CoolGraph

2023: Biliotheca Issue

In early December 2023, the first open library in Russia was released to create graph neural networks. We are talking about the CoolGraph project, which is developing MTS within its Big Data center.

The CoolGraph source code is published in the Github repository. The new solution, intended for specialists in deep training of neural networks, allows you to build graph neural networks that differ from standard ones using several lines of code.

The first open library in Russia for creating graph neural networks has been released

As MTS explained, graph neural networks differ from conventional machine learning models in that they take into account any relationships in the data and summarize all previous approaches. Therefore, the neural network is "more flexible and variable." For example, a chatbot will be able to analyze each word in the context of the previous ones, giving the most accurate answer, the company noted.

They also clarified that the CoolGraph library is intended for both experts and novice specialists in the field of deep learning of neural networks that have access to graph data with which they can build various models. Big Data MTS specialists collected in the library developments in the field of building custom models.

To achieve breakthroughs in machine learning, a large number of talented people experimenting with technology are needed. Our library solves this problem by lowering the entrance threshold to graph neural networks. This technology, on the one hand, combines well-known models of deep learning as special cases, and on the other hand, provides a tool for new research, "said Victor Kantor, director of the MTS Big Data Center.

You can use CoolGraph using two lines of code, and the user can track all the results of experiments in MLFlow. In the future, the developers plan to replenish the CoolGraph library and open access to their other libraries.^[1]

Notes