs with other users or collaborating in teams on machine learning projects.
Often times multiple people are working on a single machine learning project. This way, they can monitor what trainings were started by others and share the computation results right inside our application!
Ocean Protocol Marketplace Integration
Next, we are releasing an npm package that will provide some of the FELT Labs functions, such as:
- Starting federated learning and other algorithms
- Listing and monitoring started jobs
That way, market owners can install one package into their Marketplaces to support federated learning. The package provides components with the same functionalities as in our application. This will bring federated learning directly to marketplaces and closer to the end users.
We will talk more about the exact details in a future blog post. If you are running your own Ocean marketplace, feel free to contact us, and we can help you integrate FELT Labs!
Python Interface
Another important feature is a Python interface for the FELT Labs. Starting computation through a browser works well for a small number of datasets but can get tedious as the number of datasets grows. For that, we provided a command line tool written in Python, which allows users to start the computation using one command. Furthermore, the script is connected to our web application. That way, you can still monitor the computation through our app, even if it’s being started elsewhere.
More details about the interface can be found here:
Custom Algorithms
Previously we supported only our algorithms. From now on, you can run your custom algorithms by specifying IDs of your own training and aggregation algorithm. Custom algorithms provide greater flexibility and allow users to use FELT for different settings. This is especially useful if you need to use custom data loading methods and different machine learning models. You can contact us if you develop your own algorithm and you want to share it with other users. We can then include the custom algorithm directly in our application.
What’s Next?
There are many more minor changes that didn’t make it into the post. There are even more changes that still need to be implemented. The user accounts are definitely the most important addition, as it has great potential for extending the application and allowing us to add many new features for user collaboration.
Sounds interesting? Go to app.feltlabs.ai and test it yourself. We appreciate any feedback or bug reports.
