Oracle announced that it has open-source GraphPipe to improve machine learning applications. The goal of the project is to improve the implementation results for machine learning models, said Project Leader Vish Abrams.

GraphPipe consists of a set of libraries and tools to follow an implementation standard. It is an attempt to standardize at the client end and transport layers of machine learning.

Other vendors, such as Google and Microsoft, have tried to develop similar capabilities, but not through open source. Developers have progressed in the construction of machine learning in applications in recent years, but successfully implementing a model requires overcoming several problems.
Oracle’s decision to open this project could benefit development communities.

That it makes?

GraphPipe is a network protocol that simplifies and standardizes the transmission of automatic learning data between remote processes. There is no dominant standard on how stress data should be transmitted between components in a deep learning architecture.

Developers commonly use protocols such as JSON. But that solution is inefficient. TensorFlow uses multiple protocol buffers, which makes it a large and complex software.

The GraphPipe design solves both limitations by providing efficiency through a binary format, mapped in memory, while being simple and light in the dependencies. It includes simple implementations of clients and servers that make the models of implementation and consultation of machine learning from any framework less complicated.

Oracle has been on the path to becoming a strong advocate for collaboration and transparency, he suggested. The company has focused on finding the best comprehensive solutions for customers, and that is an indication that its adoption of open source is authentic.