What is Thuja?

Thuja is a declarative deep learning framework designed to be embeddable into other software.

Thuja is intended for the tasks when deep learning is a tool to solve a task, not a task itself. This means that Thuja is not intended for designing state-of-the-art deep learning architectures or researching the properties of new layers or model families. However, Thuja is good for applied deep learning research in any field of science, when a deep learning model is designed for a specific task and solving this task is the purpose of the research.

Thuja is good for the cases when you have a problem you want to solve using deep learning, but you do not want to dive deep into the theory or mathematics behind deep learning. You do not need a PhD in AI to use Thuja.

Thuja is also designed to be embeddable into other software. Once you have finished training your model with Thuja, you can use Thuja as a runtime to compute your model's predictions, even from your application. Thuja is distributed as a single shared library with C interface and can be embedded into any application that supports calling C functions from shared libraries. The same library is used both for training an deploying the models, so the behaviour of your model stays the same.

What is its current state?

Thuja is under active development. It is not publicly available yet, and we do not accept applications from potential testers and early adopters.

Thuja will become publicly available once it reaches the MVP state.