Skychain Beta first details!
Hello, Skychain community! It is time to share with you the first details of our long-waited Beta release!
First of all, Skychain Beta is a platform, which will feature the full functionality and capabilities for data providers, developers and consumers to host, train and use neural networks.
The whole Beta experience starts with the registration process. Right after it, the user will be offered an opportunity to create Skychain profile, which provides the access to Developer and Data provider sections additionally to Consumer section with an opportunity to use neural networks. Registration of this profile requires some additional information to be added to your account info.
Medical data provider
In medical data provider section, a user sees the list of his uploaded datasets and is able to upload a new one.
To upload the dataset, the provider should give the name and a broad description of the images types and diseases, which are marked out on the images. As the image types might significantly differ (from .img to .csv), we are going to implement the features, which will more or less standardize the uploaded data.
Dataset, which is already uploaded into the marketplace, will also provide the link on provider’s profile for potential developers.
In terms of data security, we decided to implement the dataset subdivision as one of the measures to prevent data leaks and to allow the developer check the data for relevance.
The issue is that any dataset is a private property of its owner, so we shouldn't give the developer a full access to it. But the developer needs to have at least partial access to the data to make sure it is relevant.
A dataset will be divided into three parts:
- public verification and EDA part;
- private training part;
- private validation part.
The public part is required to make sure that the dataset is properly labeled and does not content any irrelevant information. Its other purpose is to be used during exploratory data analysis, the stage when the developer tests the data. The private part for training is not used for purposes other than the training itself. The private part for validation is needed to evaluate the results of training using different metrics.
Moreover, Skychain is going to implement its own tool for automatic verification of datasets within the marketplace. It will automatically rate a dataset in accordance with some internal parameters, so any developer could see how good the images are.
Update/removal of a dataset
A dataset can be updated (by adding more files) or completely removed from Skychain. This can be done by its publisher in his profile.
There are two main blocks in Developer section: the list of developed neural networks and the list of ongoing projects.
Beginning a new project
AI Developer, interested in starting a new project, must find a potential dataset (or datasets) that suit him best for the research. Then, he starts a new project, and he sees the working area with several notebooks, which will be used by the developer through the EDA (exploratory data analysis) on a public part. After the EDA, when the developer finished building the main algorithms and layers of a neural network, he creates a checkpoint and starts training on a private training part. In case of a failure, the developer is able to come back to a checkpoint and rerun the process. When the training done right, the developer verifies it on the private validation part and evaluates the research.
In case of a successful training, the developer publishes the neural network, gives it a name and a description. The profile of the neural network will contain the links to the developer’s profile and to the datasets it was trained on.
The update/removal of the neural network
The neural network can be updated (by using additional datasets for training) or completely removed from Skychain. This can be done by its publisher in its profile.
The consumer’s interface will not be significantly changed since the Alpha’s one. Users will be able to use neural networks, which are created within Skychain.
This is pretty much all the information about the Beta release. You also might have noticed that Beta will not include blockchain, as it will be implemented in the final version of the project. And of course, some parts of the Beta can be changed as it is still in development.
We are doing our best to finish Beta in December. However, assessing the current situation, we realize that we might not make it in time, so the Beta can be released in January.
Today, project’s team has held a meeting, where the final date of the Beta release was finally settled. Skychain Beta will be released on January 31, 2019.
According to project’s roadmap, our Beta was scheduled to be released in December 2018. We apologize for not following the roadmap, but the project’s team is willing to deliver a high-quality product even at the Beta stage of development. We hope for your understanding.
Alexander Oksanenko, Marketing Director
If you have any questions about Skychain, don’t hesitate to write me on Telegram and on email: email@example.com.